Research on Object Detection Algorithm Based on Deep Learning

Abstract

The current target detection models have the characteristics of large storage, large demand for computing resources and large number of parameters, which are difficult to be implemented on the platform with low computing performance and small storage capacity. In order to reduce the size of the model and improve the detection speed, this paper proposes a new network architecture of mobilenetv2-yolov5s by combining the lightweight network mobilenetv2 with yolov5s Compared with other target detection algorithms, the improved yolov5s has better detection effect. The mobilenetv2-yolov5s network is tested on MS coco data set, and the mAP value is 55.1. While ensuring the map, the detection speed of the algorithm is 31fps, which is 25fps higher than yolov5s.