SSD Target Detection Algorithm Based on Multi-Scale Fusion and Attention

Abstract

Aiming at the problems of weak effective information in feature maps and high miss-detection rate of difficult targets when traditional SSD target detection algorithms perform target detection, we propose an improved SSD target detection algorithm. First, add a CBAM module after each feature layer of the SSD. CBAM is a hybrid module that combines spatial attention and channel attention. This module strengthens the network's ability to discriminate targets and backgrounds, improves the expression of effective feature weights, and suppresses interference from irrelevant information; then, adopt the idea of FPN to construct a feature fusion module, which effectively integrates feature layers of different scales, thereby improving the network's ability to detect difficult targets. Verifying the method proposed in this paper on the PASCAL VOC data set fully proves that the improved network performance has been greatly improved.