一种基于改进DeepSORT的淡水环境下鱼类计数方法OA

This paper proposes a fish counting method based on an improved DeepSORT algorithm for freshwater environments.Underwater fish videos are captured,sampled,labeled,and used to analyze the adaptability of different image enhancement algorithms to turbid underwater environments,and the optimal enhancement method is selected to assist in improving detection accuracy.In the tracking and counting process,we replace the detector of the DeepSORT algorithm with the YOLOv4 detection algorithm to improve detection speed,and use complete intersection over union(CIoU)instead of intersection over union(IoU)in cascade matching to enhance the accuracy of target matching.When this method is applied to track and count fish in underwater videos,the results show excellent tracking and counting performance.In a 20 s video,the average counting accuracy of the algorithm reaches 91.6%,with a detection speed of 20.83 ms per frame.The results show that this method is suitable for the evaluation of fish yield in aquaculture environments.