A network compression method based on network architecture search for SAR target detection

With the development of high-resolution satellites, real-time processing of satellite data has become a major development trend. However, due to the limitations of onboard resources, onboard real-time processing faces problems such as insufficient storage space, insufficient operating memory, and high computational complexity. To address the above issues, we propose a Synthetic Aperture Radar (SAR) image detection algorithm based on neural network structure search. By designing efficient and flexible search spaces, search strategies, and subnet evaluation modules, lightweight detection networks that meet specific task requirements can be searched from the original detection network. The proposed method was tested on three SAR ship datasets, and the results showed that the proposed method can reduce model complexity and improve inference speed while keeping detection accuracy nearly lossless.