This PDF file contains the front matter associated with SPIE Proceedings Volume 11059 including the Title Page, Copyright information, Table of Contents, Introduction, and Conference Committee listing.

Through wall indoor surveillance and monitoring is an emerging area of research and developments in many real-life applications, including security and search and rescue. In this context, it is important to deploy remote sensors able to glean information about the presence of people in indoor areas. Radar devices based on the exploitation of electromagnetic signals represent an attractive sensing modality. When operating at frequencies from hundreds of MHz to few GHz, they are capable to penetrate common construction materials enabling the detection of people in inaccessible environments. Two different types of surveillance radars are considered in this paper. The first one exploits the synthetic aperture concept with the aim to detect, locate, and track multiple subjects in the scene. The second one illuminates the scene from a fixed position and exploits the Doppler effect. Despite the limited information available with respect to aperture radars, Doppler radars allow the detection of moving targets behind walls and inside enclosed structures. This work presents a brief overview of recent developments in the field of radar data processing for indoor monitoring together with their assessment by means of numerical and experimental tests.

As is well known, in through-the wall imaging one needs to estimate the wall electromagnetic parameters in order to get properly focused images. Even under the simplest case in which the wall can be assimilated as a single homogeneous slab, this problem puts some difficulties since the wall dielectric permittivity and thickness are nonlinearly linked to the reflected field data. The usual way to go on that problem is through some optimization iterative procedure which can be time consuming and can suffer from false solutions. In this contribution we propose a method that avoids the previously mentioned drawbacks by leveraging on a MIMO configuration. The main idea is to estimate the wall transmission coefficient rather than its electromagnetic properties. This way, one estimates the kernel of the relevant (for imaging) scattering operator instead of constructing it after wall parameters have been estimated. More in detail, it is shown that the characterization stage is cast as a linear inverse problem which is solved by a Truncated-Singular Value Decomposition method. The proposed method avoids optimization but in principle can be applied only for lossless walls. However, multi layered walls can be dealt with as well. In this contribution we focus only on the wall transmission coefficient estimation; once it has been obtained imaging can be achieved by standard back-propagation algorithms. In particular, the study is developed for a single wall and 3D vector case and some numerical examples are reported to check the theory.

Modern radars systems for anti-air and anti-ship application naturally had born interest how to do the antidote against of it. Stealth or how to do the reduction of the radar cross section concerns mainly two decisions: a) The creation specially oriented surfaces which reflect radar radiation to the space or down – some kind as “in milk” (on technical language). b) Using special absorption coating on the surfaces and places which is impossible or very difficult to exclude from the reflection of the radar radiation. Absorption in this case means – sounding radar power will be absorbed by the material/s of such coating. Such coating works as black body or matched loading if to talk on microwave language. For the microwave radiometer such surfaces and coating are not antidote for the disclosing according to the physics of radiometer’s job. Nature produces radiation, which re-reflect from the various surfaces to an antenna of the radiometer from the appropriated directions from the sphere of sky or the earth with power depending on, so named, radio-brightness temperatures. And the coating if it the matched loading produces microwave radiation corresponding to her physical temperature. For the both cases a) and b), Stealth object can be disclosed by radiometer in some real cases. Keywords: stealth coating, microwave radiometer,

Passive wireless sensors provide an attractive technology in niche applications where battery powered sensors are not applicable. Surface acoustic wave technology provides an optimum implementation of passive wireless transducers acting as cooperative targets to short range radar systems: the signal carrying the information is delayed beyond clutter, with a piezoelectric substrate converting the electromagnetic wave to an acoustic wave whose properties are dependent on the environment. We have tackled the issue of short range radar certification by considering a passive radar approach in which existing non-cooperative radiofrequency sources are used to power the sensor, and the reader is made solely of passive receivers aimed at recording the reference signal generated by the non-cooperative source and the backscattered signal returned by the sensor. A passive radar approach only requires a cross-correlation between the reference and surveillance signals to identify the time delay between the incoming and backscattered signals and hence the recovery of the physical quantity sensed by the acoustic transducer through acoustic velocity modulation. Practical sources do exhibit some short term correlation. Hence, strong copies of the reference signal with delays shorter than the one introduced by the sensor must be canceled by the receiver to allow for the weak sensor response to be extracted. This classical problem is called Direct Signal Interference (DSI) suppression. In a post-processing approach with little computation time limitation, this problem is solved using a least square error optimization approach. In the context of real time sensor measurement using a Field Programmable Gate Array (FPGA) implementation, data recording, frequency transposition and decimation are readily implemented in the gate array matrix. DSI removal appears as the limiting factor for a full FPGA implementation of the short range passive radar reader. We address the challenge by the iterative process of DSI suppression using Orthogonal Matching Pursuit (OMP) algorithm, and additionnally consider the FPGAfriendly Stochastic Gradient Descent (SGD) approach to try to recover DSI coefficients from a pipelined algorithm using streams of measurement data.

