In order to improve the automation and real-time analysis ability of fish resources monitoring in waters， this paper combines YOLOv8X（You only look once version 8 - Extra Large） target detection algorithm， ByteTrack（ByteTrack： a strong baseline for multi-object tracking）algorithm and DIDSON sonar（Dual-frequency identification sonar） data to develop a fast and accurate fish target recognition and counting method. The experimental results show that the combined method of YOLOv8X and ByteTrack is close to the traditional Echoview software recognition accuracy （the deviation rate is only 1.36%）， but the processing time is significantly reduced （ the single line is reduced from approximate 30 minutes to approximate 3 minutes ）， showing strong real-time processing ability and generalization performance. Meanwhile， the stability of the method is verified by repeated experiments， and its reliability in different scenarios is confirmed. The research methods and results of this paper provide reliable technical support for the automatic monitoring of fish resources in waters， and can be widely used in the monitoring and management of fishery resources in a wide range of high frequency.