Threat target image detection method based on two-stage R-CNN network

Military threat target image detection is one of the research difficulties in the field of image processing and target recognition. The characteristics of small target, complex background and similar features make it difficult to identify unmanned intelligent equipment. Aiming at this problem, this paper proposes a new threat target detection algorithm based on two-stage deep network. The algorithm can identify multi-threat target images. Faster R-CNN has been improved using two methods: feature pyramid network and Mosaic image enhancement. The accuracy rate is 7.3% higher than that before improvement, which provides new ideas and methods for military threat target image detection.