Droplets atomization by shockwave can occur in different issues commonly encountered in the industry such as leak, tank leakage or triple aggression (high speed impact, rupture and surrounded secondary explosion) of tanks. For the last case, shockwave can interact with liquid jets of drops and propagates the liquid far away from the container zone. Very fine secondary droplets can be produced in the worst case of atomization. These small particles can generate secondary effects like explosion in case of petrol derivatives in fire or toxic effects in case of direct breathing. High speed imaging is well suited to study transient phenomenon like explosions and shockwave. A dedicated shockwave generator has been designed to cope with interferometric measurement on holographic bench. This demonstrator is made of thick plastic tubes. The high pressure chamber is isolated from the guiding tube by domestic aluminum foils, the thickness and number of which drive the pressure rupture. Previous works have been carried on by time resolved shadowgraphy to characterize generated shockwave at the guiding tube outlet. This paper deals with time resolved digital holography to perform higher accuracy measurements and of course to reach 3D reconstruction of the whole phenomenon. Lensless in-line digital holography is carried on to improve the stability of the holographic set-up. Different Phantom high speed cameras have been tested as recording sensors, following pixel pitch, pixel size and of course the maximal throughput, from 7kfps (frame per second) up to 26kfps at full 1Mpixel resolution. Different regimes of droplet trains and droplet sizes have been tested. This has also been carried on for different liquids to show the effect of the physico-chemical properties of the liquid subjected to shockwave.

Nowadays, digital holography can be considered as one of the most powerful imaging modality in several research fields, from the 3D imaging for display purposes to quantitative phase image in microscopy and microfluidics. At the same time, machine learning in imaging applications has been literally reborn to the point of being considered the most exploited field by optical imaging researchers. In fact, the use of deep convolutional neural networks has permitted to achieve impressive results in the classification of biological samples obtained by holographic imaging, as well as for solving inverse problems in holographic microscopy. Definitely, machine learning approaches in digital holography has been used mainly to improve the performance of the imaging tool. Here we show a reverse modality in which holographic imaging boosts the performance of machine leaning algorithms. In particular, we identify several descriptors solely related to the type of data to be classified, i.e. the holographic image. We provide some case studies which demonstrate how the holographic imaging can improve the performance of a plain classifier.

Friction stir welding is a solid-state welding process. The technology is used in high precision applications such as aerospace. Thus, monitoring the weld quality is highly relevant for detecting inaccurate welds. Various studies have shown a significant dependence of the weld quality on the welding speed and the rotational speed of the tool. Frequently, an unsuitable setting of these parameters can be detected by visual examination of the resulting surface defects, such as increased flash formation or surface galling. The visual inspection for these defects is often conducted manually and is therefore associated with increased costs and personnel effort. In this work, a deep learning approach to automatically detect irregularities on the weld surface is introduced. For training and testing of the artificial neural networks, 112 welds with a total length of 18.4 m were produced. Color images of the welds were taken using a digital camera and images of the weld surface topography were made with a three-dimensional profilometer. The approach consisted of a two-step procedure. First, an object detector using a neural network localized the friction stir weld on the image. Second, a neural network classified the surface properties of the weld seam. The object detector localized the friction stir welds with an intersection over union up to 89.5%. The best result in classifying the surface properties was achieved by using the topography images. Thereby, a classification accuracy of 92. % was reached by the convolutional neural network DenseNet-121. The results are the basis for the future development of an inline quality monitoring and parameter control method for friction stir welding.

Meeting the demands of Industry 4.0 and Digital Manufacturing requires a transformative framework for achieving crucial manufacturing goals such as zero-defect production or right-first-time development. In essence, this necessitates the development of self-sustainable manufacturing systems which can simultaneously adapt to high product variety and system responsiveness; and remain resilient by rapidly recovering from faulty stages at the minimum cost. A Closed-Loop In-Process (CLIP) quality control framework is envisaged with the aim to correct and prevent the occurrence of quality defects, by fusing sensing techniques, data analytics and predictive engineering simulations. Although the development and integration of distributed sensors and big data management solutions, the flawless introduction of CLIP solutions is hindered specifically with respect to acquiring and providing in-process data streams at the required level of: (1) veracity (trustworthiness of the data); (2) variety (types of data generated in-process); (3) volume (amount of data generated in-process); and, (4) velocity (speed at which new data is generated in-process) as dictated by rapid introduction and evolution of coupled system requirements. This paper illustrates the concept of the CLIP methodology in the context of assembly systems and highlights the need for a holistic approach for data gathering, monitoring and in-process control. The methodology hinges on the concept of “Multi-Wave Light Technology” and envisages the potential use of light-based technology, thereby providing an unprecedented opportunity to enable in-process control with multiple and competing requirements. The proposed research methodology is presented and validated using the development of new joining process for battery busbar assembly for electric vehicles with remote laser welding.

Closed loop dimensional quality control for an assembly system entails controlling process parameters based on dimensional quality measurement data to ensure that products conform to quality requirements. Effective closed-loop quality control reduces machine downtime and increases productivity, as well as enables efficient predictive maintenance and continuous improvement of product quality. Accurate estimation of dimensional variations on the final part is a key requirement, in order to detect and correct process faults, for effective closed-loop quality control. Nowadays, this is often done by experienced process engineers, using a trial-and-error approach, which is time-consuming and can be unreliable. In this paper, a novel model to estimate process parameters error variations using high-density cloud-of-point measurement data captured by 3D optical scanners is proposed. The proposed model termed as PointDevNet uses 3D convolutional neural networks (CNN) that leverage the deviations of key nodes and their local neighbourhood to estimate the process parameter variations. These process parameters variation estimates are leveraged for root cause isolation as a necessary but currently missing step needed for the development of closed-loop quality control framework. The proposed model is compared with an existing state-of-the-art linear model under different scenarios such as a single and multiple root causes, and the presence of measurement noise. The state-of-the-art model is evaluated under different point selections and results are compared to the proposed model with consideration to an industrial case study involving a sheet metal part, i.e. window reinforcement panel.

