A water area illegal intrusion detection algorithm based on yolov3 algorithm modification with higher detection accuracy

Abstract

The water environment is currently facing many problems, and video surveillance technology can prevent many behaviors that damage the water environment, such as overfishing. This paper proposes an improved water area illegal intrusion detection algorithm based on yolov3 algorithm. By introducing a network structure combining residual network and dense network to replace the original residual network of yolo algorithm, it solves the problem of yolov3 algorithm for large targets identify problems with poor results. The algorithm is also verified on the public data Pascal Voc and Data set of illegal water invasion behavior. Compared with the similar one-stage target detection algorithm SSD512 and the original YOLOv3, The map value on the Pascal Voc data set has increased by 4.2% and 0.9% . The map value on Data set of illegal water invasion behavior has increased by 6.4% and 3%, which is a good improvement.