Research And Implementation Of Rice Leaf Disease Detection Based On Deep Learning

Rice is an important food crop in China,and its yield has been restricted by diseases.It is particularly important to identify the types of rice diseases quickly and accurately.Traditional identification is mainly based on eye observation,which is subjective to a certain extent and requires in-depth field observation in rice fields,which is time-consuming and labor-intensive,and cannot meet the real-time requirements of rice disease identification.In recent years,deep learning technology has developed rapidly.How to apply deep learning to crop disease recognition has become a research hotspot.In this paper,based on the target detection algorithm,deep learning technology and methods are introduced to carry out research from the actual needs of rice disease detection.In order to solve the problem of low accuracy and efficiency in rice disease detection,this paper designed a lightweight YOLOv5 network model,which can fully extract the feature information of images,maintain high average detection accuracy and detection speed in the face of complex or small targets of rice disease images.In the research process,firstly,the YOLOv5 algorithm is selected as the basic network model,and the lightweight MobileNetV3 network is introduced into the network model.The standard convolution of CSPDarkNet53,the backbone network of YOLOv5 algorithm,is replaced by the deeply separable convolution,the number of parameters and computation are greatly reduced,and the detection speed is significantly improved.At the same time,the residual block structure of backbone network CSPDarkNet53 is replaced by the inverted residual block structure,which enhances the feature extraction ability of the model and improves the detection accuracy of the model.Secondly,the feature fusion method of YOLOv5 model is improved by adopting channel enhanced feature pyramid network and introducing channel attention mechanism to improve the accuracy of small target detection.Finally,in the regression box screening stage,the regression box screening method NMS is replaced by WBF method,which significantly improves the average detection accuracy of the network model.Through experimental comparison,it is proved that the lightweight network model designed in this paper has higher average detection accuracy and detection speed for rice diseases,with less parameters and computation,and can be deployed on mobile devices such as mobile phones,with strong generalization ability.At the same time,using deep learning technology to realize automatic recognition and detection of rice diseases can help rice growers quickly and accurately identify the types of rice diseases,which has very important practical significance for guiding the prevention and control of rice diseases.