Metrology assisted assembly systems constitute cyber physical production systems relying on in-process sensor data as input to model-based control loops. These range from local, physical control loops, e.g. for robots to closed-loop product lifecycles including quality management. The variety and amount of involved sensors, actors and data sources require a distinct infrastructure to ensure efficient, reliable and secure implementation. Within the paradigm of Internet of Production a reference architecture for such an infrastructure is established by four layers: Raw data (1), provisioning of proprietary systems (2), data aggregation and brokering (3) and decision support (4). In modern metrology assisted assembly systems, a virtual reference frame is constituted by one or multiple predominantly optical Large-Scale Metrology instruments, e.g. laser trackers, indoor GPS or multilateration based on ultra wideband communication. An economically efficient implementation of the reference frame can be achieved using cooperative data fusion, both by increasing the operative volume with existing systems and by optimizing the utilization of highly precise and therefor typically cost-intensive instruments. Herewith a harmonization is required as well from a physical perspective as in terms of communication interfaces to the raw data provided by the individual instruments. The authors propose a model-based approach to obtain a protocol-agnostic interface description, viewing a Large-Scale Metrology instrument as an abstract object oriented system consisting of one or multiple base units and mobile entities. Its object-oriented structure allows a realization of the interface in arbitrary structured communication protocols by adhering to fixed data transformation schemes. This approach is evaluated for MQTT (structured by topics), OPC UA (structured by data model) and HTTP/REST (structured by URLs) as key protocols within the internet of things. Moreover, the transformation between different protocols decouples software requirements of measurement instruments and actors, generally allowing a more efficient integration into cyberphysical production systems.

A single weld defect in a safety-critical engineering structure has the potential to incur high monetary costs, damage to the environment or loss of human life. This makes comprehensive non-destructive internal and external inspection of these welds essential. For non-destructive internal inspection the ultrasonic phased array supports a number of methods for producing a cross sectional image at a fixed location. Full coverage of the weld requires a sequence of images to be taken along the full length, each image at a unique incremental step. If the weld has a geometrically regular structure, such as that corresponding to a long linear section or the circumference of a pipe, automation becomes possible and data is now provided for post processing and auditing. Particularly in a production process this may provide many thousands of images a day, all of which must be manually examined by a qualified inspector. Presented in this paper is an approach for rapid identification of anomalies in sequences of ultrasonic sector images taken at equally spaced index points. The proposed method is based on robust principal component analysis (PCA). An assumption is that most sectors are anomaly free and have a statistically similar geometrical structure. Unsupervised multivariate statistical analysis is now performed to yield an initial low dimensional principal subspace representing the variation of the common weld background. Using the Mahalanobis distance outliers, observations with extreme variations and likely to correspond to sector scans containing anomalies, are removed from the reference set. This ensures a robust PCA-based reference model for weld background, against which a sectorial scan is identified as defect free or not. Using a comprehensive set of sector scan data acquired from test blocks, containing different types and sizes of weld defects at different locations and orientations, the paper concludes that PCA has potential for anomaly detection in this context. Although trimming improves the accuracy of the system eigenvectors, it is shown that greater accuracy of the low rank subspace is possible through principal component pursuit (PCP). This is evident by an almost 100% anomaly detection rate with a false alarm rate of well below 10%.

The paper deals with subsurface imaging via radar systems mounted onboard aerial platforms. Specifically, the attention is focused on a radar prototype installed on a small unmanned aerial vehicle (S-UAV), previously proposed by few of the authors. In particular, the challenges in terms of electromagnetic modeling and flight dynamics knowledge and control are here tackled. In this frame, an ad-hoc designed data processing strategy is presented; this strategy involves a preprocessing step and a reconstruction step. The pre-processing is performed in time domain and, beyond filtering procedures commonly exploited in radar imaging, involves a procedure devoted to compensate flight altitude variations and to account for the S-UAV trajectory, which is estimated by processing measurements collected by an onboard GPS receiver. In addition, the reconstruction of the investigated scenario is performed by means of a microwave tomographic approach based on a linear model of the electromagnetic scattering and the concept of equivalent dielectric permittivity for the propagation path. This latter allows us to properly face the imaging of buried objects, while avoiding the mathematical complexity introduced by the presence of the air-medium interface. Accordingly, the imaging is faced as a linear inverse scattering problem formulated in the spatial domain similarly to the case of a homogeneous scenario and, thanks to the concept of equivalent permittivity, depth and horizontal position of buried objects are retrieved properly. This is corroborated by means of a numerical analysis accounting for synthetic data.

In last years the Frequency Modulated Continuous Wave (FMCW) technology has been playing an ever greater role in the realization of compact, lightweight and cheap Synthetic Aperture Radar (SAR) systems to be mounted onboard small, low altitude platforms such as airplanes, helicopters and drones. In this work, we present the imaging capabilities of the AXIS (Airborne X-band Interferometric SAR) system, which is an Italian airborne interferometric SAR system operating at X- Band and based on the FMCW technology. The system has been developed in the frame of a cooperation between a public research institute (IREA-CNR) and a private company (Elettra Microwave S.r.l.). The first flight tests of the system have been carried out at the end of April 2018 over the Salerno coastline, in the South of Italy, onboard a Cessna 172 aircraft. In this paper, we show a preview of the obtained preliminary results, comprising a description on the mission, an analysis of both the SAR images and the interferometric products.

