Multi-class objects detection in remote sensing image is attracting increasing attention recent years. In particular, the method based on deep learning has made outstanding achievements in the object detection. However, the deep network is easy to overfitting for the insufficient of remote sensing image dataset. What’s more, the current deep learning- based methods of object detection in remote sensing image usually ignore the context information of the objects. To cope with these problems, a novel object detection method based on regularized convolutional network and context information are proposed in this paper. A form of structured dropout method is used in convolutional layers to dropping continuous regions. To address the problem of lack of context, spatial recurrent neural networks are used to integrate the contextual information outside the region of interest. Comprehensive experiments in a public ten-class object detection data set show that the proposed object detection method has an outstanding detection accuracy under different scenarios.
KEYWORDS: Target detection, Time-frequency analysis, Clouds, Signal to noise ratio, Fourier transforms, Submerged target detection, Linear filtering, Interference (communication), Sensors, Digital breast tomosynthesis
The signal-to-clutter ratio couldn't be improved effectively for conventional spatial or temporal high-pass filter is
difficult to remove clutter, especially clutter edge, and a dim moving target detection method based on the timefrequency
characters difference among target, noise and clutter is presented in this paper. Theoretical analysis shows that
the waveform at the target location in time-frequency domain is a small wave packet, the magnitude of the wave packet
is consistent with the target amplitude, and the width of the packet is inversely proportional to target speed; the
waveform at clutter edge is an "uphill" or "downhill". This target detection method takes two-stage filters to detect dim
target. At first, a threshold based on false alarm ratio criteria in the time-frequency is adopted to remove noise, and then
the ratio between "main lobe" and "side lobe" in the wave packet is counted to remove clutter and detect target. The dim
target detection experiment under cloud is included and the result shows that the method is effective.
While tracking dim and small moving targets in the electro-optical (EO) tracking system, the numerous false alarms resulted from the low signal-to-noise ratio would seriously debase the performance of target recognition and tracking. The probabilistic data association filter in conjunction with a maximum likelihood approach (PDAF-ML) has been applied effectively to low observable or dim target motion analysis. Whereas, the PDAF-ML supposes that the amplitude of target is not correlative among different sampling instants, and that the greater the amplitude value is, the greater the probability of being the target of interest would be. In the EO imaging tracking system, the amplitude information and the motion of target are consistent and highly correlative in a short period. To resolve the problem that the PDAF-ML is inconsistent with the EO imaging tracking system, the two features, namely, the amplitude information and the motion as well as their consistency, are modeled as Markov stationary random signals and are fused by means of PDAF. Experiments are carried out, and the results show that, with the proposed approach, the uncertainty of trajectory association would be largely decreased, and the performance of target recognition and tracking could be significantly improved.
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.