Differentiable Neural Architecture Search for SAR Image Ship Object Detection

SAR image ship detection has important applications in military and civil fields. With the wide application of deep learning in the field of computer vision, more and more SAR ship detection algorithms have applied deep learning. Neural network based on deep learning can automatically extract features instead of hand-crafted features, and has achieved great success. But the design of deep neural networks is time-consuming and complex, which need a lot of professional knowledge and experience. Neural architecture search (NAS) has shown great potential in automated design networks. NAS can find the most suitable architecture on the specified dataset within a given search space. In object detection, the scale of the object varies, which is usually solved by designing feature fusion network. So in this work, we proposed to use differentiable architecture search to search the feature fusion module for SAR ship detection. The experiments on SAR ship detection dataset (SSDD) showed that the discovered architecture can replace the corresponding part of the detection network with 1% performance improvement.