Research On Feature Extraction And And Detection,Registration Algorithms Of SAR Images

Synthetic aperture radar(SAR)is an active sensor using microwave.SAR is different from the optical,infrared and other sensors in that it can achieve remote imaging with its all-weather and all-time capabilities,not subject to weather and light conditions.And it has found wide applications in national defense construction,national economy and social development.However,due to the influence of the characteristics of speckle noise and geometric distortion,the discrimination of targets of interest in SAR images is reduced,and the robustness and effectiveness of information expansion techniques of SAR image target detection,image registration and change detection is also greatly reduced.As a result,the SAR image's incapability of understanding and interpretation has become a bottleneck restricting the application of SAR.It is,therefore,urgent to develop the automatic interpretation of the SAR image.As one of the key technologies for image understanding and interpretation,SAR image feature analysis and extraction is the link between SAR image acquisition and application,which plays an important role in image engineering.In this paper,we study the feature analysis and extraction of SAR images based on different application scenarios.1.Aiming at the problem of low independence of target features in complex scenes,a detection algorithm for SAR targets combining multiple features is proposed.First,the effective features are extracted,which reflect the physical properties of targets,including scattering intensity,size and boundary change.Then the image background,natural clutter,man-made clutter are eliminated in sequence using the discrimination detection method,which contains two layers,the initial target detection layer and the potential target identification layer.The proposed algorithm can reduce the number of false alarm targets and improve the detection efficiency with a smaller number of features.2.To solve the problem that SAR target features can't be completely described in single scale,a multi-scale feature extraction algorithm for SAR targets is proposed.The multi-scale SIFT features are extracted using the Gauss scale space and d multi-group of seed points;the feature dimension is reduced by extracting the principal component of the multi-scale features;the most optimal parameters of multi-scale factor and number are fixed by calculating the image degradation rate.The multi-scale features contain both the overall target contour information and the image details.At the same time,the traditional SAR target detection based on the training set with a large number of samples has difficulty in sample acquisition,and the detection and matching method for SAR targets based on single sample feature extraction is proposed.The new method can detect targets of interest without enough samples and other prior information.And it is of great significance for detection of targets of interest in SAR images with complex background,especially when the target sample and detection image are at different scales and imaging views.The new method can avoid the calculations and selection of a lot of target samples from different scale and view angles,making it unnecessary to obtain a lot of target training samples.3.The occurrence of speckle noise and geometric distortion within SAR images usually leads to limited effectiveness,challenging the stability of the scale-invariant feature transform(SIFT)and its variants in actual applications.Aiming at this problem,a scheme is proposed to enhance the description of keypoints with improved dominant orientation assignment and support region.In addition,aiming at the problem of false matching and mismatching in the fixed distance ratio matching method,an eliminating interference nearest neighbor distance ratio matching mehotd and an optimal Euclidean distance matrix matching method are proposed,respectively.The proposed matching methods can eliminate the mutual interference between the keypoints,and then reduce the leakage matches,which increase the number of matches greatly.4.The accuracy of orientation assignment in SAR images suffers a great challenge because of serious speckle noise and geometrical distortion.Aiming at this problem,the polar scale-invariant feature transform(PSIFT)descriptor and polar speeded-up robust features(PSURF)descriptor are proposed.The novel descriptors are invariant to rotation without the step of dominant orientation assignment.In PSIFT and PSURF,the polar-transformed support region is adopted to calculate the gradient magnitudes and orientations and further sampled in the radius and angular directions with different scales.Then,the final descriptor is built with the orientation bins covering the omnidirectional space.Furthermore,an improved dual-matching method is proposed to achieve sufficient correct matches with PSIFT and PSURF descriptors,and the matching method can greatly increase the matching numbers in SAR image registration,which is meaningful when more matches are required.