Using Neural Network To Recognize Fish Active States For Water Quality Monitoring

With the rapid development of social science and technology,the problem of water pollution is serious day by day,and more countries and people are concerned about this.Facing the growing water pollution problem,especially the presence of toxic pesticide substances in the water,the traditional physical and chemical analysis technology,because of its long testing time,can not effectively make a real-time effective warning of water pollution.Therefore,a rapid and efficient biological water quality monitoring and early warning has become an important subject in the field of the safety of water quality research,and it has been widely researched and applied at home and abroad.In this thesis,red crucian carp as a biological water quality warning object in the water samples for automatic identification tracking.In the different types of pesticides(pesticides,chlorpyrifos,dimethoate,etc.),the red carp showed different active state.And the active state of the fish target was judged by the neural network model,according to this we developed the water quality monitoring and warning system.The main contents of this paper are as follows:Firstly,a method of fish target tracking based on selective force attention mechanism is proposed.When a human identifies a target,it always focuses its attention on the target to be identified.According to this principle,the area of attention of the visual target is designed,in which the target can be effectively segmented.In order to continuously track the fish target,a focus area update algorithm is also designed.The experimental results show that the method can effectively track the individual target in the fish group in a complex background.Secondly,a method of judging the active state of fish target based on neural network model is proposed.The extracted fish target is used as the input of the neural network,and the training network is used to identify whether the fish target is active(or dead)according to the error back propagation algorithm.The experimental results show that the model can quickly and accurately judge the state of the fish in the water.The active state of fish targets,which exposure in the pesticide insecticides,chlorpyrifos and dimethoate,can be determined within the set threshold time.This method is equally effective for the presence of dead individuals in the cluster of fish.Therefore,according to the neural network to judge the active state of the fish target of water can be toxic for early warning.Thirdly,a water quality monitoring system based on the determination of fish active state was developed.Based on the Visual Studio 2013 development platform,the OpenCV library of C++ is used to preprocess the fish target video data,and then the recognition,tracking and classification of the fish target are completed.The graphical user interface of the platform is designed based on Qt.The development of this system is currently used in a Shaoxing water company and a Xiaoshan water company.