Multi-Scale Ship Detection in SAR Images Based on Multiple Attention Cascade Convolutional Neural Networks

Abstract

with the development of synthetic aperture radar (SAR) technology, accurate detection of target in SAR images has become a challenging task, such as multi-scale ship detection. Detection of different scale ship target in SAR images is widely used in military and civilian field, but for small ships with few pixels and low contrast, the traditional detection algorithms are difficult to accurately detection. In order to solve the problem of multi-scale ship detection, the multiple attention cascade convolutional neural networks (MAC-CNNs) is proposed. This algorithm based on the YOLOv3 network and attention mechanism, introduces channel attention and spatial attention during the feature extraction stage, and then uses the filtered weighted feature vector to replace the original feature vector for residual fusion. Experiments on the SAR ship detection datasets which including multi-scale ships in various SAR images, and the results shown that the proposed algorithm can detect multi-scale ships in SAR images with extremely high accuracy.