Low cost solutions to infrastructure’s thermal monitoring has become possible thanks to the development of uncooled infrared cameras over the past few years. However, in order to get a temperature value, one need to convert the radiative fluxes received by the camera. Such a step depends on spatial parameters, environment and the observed target characteristics which are usually unknowns or roughly estimated. To overcome this lack of knownledge, one solution is to combine the infrared camera measurements with other sensors. The purpose of this study is to examine the influence of the different parameters on the temperature conversion process and finally see to what extent complementary data can improve the estimation of the targeted object’s temperature.

Smart monitoring of critical civil engineering infrastructures has become a priority nowadays as ageing of construction materials may have dramatic consequences on the community. The issue is exacerbated as it applies to many structures at the network level rather than to single structures or limited areas. To this effect, catalogues for assessment of decay conditions and identification of maintenance actions are crucial pieces of information for infrastructure management purposes. Within this context, innovative non-destructive methods, such as space-borne techniques, have been increasingly used for monitoring purposes in the past two decades. Among these, the Interferometric Synthetic Aperture Radar (InSAR) imagery technique is gaining momentum nowadays. This method is used to monitor ground and infrastructure displacements at the large scale with a millimeter resolution. It can compare radar satellite images over time, and it is capable to measure variations accurately using interferometry. An advantage of using InSAR techniques is that these are not affected by cloudiness as well as lighting conditions, since data can be also collected at night time. On the contrary, InSAR is computational-demanding as it requires to filter out unnecessary information from the entire captured area to obtain data on the target domain. Most common applications include landslide assessment and monitoring of surface deformations following major seismic events. This paper reports a methodology for the assessment of surface deformations of viaducts by reducing drastically computational time. To this purpose, a multi-stage automatic bridge monitoring protocol is developed at the network level by integration of information into Geographic Information System (GIS) catalogues and use of the InSAR imagery technique. The first stage locates the viaducts in the area of interest by querying open data and inputting results into a GIS catalogue. On a second stage, an InSAR analysis of the identified bridges is performed. This approach allows an estimate of surface displacements as well as an identification of bridge areas affected by millimeter-scale settlements. Fundamental theoretical and working principles of the two methodologies are first introduced in the paper. Advantages against drawbacks of each technique are then discussed. The last Section reports a case study and a discussion of the main results including conclusions and future prospects.

A novel methodology for utilizing any arbitrary antenna as a source of radiation in Ground Penetrating Radar (GPR) simulator gprMax is proposed. gprMax is a numerical tool that solves GPR forward problem. The source of electromagnetic radiation that can be combined with gprMax, even in the latest version, is limited in nature and many practical antennas used in GPR systems are not amenable to use in gprMax. On the other hand, gprMax has turned out to be an excelled tool in GPR research. The proposed methodology is capable of examining the response of any newly developed antenna in a GPR simulation environment. Validation of the developed methodology is also presented in this work.

High-frequency spectroscopic observation methods using small satellites and drones for monitoring of plankton in the ocean and vegetation activity have recently attracted considerable attention. However, in multi-directional spectroscopic imaging, the spectroscopic characteristics vary depending on the observation and illumination angles. Therefore, huge quantities of spectroscopic data were previously required for every conceivable combination of zenith and azimuth angles to identify plant species. The method proposed here can identify any plant species from near-surface and internal reflectance spectroscopic data, regardless of the zenith and azimuth angles. We assume that the observed spectral intensity can be calculated as a linear sum of the near-surface spectral reflectivity and the internal diffusion spectral reflectivity multiplied by the light-source spectral intensity and the reflection correction coefficients a and b. We acquire the near-surface and internal reflected light as basic spectroscopic data using the orthogonal polarized light illumination method. The coefficients a and b can be calculated from basic spectroscopic data. We obtain m-sets (a_i, b_i) (i =1-m) using combinations of the numbers of λ1…λn. If the reflection correction coefficient of the m-sets (a_i, b_i) is close to one, we identify the observed plant as a plant species contained in the basic data. If the two species are different, the m-sets (a_i, b_i) have uncorrelated values and the m-sets (a_i, b_i) reflection correction coefficient decreases towards zero. In this work, we performed feasibility demonstrations using two types of plant and successfully determined from the basic data that the observed plant is the correct plant species.

Accurate classification is one of the most important prerequisites for hyperspectral applications and feature extraction is the key step of classification. Recently, deep learning models have been successfully used to extract the spectral-spatial features in hyperspectral images (HSIs). However, most deep learning-based classification methods are supervised and require sufficient samples to guarantee their performance. And the labeled samples in HSI are limited. To solve this problem, unsupervised feature extraction based on improved Wasserstein generative adversarial network (WGAN) is proposed in this paper. Further, in order to fully explore the spectral-spatial features in HSIs, a three-dimensional (3D) model is designed. Considering that its difficult for GANs to generate high dimension data, the dimension of HSI is reduced firstly by principle component analysis (PCA). Then, an improved WGAN is trained unsupervised with HSIs after dimensionality reduction to achieve the stable status. Finally, the discriminator of the trained improved WGAN is used as a feature extractor and followed by a classifier, which can be used to classify the HSIs. In order to evaluate the performance of the proposed method, PCA-based methods and GAN-based methods are compared in the experiment of a real-world HSI. Experimental results have shown that our proposed method gained more promising results which has great potential prospects in HSI classification.

