Research On Corn Disease Detection Based On Improved YOLO Algorithm

Corn yields have been affected by pests and diseases,which in turn affect people’s lives.Corn has good nutritional value and medicinal value,and corn can survive in harsh growing environment,in the first half of the 16 th century,it was introduced into China from Europe in the first half of the 16 th century.China has been a big agricultural production since ancient times,so corn has been one of the most important grain crops because of its nutritive value.With the gradual maturity of the idea of replacing human beings with machines and the development of science and technology,the method of disease identification by artificial intelligence has become a major innovation point and developed rapidly.This technology can liberate the labor force,accurately identify,and provide effective and quick help for agricultural disease and pest management.Therefore,the use of computer vision technology to identify diseases,improve the productivity of farmland,and can detect and diagnose diseases in the early stage has a very important significance.In this thesis,based on the onestage detection algorithm,YOLOv5 s target detection algorithm,the algorithm is improved,and the actual application of maize disease detection is carried out to complete the task of disease detection.The contents of this thesis are as follows:First,collect and enhance the data set,process the data set through mirroring,affine transformation and other methods,and divide the processed data set into training set and test set in a ratio of 9:1.Secondly,according to the problems of backward maize leaf disease identification technology,low efficiency and high precision,YOLOv5 s algorithm was improved by two methods.The first improvement method improves the prior box and the convolution module:the most appropriate prior box is obtained through the method of data cluster analysis,so as to make the disease target location more accurate and improve the detection accuracy.A new convolution module is introduced to accelerate the budget speed of the model.Experiments show that the first improved method can effectively improve the detection speed,reduce the parameters and complexity of the model,and slightly improve the accuracy of the model.The second improvement method mainly improves the feature fusion network,attention mechanism and loss function: improved the feature fusion network,added cross-scale connection,fused more features,and added a small target detection layer to enhance the effect of feature fusion.Introducing attention mechanism to improve model performance;The loss function is improved to make the model focus more on the difficult-to-classify samples and reduce the influence of negative samples on the algorithm training.Experiments show that the second method increases the complexity and size of the model very little,improves the detection speed of the model,and has a better effect in the frame of the target more accurately,higher confidence in the prediction.Third,in order to simplify the corn disease detection and recognition operation,a maize disease detection system based on the improved YOLOv5 s algorithm was developed,and the improved YOLOv5 algorithm was applied to the corn disease detection system.Py Qt5 tool and Py Charm software were used to realize the GUI human-computer interaction interface design,and then the generated model of the improved YOLOv5 algorithm was systematically debugging.The results show that the designed corn disease detection system is easy to operate,simple and intuitive.