Research On Data Enhancement And Cassava Leaf Disease Classification And Recognition System Based On LeafGAN

China is an important importer of cassava products.Cassava was introduced and cultivated in the 1920 s,and is mainly distributed in Guangxi Zhuang Autonomous Region,Guangdong Province,Hainan Province,and other places in China.The status of cassava industry in agricultural efficiency and farmers’ income has become increasingly prominent,but cassava production efficiency has been increasingly volatile and declining,and cassava planting risk is one of the main reasons.Factors such as climate,geography,temperature,diseases,and soil moisture can all affect the yield of cassava.Among them,the invasion of diseases can reduce the yield and quality of cassava,making China’s cassava uncompetitive in the international market.Therefore,the identification of cassava diseases is very important.During the growth process of cassava,most of the infected diseases will be characterized in the leaves.Aiming at the identification problem of possible leaf diseases in cassava,this article takes four common leaf diseases,namely bacterial blight,brown stripe disease,green mottle disease,and mosaic disease,as the research object,and adopts data enhancement based on improved LeafGAN,A cassava leaf disease detection method based on the YOLOv5 v6.2 classification and recognition algorithm of the CBAM attention mechanism module was studied.In terms of data enhancement,the unlabeled leaf segmentation module(LFLSeg module)proposed by LeafGAN can be used to identify disease-related areas in leaf images,effectively segmenting them from the background and transforming them into the target domain,helping the LeafGAN model pay more attention to ROI(regions of interest).Not only can high-quality images be generated,but these generated new images can also be added as training resources,Thus improving the recognition accuracy of the network model.In terms of cassava leaf disease identification and detection algorithms,a deep convolution neural network based cassava leaf disease identification method was explored based on the characteristics of cassava leaf disease datasets.The YOLOv5 network structure has been adjusted and optimized,and an attention mechanism module has been embedded in the training of the network model.At the same time,experiments were conducted to compare different network structures and training methods,and the effects of different optimizers,learning rates,iterations,and batch sizes on recognition accuracy were studied.The accuracy rate for identifying four types of cassava leaf diseases can reach 97.34%.Finally,an identification application system was developed based on the We Chat applet developer platform,providing a mobile cassava leaf disease identification method that can quickly and accurately identify in the field.The system implements functions such as disease photo shooting,disease identification,and expert guidance,which has reference value for cassava leaf planting and disease control.