Analysis Of Water Quality Abnormal Stste Based On Convolutional Neural Network And Fish Trajectory Characteristics

Water quality safety is related to national security and social and economic development.It is everyone's responsibility to create a good ecological environment.Current biological monitoring methods mainly rely on visual methods to extract the characteristic information of indicators in water,man-made characteristic factor based on experience knowledge,combining many characteristic factors to form characteristic database of water quality state evaluation,constructing classification model to identify water quality state.But these methods have a lot of drawbacks,artificial tectonic features are slow and limited in number,moreover,the non-linear expression ability of these features is not good enough to make an accurate analysis of water quality.To solve these problems,in this paper,the problem of water quality anomaly analysis based on convolution neural network and fish trajectory characteristics is studied,the convolution neural network is used to extract the original state of the indicator in water,and then the online state analysis is carried out.The main research contents include:1.Adopting biological monitoring method,using computer vision theory to eliminate image distortion and noise signal.The image of fish movement information is preprocessed.Detection of moving fish targets using YOLO algorithm,tracking-Learning-Detection algorithm is used to track the target on the premise of detecting the target,recording position change information to extract fish body trajectory,it lays a foundation for water quality state analysis.An effective way of water quality state analysis is put forward.2.A water quality classification model based on convolution neural network with dense connection is studied.An end-to-end feature extraction network is designed.Using cross-validation to test the accuracy of the algorithm;setting up parallel experiments in multi-group environments,comparing with other water quality evaluation models,it is concluded that: the new model has significant advantages in classification accuracy and feature extraction.3.According to the actual situation of water quality monitoring,a new model updating method is designed.Using the output of the standard model softmax function as feedback,it can judge the deviation between the current model and the actual result more intuitively.Self-calibration in time according to the output deviation of the model,re-trained water quality state analysis model in line with the current actual situation.By comparing the experimental results,it is found that,the water quality classification model based on feedback self-tuning system can evaluate the deviation of the model and adjust it immediately,to restore the accuracy of water quality classification model.Finally,the advantages and disadvantages between the existing models and the proposed methods are analyzed from various perspectives.