Synthetic Aperture Radar image target recognition based on hybrid attention mechanism

Abstract

Deep learning algorithm has been more and more applied in the image field, but its application in the SAR image target recognition field is still faced with some problems, such as poor instantaneity and low precision. On this basis, this paper puts forward a convolutional neural network algorithm based on hybrid attention mechanism . The basic module of this model is composed of the trunk branch and the soft branch. The trunk branch composed of the residual shrinkage network and the improved channel attention mechanism is responsible for extracting the main characteristics. Soft branch composed of up sampling and down sampling is responsible for extracting the mixed attention weight, which can enhance the mapping capacity from input to output. The recognition rate of MSTAR dataset with this model is 99.6%. According to noise analysis, this model is of strong robustness for images with impulse noise added .