Research On Fish Detection Algorithm Based On Density Peak Clustering

At present,the development of fishery in China is facing some very severe situations.Due to the continuous expansion of the scope of human economic and social activities,the problem of water pollution has become increasingly prominent,and the frequency of various diseases in fish has been increasing.Fish behavior is very sensitive to their own diseases or changes in the external environment,once the fish behavior is obviously abnormal often indicates the health of fish or water environment problems.In addition,there is still no automatic alternative to the problem of timely and appropriate feeding in large-scale intensive ponds.At present,some scholars have used computer vision technology for fish video analysis to guide fishery culture and improve the quality and efficiency of fishery culture,but there are still many problems,including the following three points:(1)how to accurately track the path of fish.(2)how to identify the abnormal behavior of fish schools on the basis of path-tracking of the fish.(3)how to carry out accurate feeding on the basis of identifying the behavior of fish schools.Aiming at the above problems,this paper mainly carries out the research work from three aspects: the tracking of fish movement,the detection of abnormal behavior of fish and the construction of accurate feeding model of fish.The main contents are as follows:(1)The video motion tracking method of fish group based on genetic algorithm and long-term and short-term memory network is studied.Firstly,aiming at the fish video captured by the camera in the high turbidity pond,the fish detection and coordinate data are obtained by subtracting the sampling frame from the reference frame,image preprocessing and BLOB analysis.Then,the genetic algorithm and short-term memory artificial neural network are used to predict the fish trajectory.The experimental results show that the average absolute error of fishs path prediction using genetic algorithm and LSTM is within an acceptable range,and the calculation accuracy of LSTM method is generally better than that of genetic algorithm,which can provide effective data support for fish video motion tracking and abnormal behavior detection.(2)A fish group abnormal behavior detection method based on density peak clustering is proposed.In view of the sensitivity of current clustering algorithms to outlier information,an improved clustering algorithm LSNN-DPC,which combines outlier detection algorithm LOF and clustering algorithm SNN-DPC,is proposed and applied to fish school abnormal behavior detection.Firstly,the similarity matrix of the fish school is transformed into the distance matrix through the conversion function,and then the proposed algorithm LSNN-DPC is used to detect the abnormal data of the fish path.The experimental results show that all the indexes of clustering algorithm LSNN-DPC are better than SNN-DPC algorithm,and can be used to cluster the path data of most fish in ponds.By analyzing the behavior characteristics of most fish,it is helpful to analyze the relationship between fish trajectories and provide a reliable basis for fishery construction.(3)A precise feeding model of fish groups based on machine learning is built.Taking the acceleration and angular velocity data sets obtained by the data recorder as the data source,the Freeman chain code generation algorithm is used to extract the chain codes of fish escape,swimming,eating and daily activities.Then,the extracted sequence vector set is processed based on discrete Fourier transform,and the individual activity of fish is represented by Fourier descriptor;the accurate feeding model of fish based on multi-layer perceptron neural network is constructed,and the Fourier descriptor is used as its input and the individual activity category of fish as the output to classify fish behavior and distribute feed according to behavior category.The experimental results show that the detection accuracy of the model is 100%,and can effectively support the feeding process based on fish behavior patterns.The research shows that the accurate tracking algorithm of fishs path,the identification algorithm of abnormal behavior of fish and the accurate feeding model of fish have been effectively improved,which has the economic and social value to improve the quality and efficiency of fishery culture.