In the field of hyperspectral image (HSI) processing, shadow regions in HSIs are often ignored or simply treated as a category because of their low reflectivity and complex information. There have been some studies on shadow regions of HSIs but there are still few effective algorithms to detect the real substances under the shadow regions. In view of this problem and the good performance of convolution neural network (CNN) in HSI classification and target detection, this paper improved target detection method in shadow regions in HSI which combines the CNN and the adaptive coherence/cosine estimator (ACE) of the spectral derivative image. This method includes three main steps: firstly, shadow region would been determined by CNN model whose main parameters have been adjusted to optimize the network performance; secondly, the derivative data of hyperspectral image would be obtained by deriving the shadow region of hyperspectral image; finally, due to the prominent performance of ACE algorithm in target detection of HSI, this algorithm could be applied to detect the substances contained in the shadow regions. To assess the performance of the proposed method, one widely used HSI dataset is used in the experiments. The numbers of experiment results show that the proposed method can detect the substances under the shadow regions in HSIs and it also has promising prospect in the field of HSI data processing.

Multi-spectral imagery provides wide possibilities for improving quality of object detection and recognition due to better visibility of different scene features in different spectral ranges. To use the advantage of multi-spectral data the relation between different types of data is required. This relation is provided by capturing data using calibrated, aligned and synchronized sensors. Also geo-spatial data in form of geo-referenced digital terrain models can be used for establishing geometric and semantic relations between different types of data. The presented study considers the problem of object recognition based on two data sources: visible and thermal imagery. The main aim of the performed study was to evaluate the performance of different convolutional neural network models for multimodal object recognition. For this purpose a special dataset was collected. The dataset contains synchronized visible and thermal images acquired by several sensor based on unmanned aerial vehicle. The dataset contains synchronized color and thermal images of urban and suburb scenes gathered in different seasons, different times of day and various weather conditions. For convolutional neural network training the dataset was augmented by model images created using object 3D models textured by real visible and thermal images. Several convolutional neural network architectures were trained and evaluated on the created dataset using different splits to estimate the influence of training data on object recognition performance.

Manual assembly remains an important task in production with changing requirements for the worker due to i.e. mass personalized products. To keep up with these requirement changes, modern assistance systems are suitable to support the workers. The goals of these assistance systems are to detect errors in the process, guide the worker through new processes and document the process. To achieve this, the assistance system needs to follow each step of the worker reliably. This can be realized with a visual scene analysis based on machine learning. In the presented work, one 3D-sensor (active stereo vision with an additional RGB camera) is used for the scene analysis. The presented work evaluates existing methods for hand localization and hand pose recognition for the application in manual assembly assistance systems. A new procedure for hand localization based on current machine learning techniques is developed specifically for the scenario of recognizing hand joints (also called hand pose) in a manual assembly scenario. Additionally, different methods for hand pose recognition are compared in the application scenario. Based on the results of the developed hand localization method, current hand pose detection methods were evaluated (Dense-Regression and DeepPrior++). While the methods worked well in a scenario comparable to the dataset, they did not perform very well in a manual assembly scenario. Improvements can be made using pixel-wise segmentation or using specific datasets for training containing data from manual assembly scenarios.

Learning to automatically recognize objects in the real world is a very important and stimulating challenge. This work deals with the problem of detecting aluminium profiles within images, using hierarchical representations such as those based on deep learning methods. The use of regional CNN, a conceptually simple, flexible, and general framework for object instance segmentation, allows to exceed the previous state-of-the-art results. This approach efficiently detects objects in an image while simultaneously generating a high-quality segmentation mask for each instance. Neural network training uses ResNet networks of depth 50 or 101 layers. In particular, the training dataset consists of synthetic data generated by CAD files. The Dataset creation process is fundamental: experimental results show that trivial datasets lead to poor detection performance. A rich dataset, instead, including more complex images, allows the network to learn more and better guaranteeing excellent results. How to get more data, if you do not have more data? To get more data, we just need to make minor alterations such as flips, scale or rotations to existing dataset. This process is known as Data augmentation. The performance of the proposed system strongly depends on the dataset used for training and on the backbone architecture used. This why, adopting a strategy that generates a priori a very large dataset the time required to create the annotation file grows almost exponentially. The validation and test dataset, on the other hand, consists of real images captured by cameras. The distinctiveness of this work consists to train a deep neural network with synthetic data (aluminium CAD files) and verify if it on real data. Experiments show that the implementation of architecture, as described above, leads to good performance in automatic detection and classification. Future work will be addressed to improve the network training process together with the architecture, the algorithms and the dataset creation process. The latter is proved to be fundamental for the balance and optimization of the whole process. The way is to develop not much augmented datasets focusing on online data augmentation during network training.

Estimation of stereovision disparity maps is important for many applications that require information about objects’ position and geometry. For example, as depth surrogate, disparity maps are essential for objects’ 3D shape reconstruction and indeed other applications that do require three-dimensional representation of a scene. Recently, deep learning (DL) methodology has enabled novel approaches for the disparity estimation with some focus on the real-time processing requirement that is critical for applications in robotics and autonomous navigation. Previously, that constraint was not always addressed. Furthermore, for robust disparity estimation the occlusion effects should be explicitly modelled. In the described method, the effective detection of occlusion regions is achieved through disparity estimation in both, forward and backward correspondence model with two matching deep subnetworks. These two subnetworks are trained jointly in a single training process. Initially the subnetworks are trained using simulated data with the know ground truth, then to improve generalisation properties the whole model is fine-tuned in an unsupervised fashion on real data. During the unsupervised training, the model is equipped with bilinear interpolation warping function to directly measure quality of the correspondence with the disparity maps estimated for both the left and right image. During this phase forward-backward consistency constraint loss function is also applied to regularise the disparity estimators for non-occluding pixels. The described network model computes, at the same time, the forward and backward disparity maps as well as corresponding occlusion masks. It showed improved results on simulated and real images with occluded objects, when compared with the results obtained without using the forward-backward consistency constraint loss function.

