Research And Application Of Corn Leaf Disease Identification Method Based On Improved Mobilenetv3

Corn Disease is one of the problems affecting the yield of corn,which will affect the yield and quality of corn in the growth process.How to efficiently and accurately diagnose and identify corn disease,control and prevent it effectively is very important in the process of corn production.Maize disease identification is usually carried out by experts collecting samples for experimental analysis or field observation.This method is too time-consuming and laborious to be applied to promotion.The traditional maize disease diagnosis and identification methods have low accuracy and weak robustness due to the problems of mutual shielding and uneven spot area of maize disease leaves.In recent years,deep learning technology has been widely used in the field of computer vision,and has achieved good results in many recognition studies,which provides strong support for the field of agricultural disease diagnosis and recognition.In this paper,the CNN algorithm based on deep learning was used to design and implement the maize disease recognition software application,which could accurately identify northern leaf blight,dwarf Mosaic virus,rust,gray spot and healthy leaves.The main work of this paper is divided into the following aspects:(1)This paper collects and integrates maize leaf disease images from multiple open source data sets,classifies and labels the data sets,and then carries out data enhancement operations.The enhanced data set contained 33409 images,including 4diseases and 1 healthy leaf.The training set and the test set are divided by 8:2 ratio and used in the training and testing of the model.(2)An improved Mobilenetv3 method for maize leaf disease recognition was proposed in this study.Compared with heavyweight model and lightweight model,the improved model CD-Mobilenetv3 proposed under the same data set has better results in all aspects.The ablation and visualization experiments showed that the accuracy of CD-Mobilenetv3 test was 2.69% higher than that of Mobilenetv3,and the number of parameters was 20.44% lower,run time decreased by 0.82 s.It can be seen that the method can effectively improve the identification effect of maize leaf diseases.(3)The maize disease recognition software based on CD-Mobilenetv3 proposed in this paper.The C/S architecture is adopted.The client uploads the picture data,and the server processes the picture data and returns the result.The user can identify the corn disease by taking pictures or selecting from the memory.The test results show that the software has stable performance and high accuracy,and is suitable for real scenarios.