Deep Learning for Target recognition from SAR images

International audienceThis paper deals with the problematic of automatic targetrecognition (ATR) using Synthetic Aperture Radar (SAR)images. In this work, the Deep Learning (DL) architecture isproposed and applied in order to recognize military vehiclesfrom SAR images. We propose mainly in this work the deeplearning algorithms based on convolutional neural networkarchitecture. In the second step and in order to optimizethe convolution of DL steps, we propose to use a convo-lutional auto-encoder which may be better suited to imageprocessing. Its use provides several areas of the best resultsin the presence of noise on shifted and truncated images.To validate our approach, some experimentation results aregiven and compared. The obtained results show that theproposed approach of DL achieves a height recognitionaccuracy of 93%