Research on Target Detection Based on Deep Learning

In modern war, the environment is complex and changeable, so how to detect and attack targets automatically and effectively is of great significance. In this paper, through the analysis of the characteristics of targets under complicated environment, six targets, the aeroplanes, bridges, vehicles, ships, submarines and tanks are selected as the objects to be detected, and a large number of corresponding images of them are collected via Internet, and with reference to the dataset format of PASCAL VOC [1], the collected six types of target images are manually annotated to set up the dataset. Then, a corresponding detection model which will be used to detect the six types of targets is built. Base on the dataset built earlier, the detection model is trained and improved by modify its anchors properly.