Image classification behind complex inhomogeneous media is a pervasive problem in computational optics. In recent years, optical neural networks have shown high accuracy and little computation costs in image classification. However, the improvements in scalability and complexity are still challenging. This paper presents an optronic speckle transformer (OPST) for image classification through scattering media. We utilize the optical self-attention mechanism to extract the speckle pattern’s local and global properties. We realize excellent speckle classification results with minimal computation costs based on OPST. The OPST improves the classification by more than 8% and reduces the network’s parameter by more than 30%, compared with optronic convolutional neural networks (OPCNN). Moreover, our OPST demonstrates high scalability with existing optical neural networks and is adaptive to more complex tasks. Our work paves the way to an all-optical approach with less computational costs for object classification through opaque media.
Object classification behind a complex inhomogenous medium remains a significant challenge in many fields. Valid information is hardly extracted from speckles owing to the distortion of scattering media. Recent years deep learning has shown powerful capability in classifying object from scattered speckle patterns. However, largescale computations and pure digital procedures set a challenging task for deep neural networks running in optics. Here, we present an optronic technique for object classification through random diffuser media. A group of Fourier lens and a programmable spatial light modulator, form an optronic convolutional neural networks(OPCNN) with optimized kernels. The CMOS camera not only works as an image detection sensor but a non-linear activation layer by the curve built-in. We demonstrated the technique classification result using the airy disk intensity of every candidate channel. The trained OPCNN shows high-quality object predictions on the speckle patterns. Our work paves the way to an all-optical approach for imaging through scattering media.
Deep learning technology has been widely used in synthetic aperture radar automatic target recognition (SAR ATR) tasks due to its good performance and high efficiency. However, the development of semiconductor technology cannot completely satisfy the demand for electronic hardware with high computational capability to conduct deep neural networks. To solve this dilemma, we propose an optronic convolutional neural network (OPCNN) that can perform SAR ATR tasks directly in optical platform. In OPCNN, all computational operations are implemented in optics with high speed and low energy cost. Electronic platforms are used only to control devices and transmit data information without a massive computational burden. As in digital CNNs, the convolutional layer, downsampling layer, nonlinear activation layer, and fully connected layer are all contained in OPCNN. The simulations demonstrate the feasibility of our OPCNN in solving SAR ATR problems. The good performance in experiments, which achieves 87.4% and 93.8% recognition accuracy on original and denoised moving and stationary target acquisition and recognition dataset, validates the application ability of OPCNN in practical scenario. In one-time recognition tasks, the processing time of our OPCNN is only 0.26 s with the speed of light and the power consumption of our prototype is also far less than digital processer, which is 954 W. Through analysis, our OPCNN obtains the higher processing speed and lower energy cost than digital CNNs with the same structure due to the advantages of optical technology. Also, the scalability of optical structure contributes to build more complex networks to solve complicated dataset without the demand of advanced electronic hardware.
Scattering media would scratch light propagation, and images would degenerate into unrecognizable speckle patterns. Conventional target recognition through scattering media is composed of two steps, i.e., reconstruction and recognition. Here, combining the compressive sensing with feature extraction, a method of efficient speckle based compressive target recognition through scattering media is proposed. In the paper, autocorrelation of speckles is proved to have the same singular values as that of their corresponding objects, and then speckle based recognition is introduced. Compressive sensing can be used to retrieve signals with measurements fewer than those required by Nyquist-Shannon theory. With the proposed method, scattered object recognition can be replaced with speckle recognition, bypassing the conventional object reconstruction procedure. Performances are validated through relevant experiments. Besides, benefited from the conclusion, domain adaption based support vector regression method is proposed and utilized for imaging through scattering media then. Domain adaption is introduced to transfer leaning samples and testing samples into a new space where the distance between them is much closer, leading to high reconstruction fidelity in the followed support vector regression based inverse scattering stage. Principle component analysis is also considered to help decrease dimension and thus improving efficiency. Experiments validate that the presented technique owns a higher image reconstruction efficiency and fidelity, compared with our previous researches. Since the target recognition and reconstruction is mainly based on ground truth images, the work is valuable and meaningful for remote sensing applications, especially for object detection or monitoring when scattering is occurred.
