A deep learning model S-Darknet suitable for small target detection

Abstract

In order to solve the problem of low accuracy of small target detection in target detection, a small target detection model S-Darknet is proposed. The algorithm is designed based on the Darknet-53 network. First, a new backbone network is proposed, which fully extracts the feature of small objects and adapts to multi-scale detection. Then, in order to enhance the feature information of the target after the fusion, a feature enhancement module is added before each upsampling. Finally, the proposed algorithm was verified on the VOC2007, VOC2012 data sets and the actual data sets of railway freight locks. Experimental results show that this method has high detection accuracy under the premise of ensuring real-time performance.