YOLO-based Detection Technology for Aerial Infrared Targets

The anti-interference detection of traditional infrared imaging air-to-air missiles is limited by the experience of artificial feature design, and it is difficult to cover all air combat against environmental problems. Therefore, the deep learning method was proposed, and used to extract features autonomously and achieve target detection under interference environment. The target detection principle of YOLO network was analyzed, the backbone network of YOLOV2 was rebuilt by using the densely connected convolutional network, the ability of network to extract and transmit feature information were improved, and the network performance was improved. The simulated infrared countermeasure data was used as the training set and test set of the network, and the network was retrained. The test results show that the precision of the YOLOV2-D network is higher than 99%, which is 7.5% higher than YOLOV2. The recall rate of the YOLOV2-D network is higher than 96%, which is 2% higher than YOLOV2. The detection speed reaches 180 frames per second, which is three times that of YOLOV2. Even in the case of overlapping targets, the YOLOV2-D network can detect targets accurately and quickly.