Design And Implementation Of River Surface Pollution Identification System Based On Deep Learning

With the development of "Carbon Neutrality","Peak Carbon Dioxide Emissions" and high-quality development goals,the state is gradually increasing her efforts to protect the environment.Effective protection of the river environment can improve the ecological carbon sequestration capacity of rivers and improve the quality of social development.River pollution monitoring is an important part of river environmental protection.At present,the main means of monitoring pollution in rivers is water quality testing by monitoring station sensors,while the surface pollution is mainly detected and dealt with by regular manual inspections,which is inefficient,costly and dangerous.In order to improve the river pollution monitoring capability,this paper designs and implements a river surface pollution identification system using river water surface images taken by river monitoring cameras deployed by environmental protection departments,from the new perspective of river water surface,combined with deep learning technology,that can detect pollutants in a timely manner,reduce environmental protection costs,protect river resources,and provide a new solution for river pollution monitoring.The main work of this paper is as follows:(1)In order to exclude the influence of interference factors outside the river region on river pollution class classification,an improved U-Net-based river semantic segmentation algorithm is proposed to extract the river region in the image as the data picture for subsequent river pollution class classification.The improved U-Net adds multi-scale feature fusion to the downsampling part of the network to fuse the highlevel extracted features with the underlying features to enhance the semantic,spatial representation of the feature MPA.In addition,a Dupsampling structure is used to decode the feature MPA in the upsampling part of the network to improve the efficiency and accuracy of feature upsampling.Experiments show that the improved U-Net network works better than the original U-Net,FCN and Seg Net in the river semantic segmentation task.(2)To improve the accuracy of the model in classifying river pollution levels,the Grouped convolution dual attention(GCDA)module is proposed and deployed on Res Net50 to classify the pollution levels of river water surface conditions.The use of grouped convolution reduces the number of parameters and computational effort of the network,and the combination of both channel and spatial attention mechanisms enhances the feature extraction ability of the network,thus improving the image classification results.Experiments show that the algorithm has significantly improved its effectiveness in classifying the pollution level of river water surface compared with the original Res Net50,Res Net50 with the addition of spatial attention mechanism or channel attention mechanism respectively.(3)With the improved U-Net and Res Net50 deployed with the GCDA module as the core algorithm,a river surface pollution identification system is built through four steps of demand analysis,system design,system implementation and system testing,which realizes video monitoring,system management and river pollution identification,and other functional modules,which can identify and give early warning to monitoring and acquisition images in real time.It provides a new solution for river pollution monitoring.