Deep Learning Recognition Algorithms For Image Of Microfiber In River

Microplastic pollution has increasingly become a topic of concern.In the water environment,tiny plastic fragments will be carried by microorganisms to migrate into the food chain,unknowingly endangering human health.Microfiber is the main form of microplastics and an important object in the research of microplastics.However,the traditional method of identifying microfibers requires expensive equipment and experienced professionals,which seriously affects the progress and results of microfiber research.Secondly,the accuracy of manually identifying microfibers is greatly reduced.Therefore,this topic is to apply deep learning to microfiber recognition,to study how to effectively learn the characteristics of microfibers,and to improve the recognition rate of microfibers in river.In this paper,two microfiber image recognition algorithms are constructed,which can effectively improve the recognition effect of microfiber in river.First,in order to improve the recognition effect of water microfibers,and at the same time better focus on learning the detailed information of microfibers,a MobileNetV2 algorithm for image recognition of microfiber in river is constructed.For the identification of microfibers,considering the size,winding shape and other factors,the feature reconstruction strategy is adopted in the feature extraction part,the global average pooling is used to obtain the global receptive field,the fully connected layers establishes the inter-channel dependence,and the weighted completion of the microfiber feature reconstruction.In order to strengthen the microfiber width information and obtain the feature area with good detailed information,the pooling fusion strategy is used to downsample the microfiber features at different scales,SO that the network can better learn the microfiber feature information.Second,in order to better obtain important local information that requires microfiber features and further improve the recognition rate of microfibers,a microfiber recognition algorithm with deep feature fusion and reconstruction is constructed.Due to the small size of the microfibers,in order to enhance the key detail information,convolution features and depth separable convolution features are fused.Aiming at the problem of low resolution of deep networks,features are reconstructed in channel and spatial dimensions.At the same time,the original feature map and the reconstructed feature map are weighted to enhance the directionality of important feature information and learn more important features information to improve the efficiency of microfiber recognition.Through network training,experimental results,comparative experiments,and feature visualization comparison analysis,it is proved that the algorithms has a good recognition effect on microfiber images.Focusing on the key information of microfibers is conducive to the effective and stable learning of microfibers by the network Important information.The accuracy of the MobileNetV2 algorithm for image recognition of microfiber in river has reached 97.96%,which is an increase of 2.54%compared with the original MobileNetV2 network.The accuracy of the micro fiber recognition algorithm based on deep feature fusion and reconstruction reached 98.77%.At the same time,compared with other classic networks,the recall,accuracy and F1-scores are effectively improved,and the false recognition rate and missed recognition rate are greatly reduced.