A clear image of an observed object may deteriorate into unrecognizable speckle when encountering heterogeneous scattering media, thus it is necessary to recover the object image from the speckle. A method combining least square and semidefinite programming is proposed, which can be used for imaging through scattering media. The proposed method consists of two main stages, that is, media scattering characteristics (SCs) estimation and image reconstruction. SCs estimation is accomplished through LS concept after establishing a database of known object-and-speckle pairs. Image reconstruction is realized by solving an SDP problem to obtain the product of the unknown object image and its Hermitian transposition. Finally, the unknown object image can be reconstructed by extracting the largest rank-1 component of the product. Structural similarity (SSIM) index is employed as a performance indicator in speckle prediction and image reconstruction. Numerical simulations and physical experiments are performed to verify the feasibility and practicality of the proposed method. Compared with the existing phase shift interferometry mean square optimization method and the single-shot phase retrieval algorithm, the proposed method is the most precise to obtain the best reconstruction results with highest SSIM index value. The work can be used for exploring the potential applications of scattering media, especially for imaging through turbid media in biomedical, scattering property measurement, and optical image encryption.
Motion compensation (MOCO) is an essential procedure for synthetic aperture radar under a complicated flight path. Wide swath and low-precision navigation data limit traditional MOCO algorithms. We present a space-variant MOCO method to solve two problems: (1) estimation of the deviations of radar during data acquisition and (2) range envelope correction (REC) before range cell migration correction. In the first problem, we apply phase-gradient autofocus with correlate input and weighted total least square jointly. In the second one, the chirp-scaling-based (CS) REC is proposed. Finally, both simulation and real data experiments validate the proposed method.
Target object image would deteriorate into unrecognizable speckle pattern when encountering with scattering media. It’s necessary to reconstruct the target object image from captured speckle. In this paper, a method combining correlation method and oversampling smoothness is proposed. It is used for target object reconstruction from scattered speckle pattern. The reconstruction is based on the Fourier transform of the target object. The Fourier amplitude of the target object can be calculated through an inverse Fourier transform of the autocorrelation of captured speckle pattern. The Fourier phase can be recovered with oversampling smoothness method. Experiments were used to comparing the proposed C-OSS method with the C-ERHIO method. The results indicate that the proposed method improves the reconstruction with lower background components than the other. The proposed method can also be applied to optical image encryption et al.
Scattering media would deteriorate an object image into unrecognizable speckle pattern. Machine learning is introduced to reconstruct the object image from speckle pattern. In the proposed method, a database containing two groups (i.e., face image-and-speckle-pattern pairs, non-face image-and-speckle-pattern pairs) is firstly established. Then support vector classification (SVC) is introduced to classify a given unknown speckle pattern into which group it belongs to. Taking advantage of support vector regression (SVR), the object image corresponding to the unknown speckle pattern can be reconstructed. Experiments are conducted to verify the effectiveness of the proposed method, as well as the necessity of the introduction of SVC.
Due to the presence of atmospheric turbulence, motion error (ME) arises and causes residual azimuth phase error (APE) during synthetic aperture radar (SAR) data acquisition. APE can degrade SAR images, especially for light-weight SAR. Moreover, different kinds of APE have different impacts on the image, which makes it hard to compensate for. A parametric autofocus based on a cost metric consisting of the modified entropy and the residual entropy (MERE) is developed to compensate the APE. This approach using the optimization transfer method aims to minimize the MERE. The polynomial decomposition is applied to fit the low-order APE while inverse discrete cosine transform model is adopted for the high-frequency case. Additionally, we also design a modified adaptive-order search strategy, and it helps to remarkably reduce the computational load while maintaining accuracy. In the case of correcting high-frequency APE, the MERE metric could effectively avoid the over-fitted problem that arises in entropy-based autofocus. The real airborne SAR data experiments and comparisons demonstrate the validity and effectiveness of the proposed autofocus.
The configuration of multiple uniformly spaced channels in azimuth can achieve high-resolution and wide-swath images for synthetic aperture radar (SAR) systems. The unambiguous Doppler spectrum can be reconstructed via digital beamforming (DBF) techniques in the azimuth multichannel SAR system. However, the performance of DBF deteriorates significantly because of the inevitable channel errors in practical applications. Since the SAR antennas are generally unweighted in azimuth, the power of the SAR signal is symmetrically distributed around the Doppler centroid in the azimuth frequency domain. Based on this observation, the covariance matrix of multichannel output in the range-Doppler domain can be regarded as a real matrix when there is no existence of channel errors and the correction of the baseband Doppler centroid is applied. Consequently, we propose a fast calibration algorithm to calculate channel phase errors. The proposed method can acquire the estimation of phase errors directly from the covariance matrix. Thus, the proposed method has the properties of high robustness and low computation load. Theoretical analysis and experiments validate the performance and the efficiency of the proposed algorithm.
