The Design And Implementation Of Activated Sludge Microbial Detection System Based On Visual Computing

With the increasingly serious environmental problems,the prevention and control of water pollution has become the focus of attention of the state and society.At present,most of the urban sewage is treated by biological treatment,and the treatment results are detected by the way of microscopic examination of activated sludge.However,the traditional microscopic examination of activated sludge is mostly manual detection,which can not process and analyze the data quickly.In recent years,as one of the important branches of artificial intelligence,computer vision has developed rapidly.The visual computing method is introduced into the process of microbial microscopic examination of activated sludge,which can quickly detect the existence,disappearance and movement characteristics of indicator organisms reflecting the water quality status,and detect the water quality status more efficiently.Firstly,through the researches of moving object detection methods,the application scenarios,advantages and disadvantages of the algorithm are analyzed and compared.The first step of microbial vision calculation is realized by combining the inter frame difference method with the background modeling method.The inter frame difference method is used to determine the moving target area,and the background modeling method is used to extract the complete target;Secondly,Res Net network is selected as the classification model of microbial identification,and the target classification is realized by deep learning,and the output layer structure of the network is adjusted according to the actual number of microbial species.Due to the limitation of the existing data sets,the data enhancement operation is carried out to improve the accuracy of the classification model;Finally,local image features need to be extracted before moving object tracking is realized by online target detection.After comparing and analyzing the limitations of general feature extraction,the Hilbert curve based feature extraction method is selected to complete the local image feature extraction.Then,partial least squares method is used to establish voting regression model and label regression model between the local feature set and the target location information to realize offline target detection.The off-line target detection is further extended to update the existing model by using the feature information of a new frame in the video sequence,and finally realize the target tracking based on online detection.Recording and collecting videos of different scenes and complexity to test the system,the results show that the accuracy of microbial target detection is about 92.9%;the accuracy of microbial classification and identification is about 86%;the accuracy of microbial tracking can reach 85% at last.The processing speed basically meets the requirement of real-time,and can be applied in actual projects.