Research and Application of Image Recognition Algorithm Based on Deep Learning

The accuracy of image recognition algorithm is greatly reduced in practical applications. To solve this problem, an architecture combining semantic segmentation and image classification is proposed. U-Net network is used to extract and classify image target regions. At the same time, the image classification algorithm based on group convolution and dual attention mechanism is used for hierarchical classification. LabelMe software was used to mark the target of the open data channel, build the image data set for semantic segmentation experiment, and complete the model test. Experimental results show that the semantic segmentation algorithm can well complete the river task extraction, and under the image classification algorithm based on group convolution and dual attention mechanism, it can also effectively classify the pollution level of river surface, and has high model generalization ability, which can provide reference and support for related research and application in practical application.