Multichannel synthetic aperture radar systems are usually employed to suppress azimuth ambiguity and realize high-resolution wide-swath imaging. However, unavoidable array errors will significantly degrade the performance of ambiguity suppression and imaging quality. This paper presents an array error estimation method based on cross correlation. First, unambiguous Doppler spectra are obtained by selecting a short length of range profiles from strong targets. Then, array errors are estimated by a proposed cross-correlation method. Finally, a preprocessing method to improve the estimation accuracy is proposed. The proposed method takes full advantage of the training samples from strong targets and estimates array errors by the coherent integration technique, which improves the estimation accuracy and robustness. Theoretical analysis and experiments based on simulations and measurements showed the validity of the proposed method, especially in low signal-to-noise ratios.
A spectrum reconstruction algorithm based on space–time adaptive processing (STAP) can effectively suppress azimuth ambiguity for multichannel synthetic aperture radar (SAR) systems in azimuth. However, the traditional STAP-based reconstruction approach has to estimate the covariance matrix and calculate matrix inversion (MI) for each Doppler frequency bin, which will result in a very large computational load. In addition, the traditional STAP-based approach has to know the exact platform velocity, pulse repetition frequency, and array configuration. Errors involving these parameters will significantly degrade the performance of ambiguity suppression. A modified STAP-based approach to solve these problems is presented. The traditional array steering vectors and corresponding covariance matrices are Doppler-variant in the range-Doppler domain. After preprocessing by a proposed phase compensation method, they would be independent of Doppler bins. Therefore, the modified STAP-based approach needs to estimate the covariance matrix and calculate MI only once. The computation load could be greatly reduced. Moreover, by combining the reconstruction method and a proposed adaptive parameter estimation method, the modified method is able to successfully achieve multichannel SAR signal reconstruction and suppress azimuth ambiguity without knowing the above parameters. Theoretical analysis and experiments showed the simplicity and efficiency of the proposed methods.
KEYWORDS: Radar, Signal detection, Target detection, Interference (communication), Signal to noise ratio, Doppler effect, Radar signal processing, Detection and tracking algorithms, Signal processing, Receivers
An optimal radar waveform-design method is proposed to detect moving targets in the presence of clutter and noise. The clutter is split into moving and static parts. Radar-moving target/clutter models are introduced and combined with Neyman–Pearson criteria to design optimal waveforms. Results show that optimal waveform for a moving target is different with that for a static target. The combination of simple-frequency signals could produce maximum detectability based on different noise-power spectrum density situations. Simulations show that our algorithm greatly improves signal-to-clutter plus noise ratio of radar system. Therefore, this algorithm may be preferable for moving target detection when prior information on clutter and noise is available.
A computational method for suppressing clutter and generating clear microwave images of targets is proposed in this paper, which combines synthetic aperture radar (SAR) principles with recursive method and waveform design theory, and it is suitable for SAR for special applications. The nonlinear recursive model is introduced into the SAR operation principle, and the cubature Kalman filter algorithm is used to estimate target and clutter responses in each azimuth position based on their previous states, which are both assumed to be Gaussian distributions. NP criteria-based optimal waveforms are designed repeatedly as the sensor flies along its azimuth path and are used as the transmitting signals. A clutter suppression filter is then designed and added to suppress the clutter response while maintaining most of the target response. Thus, with fewer disturbances from the clutter response, we can generate the SAR image with traditional azimuth matched filters. Our simulations show that the clutter suppression filter significantly reduces the clutter response, and our algorithm greatly improves the SINR of the SAR image based on different clutter suppression filter parameters. As such, this algorithm may be preferable for special target imaging when prior information on the target is available.
With the high programmability of a spatial light modulator (SLM), a newly developed synthetic aperture radar (SAR) optronic processor is capable of focusing SAR data with different parameters. The embedded SLM, encoding SAR data into light signal in the processor, has a limited loading resolution of 1920×1080. When the dimension of processed SAR data increases to tens of thousands in either range or azimuth direction, SAR data should be input and focused block by block. And then, part of the imaging results is mosaicked to offer a full-scene SAR image. In squint mode, however, Doppler centroid will shift signal spectrum in the azimuth direction and make phase filters, loaded by another SLM, unable to cover the entire signal spectrum. It brings about a poor imaging result. Meanwhile, the imaging result, shifted away from the center of light output, will cause difficulties in subsequent image mosaic. We present an SAR image formation algorithm designed to solve these problems when processing SAR data of a large volume in low-squint case. It could not only obtain high-quality imaging results, but also optimize the subsequent process of image mosaic with optimal system cost and efficiency. Experimental results validate the performance of this proposed algorithm in optical full-scene SAR imaging.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.