In sheet metal production the quality of a cut determines the conditions for a possible postprocessing. Considering the roughness as a parameter for assessing the quality of the cut edge, different techniques have been developed that use texture analysis and convolutional neural networks. All methods available require the use of appropriate equipment and work only in fixed light conditions. In order to discover new applications in the contexts of Industry 4.0, there is a necessity to go beyond their intrinsic limits as camera types and light condition while ensuring the same level of performance. Taking into account the strong increase of the smartphones features in recent years and the fact that their performance in some respect is now comparable to that of a PC with a middle-range mirrorless camera, it is no longer utopian to think of a new out-of-the-box use of these devices that employs the capability in a new way and in a new context. Therefore, we present a method that uses a mobile device with a camera to guarantee images of sufficient quality that can be used for further processing in order to determine the quality of the metal sheet edge. After the image acquisition of the sheet metal edge in real condition of use, the method uses a trained deep neural network to identify the sheet metal edge present in the picture. After the segmentation a no-reference image quality algorithm provides an image quality index, in terms of blurriness, for the image region of the cut edge. This way it is possible for the further evaluation of the cut edge to only consider image data that satisfies a specific quality, ignoring all the parts of the picture with a bad image quality.

One of the objectives of image processing is to detect the region of interest (ROI) in the given application, and then perform characterization and classification of these regions. In HyperSpectral Images (HSI) the detection of targets in an image is of great interest for several applications. However, edge detection is not considered by the community. Moreover, it is less easy to define what can be a contour in a HSI than in a conventional image where the edges are defined as a rapid change in the intensity of the pixels. In fact, because the spectral data (hyperspectral pixels) consist of several hundreds of values, it becomes more difficult to define the variations between two pixels. One can found in the literature a few edge detection works on multi or hyperspectral images, based on statistical similarity criteria (maximum likelihood, Mahalanobis distance), or based on a geometric measurement approach (study of distances or angles between pixel vectors), some works also adapt to multispectral data some classical convolution filters (Sobel gradient, etc.), or mathematical morphology methods. But all these methods are more suited to the multispectral than the hyperspectral data because they are put in difficulty by the large amount of information known as Hugues phenomenon. In this paper we propose a new multidimensional method, based on tensor modelling of HSI, and using a local approach thanks to subtensors rank estimation to perform edge detection. This could be a good way to select regions of interest which are known to be useful when small target detection is needed. Generally, when ROI containing targets is previously selected, the detection results are better. The ROI selection for large HSI containing objects with large and small sizes, is based on tensorial modelling, and an estimation of local signal subspace rank variations. Examples of the results obtained by the proposed method show that the local variations of signal subspace rank for subtensors of 5x5 pixels on 143 bands centered on each pixel of the image. The processed real-world HSI containing an almost uniform background and 4 rows of different sizes of panels. The obtained results concerning the evolution of the local rank of subtensors centered on each pixel of the HSI, show that the local rank of the signal subspace and the panels are linked. According to these promising and convincing results, in this paper we propose an algorithm for edge detection based on local rank variations.

The objective of this case study is to characterise and to numerically assess the volume of impact damages in composite laminates with multiple non-destructive testing techniques. Reliable and robust assessments of the damage volume will be used as input and validation data for the existing and developing numerical models of composite materials. In everyday work, researchers in the area of composite materials mostly use 2D assessment of the damaged area acquired with conventional inspection methods as ultrasonic c-scan. This practice is supported by the relative simplicity of the data analysis carried in 2D. It is proposed to assess the damage volume in 3D which may serve as a more reliable criterion to improve coupling with the numerical modelling. In this paper micro X-ray computed tomography (CT) was used as the reference technique for the damage volume assessment and comparison with an industrial ultrasonic c-scan. In addition, for this case study, ultrasonic phased array technique with the full matrix capture algorithms was used for the inspection together with shearography (speckle pattern shearing interferometry). A comparison of all techniques referenced to the CT data formed the guidance for their applicability. In addition, an attempt to fuse the multiple assessments was made to support the comparison of 2D and 3D inspection techniques. For the experimental part, carbon fiber reinforced polymer (CFRP) specimens of three layups were manufactured and damaged with 4 different impact energies in the range of 18-45 J. The layups were orthotropic, quasi-isotropic and a novel one adapted for the impact events. In total 12 specimens were inspected with multiple non-destructive testing techniques to numerically assess the volume of the impact damages. 3 additional specimens were used to assess the repeatability of the damage in the specimens with the same layup and impact energy. 3D CT reconstructions were made by a rotation of specimens using a cone-beam system with the voxel side size of 50 μm. A developed automatic segmentation algorithm was used for the damage volume estimation in all specimens. The phased array inspection was made with a full-matrix capture technique with the frequencies of 2.25 and 5 MHz. An available 3D shape shearography setup was used to characterise the damages both with the in- and out-of-plane surface strains with a thermal loading of the specimen. Future developments include the comparison of the inspection results together with the numerical modelling predictions and with residual strength after the impact.

