Research On Fish Behavior Monitoring Method Based On Visual Perception Technology

China is a big fishing country,and aquaculture has developed rapidly.In terms of breeding environment,circulating water breeding is gradually replacing the traditional breeding mode because of its energy saving and high efficiency.At present,with the development of image processing technology and the promotion of neural network technology,visual perception technology has been applied in many fields.Among them,using visual perception technology to monitor and analyze fish behavior in aquaculture has become a hot spot in the research field.However,in the current research,visual perception technology is only applied to the ideal experimental environment,and there are few relevant studies in the industrialized circulating water culture system,and there are few studies on the analysis of fish behavior characteristics.In view of the above situation,this paper proposes the research on fish behavior monitoring method based on visual perception technology,which makes up for the shortcomings of strong subjectivity and low accuracy of fish behavior monitoring relying on manual observation and empirical judgment in the past.The main research contents and results are as follows:1.identifying abnormal behaviors of fish is an important part of aquaculture health monitoring.In view of the low efficiency of manual inspection during breeding operation,combined with visual perception technology,this paper proposes a method to judge the abnormal trajectory in the behavior of cultured fish clusters.This method effectively combines deep learning with fish behavior,fish recognition is carried out through YOLO algorithm,fish swarm is tracked in real time based on sort algorithm,and the swimming trajectory line of fish swarm is drawn,and then cluster analysis is carried out by using the swimming characteristics of fish swarm,And use it to quantify and identify the swimming anomaly.2.In view of the lack of quantitative research on feeding activity in the actual breeding scene,a characterization method of fish feeding activity suitable for circulating water breeding plants is proposed.This method uses the information of average velocity and dispersion of fish to quantify and characterize the real-time feeding activity of fish,and uses Deep SOCIAL algorithm to visualize the density of fish.3.Aiming at the problem of identifying aggressive behavior in the process of red fin Fugu breeding,an aggressive behavior identification method based on yolov5 rotation detection model is proposed.This method designs a fish target detection model based on rotating rectangular frame through angle discretization,definition of rotating target loss function and other related technologies.Combined with the actual attack behavior characteristics of red fin fugu,the attack behavior is determined by axis angle.Finally,taking the actual culture image as the data set,the fish behavior monitoring method proposed in this paper is verified.The results show that the research on fish behavior identification method based on visual perception technology proposed in this paper can effectively identify the abnormal trajectories in the swimming of fish schools,and has lower missed detection and false detection rate and better effect than the original target tracking algorithm.The discrimination method can also accurately measure the changes of swimming speed and dispersion before and after feeding,and visualize the density distribution map of fish schools.In the identification of attack behavior of Takifugu rubripes,this method can effectively calculate the distance information and angle information of fish group,so that the breeders can know it in time.