Study On Maize Leaf Disease Recognition Based On YOLO Model

Maize is one of the crops that can adapt to a variety of growth environments and high yields,and each year brings significant economic benefits to growers,but because maize has a long cycle from growth to maturity for sale,during this period,maize diseases are many and because of the similarity of disease characteristics,resulting in maize diseases are difficult to be detected in a timely manner.At present,maize disease identification in China is mainly by farmers through visual observation and empirical judgment,and researchers through a variety of data comparison based on empirical judgment of data results,not only has limitations,but also prone to subjective errors,as well as longer identification time and other problems.By using convolutional neural networks and deep learning in the field of maize disease identification,the problems of difficult and time-consuming identification of maize spot species and the influence of external factors have been alleviated to a certain extent.Since it is impossible for farmers to carry large servers with them while working,the development of a maize disease identification system that can be deployed on mobile devices.In this paper,we propose an improved YOLOv5 n model incorporating CA(Coordinate Attention)mechanism and STR detection head:CTR＿YOLOv5n,to identify common maize leaf spot,gray spot and rust diseases.The main research of this paper has the following parts:(1)Data pre-processing of the maize leaf dataset,mosaic data enhancement,rotation,panning and cropping of the dataset to improve the quality and quantity of the dataset to some extent.Five mainstream neural network models Faster R-CNN,YOLO3,YOLO4,and YOLO5 with different network depths were built and trained on the corn leaf disease dataset to compare the performance performance.Finally,YOLOv5 n is selected as the base network for this study,and certain improvement enhancement measures are made to the base model.(2)The model accuracy is improved by adding CA attention module to the lightweight model YOLOv5 n,which has better recognition effect and higher accuracy compared with SE,CBAM and ECA,the mainstream attention mechanisms.To enhance the recognition capability of the model for small targets,a smaller x-samll target detection head is added to the original small,medium and large size detection heads of the YOLOv5 n model.STR(Swin Transformer)is used as the detection head to enhance the model’s ability to obtain global information.The average recognition accuracy of the proposed algorithm model can reach 95.2%,which is 2.8% better than the original model,and the memory size is reduced to 5.1MB compared to 92.9MB in YOLOv5 l,which is 94.5% smaller and achieves the lightweight requirement.(3)Develop the maize leaf disease identification system by Py Qt5,python and Py Charm tools to make the human-computer interaction interface simple and clear,realize direct online monitoring of leaf images and videos,and provide technical support for the subsequent development of the mobile APP.In this paper,the CTR＿YOLOv5n model is improved on the basis of YOLOv5 n to achieve efficient and easy recognition of maize leaf spots with less computational resources,which provides technical support for the development of mobile recognition of maize leaf spots.