Efficient automatic target recognition method for aircraft SAR image using supervised SOM clustering

Synthetic aperture radar (SAR) has significant advantages in providing high resolution target images, even in darkness or adverse weather. Nevertheless, human operators find target images difficult to recognize because SAR images are generated using complex-valued radio signals of around 1.0-m wavelength. To address this issue, various automatic target recognition (ATR) approaches have been developed, such as those based on neural network or SVM(Support Vector Machine). Moreover, we have already proposed the efficient ATR method using a supervised self-organizing map (SOM), where a binarized SAR image is accurately classified by exploiting the unified distance matrix (Umatrix) metric. Although this method significantly enhances the ATR performance even with heavily contaminated SAR images, it still has a significant problem requiring enormous calculational demands under expansions of scale and thus cannot handle the ATR issue using more training data. As a solution for this problem, this paper employs the A-star algorithm to accelerate the classification speed, and then newly introduces the constrained learning process in generating SOM, which enhances the robustness to the angular variation in targets. Experimental results validate the effectiveness of our proposed method.