This paper presents an accurate and robust processing pipeline for merging tridimensional datasets related to railway contexts and thus producing a comprehensive 3D model of the monitored scenario. The method is made of multiple modules able to detect the rail tracks and achieve consecutive point cloud registration. A preliminary stage is aimed at filtering out those outlier points and selecting only specific regions of interest from the point cloud. Afterwards, the procedure detects the 3D points belonging to the rail tracks, which can be considered as good candidates for attaining the final point cloud registration. A local analysis for each 3D point is performed by considering a parallelepiped-shaped voxel opportunely centered at the point under investigation. The evaluation of the spatial distributions of points inside the considered volume voxel is performed in order to establish if a seed point lies on the rail head. Further checks enable to reject false candidate points from previous steps by taking advantage of the knowledge about the rail track gauge. Finally, a hierarchical clustering completes the extraction of potential rails. The registration module uses the IterativeClosest-Point method, combined with an algorithm that iteratively reduces the overlapping regions between two consecutive point clouds, for merging the data by using the rail points. The methodology is validated on two different datasets collected by using a stereo camera developed at our laboratory. Final outcomes prove as the proposed approach enables to attain robust and accurate global 3D registration in railway contexts.

The current context of biodiversity loss is particularly marked by the Colony Collapse Disorder (CCD) of honeybees due to multiple causes, toxicological, parasitic and viral. The beekeeper has to face these difficulties in order to maintain the population of bees to save the species but also to make its exploitation profitable. According studies, one can understand what is happening inside the hive by observing what is going on outside. In this context, we propose to individually capture by video the flight trajectories of bees and then characterize the pace of the global activity in front of the hive to infer observations that will be consolidated and made available to apicultural data scientists. Thus bee are detected and tracked using image and video processing methods, then the trajectory are modeled. Then, from the extracted data outcome of the videos, curves are fitted as the ideal trajectories of each bee path in order to study and classify their behaviors. Thus, for each tracked bee, the points of extracted centered positions are time-ordered approximated on a plan. The chosen method interpolates the abscissae separately from the ordinates as time-dependent functions before plotting the parametric curve for each bee path individually. Thus, the abscissae as the ordinates are interpolated using cubic splines. The consecutive points to be interpolated are connected by polynomials of degree three. The first and second derivatives of these polynomials must be connected too. This allows the curve to look more natural by avoiding tingling and convexity discontinuities. Finally, it represents the continuity of the speed of the bees too. Experiments on synthetic and real videos show precise detections of the bee paths. Looking forward, through the collected data, the bee behavior could be understood by using machine learning and the semi supervised method must be one way to proceed.

Electronic Speckle Pattern Interferometry (ESPI) is a full-field technique based on interferometry that has found applications in a wide range of domains mostly related in the field of non-destructive testing. The continuous progress of electronics, image analysis and processing has allowed during the last years this technique to expand its application horizons. In this contribution we review the use of ESPI as optical metrology tool for new industrial applications, biomaterials and non-destructive evaluation technique for novel materials. For the first topic different examples found in literature have been shown and discussed. In fact, ESPI has been successfully employed for assessing the integrity of stainless steel kegs or for measuring the deformation of the surface of a heated mirror, and also for evaluating residual stress fields in pipes. Further interesting studies were about thermomechanical behavior of a Printed Wired Board. Biomedical field can benefit of ESPI technology since ESPI has been employed for measuring surface displacements during inflation testing of ocular tissues or to assessing strain in bone-implant interfaces. Finally, one more example of application regarding non-destructive testing technique describes some interesting results for characterizing novel “green” and thus sustainable composite materials.

The real 3D stereoscopic display system is based on the light field scanning, which is realized by reconstructing the spatial distribution of object light intensity and reducing information redundancy. The entire system is designed with a high-speed projector, a directional scattering mirror, a circular stainless steel carrying disk, a rotating shaft and a micro electric motor. In this paper, the feasibility of developing the display system is discussed in detail by analyzing the principle of 3D display of light field, and the working principle of the system, and the system architecture, and the principle of stereo imaging. The experiment proves that the projection rate of the image matches the rotating circle of the rotating mirror and the rotating speed of the motor, which can be generated a stable 3D image. The accuracy of the image depends on the angle and quantity of the selected pictures. The smaller the angle of the selected pictures, the more number of pictures, and the more realistic 3D image will be presented.

Nowadays, taking advantage of polarization allows enhancing the contrast of a detected object and extending the information of the scene compared to conventional intensity imagery, as the information added by polarimetric images is complementary to intensity ones. Using polarization has a wide interest in space exploration, earth remote sensing, machine vision and biomedical diagnosis, and is extending its use to several applications. Here, we present a basic imaging polarimeter which measures the full Stokes vector of the scene based on a division of time structure, based on a consumer CMOS camera. The polarization state of partially or fully polarized light can be represented by means of the Stokes vector, which is the goal of the measurement. However, due to its nature, the components of the Stokes vector cannot be measured directly, as they must be recovered from a set of intensity measurements. In the paper the results show the measured intensity at 632nm and the recovered Stokes data produced by the determined data reduction matrix at six reference polarization states, as well as the theoretical and recovered Stokes parameters from the calibration experiment. The root mean square (RMS) errors are below 10% . Therefore, this system can provide a well-conditioned data reduction matrix for noise immunity and the spectral range can be widened by using white light and a monochrome camera.

The highest accuracy in image classification is of utmost importance for the industrial application of algorithms based on convolutional neural networks. Empirically, it is sometimes possible to improve accuracy by increasing the size of the training set. In this work, the scaling of the test accuracy versus the size of the training set was studied for different networks. First, a network with a depth of few layers was initialized with random parameters and trained on subsets of images of variable size sampled from the MNIST dataset of handwritten digits and MNIST fashion dataset of clothes and accessories. The scaling of the accuracy versus the size of the training set may be described as the sum of two components: a power law and an offset independent on the size of the training set. Exponent of the power law appears to be the same in both dataset and independent on seeds, initial weights and number of convolutional filters. Then, the scaling of the accuracy versus the size of training set has been evaluated on a dataset of pictures of paintings, sacred icons and sculptures with the goal to correctly classify unknown pictures. The networks chosen are the ones implemented in the machine vision library Halcon 18.11, including two convolutional neural networks with unknown topology and Resnet50, pretrained on industrial images. The scaling of the accuracy versus the size of the training set seems to be compatible with the power law scaling observed on the few layers network trained on MNIST.

With the continuous development of robotic welding automation, the weld seam tracking system plays a decisive role in guiding the welding robot to perform accurate and real-time space welding operations. At present, the seam tracking is mostly performed by a laser with a laser. The laser triangulation method is used to identify and judge the position of the inflection point of the line segment in the image. However, the identification method relies heavily on the processing algorithm of the inflection point position in the image, and the recognition result lacks welding. Sewing space pose information. Therefore, how to obtain the full spatial information of the weld by the collection and processing of the line and the surface is of great significance to the development of the weld tracking. The grating projection technique based on Fourier transform profilometry can realize the full collection of the local information of the weld seam. Through the subsequent algorithm processing of the local features, the whole feature information of the weld seam is continuously extracted to guide the space position and attitude of the weld gun. Avoid the deviation of traditional weld seam tracking due to unmatched factors such as weld joints and bumps in the weld. On this basis, the correlation feature recognition of the collected spatial point cloud is proposed. Based on the proximity probability feature extraction method, accurate and real-time identification and tracking of non-standard welds are realized. Finally, the experimental results show that the weld seam tracking based on Fourier transform profilometry is more complete and accurate for the full-feature information of the weld seam, and has better recognition effect.

Different spatial alignment techniques for multimodal image fusion are presented in the paper. Main feature-based spatial alignment steps and operations as well as cross-correlation techniques are described. An experiment was carried out to compare spatial alignment methods. Experiment results are presented and conclusions are made. The text gives information about spatial alignment application for multimodal systems calibration.

Algorithms based on a clever exploitation of artificial intelligence (AI) techniques are the key for modern multidisciplinary applications that are being developed in the last decades. AI approaches’ ability of extracting relevant information from data is essential to perform comprehensive studies in new multidisciplinary topics such as ecological informatics. For example, improving knowledge on cetaceans’ distribution patterns enables the acquisition of a strategic expertise for developing tools aimed to the preservation of the marine environment. In this paper we present an innovative approach, based on Random Forest and RUSBoost, aimed to define predictive models for presence/absence and abundance estimation of two classes of cetaceans: the striped dolphin Stenella coeruleoalba and the common bottlenose dolphin Tursiops truncatus. Sightings data from 2009 to 2017 have been collected and enriched by geo-morphological and meteorological data in order to build a comprehensive dataset of real observations used to train and validate the proposed algorithms. Results in terms of classification and regression accuracy demonstrate the feasibility of the proposed approach and suggest the application of such artificial intelligence based techniques to larger datasets, with the aim of enabling large scale studies as well as improving knowledge on data deficient species.

This paper describes the development of a transverse deflection measuring system for floating dry docks. The system consists of a set of reference marks and a measurement unit. When the measurement is performed in outside for the inspection of real structures, the influence of external conditions can lead to change the position and the direction of a measurement unit. In this case, the deflection of the object cannot be measured because the effect of the camera movement is included in the measured deflection. The aim of the work is to analyze and estimate the rotation parameters of the measurement unit relative to an arbitrary axis. The result of the estimation rotation parameters can be used to compensate the error caused by the rotation of the measurement unit in the development of electro-optical system for transverse deflection measuring system for floating dry docks.

In Germany there are millions of people in retirement age which are living alone. Approx. 30% of them come into emergency situation because of a fall once a year. If the fall leads to bone fracture, the person isn’t able to stand off by itself and if there is nobody recognizing the situation, the fall can lead to death. Therefore a detector is developed, which can detect a person lying on the ground and send a signal to a family member in case of emergency. The system is built optically, which makes it insensitive to outer factors like humidity or thermal drifts. Vice versa it won’t apply any electromagnetic radiation onto medical devices situated in the room. The whole system consists of a laser, a material which shows birefringence, two polarizers and a detector. The material covers a big area in the apartment, which needs to be watched by a senor. If the senor is under strain, the polarization of the light will change its direction while passing the material and the system will measure an increase of light intensity after the 2nd polarizer. To discriminate an emergency from normal everyday situation, an algorithm is implemented, which asks for three steps to be positive ere the system triggers a signal. If the system has access to an already existing smart home-system, a signal can be activated to inform the person’s next relatives or even an emergency service. The systems works best, if the apartment is built newly