| With the expansion of the area and proportion of maize in China,the occurrence of maize diseases is more serious.The traditional methods of disease judgment mainly rely on artificial eyes,and lack professional and rapid disease judgment tools.Aiming at this problem,this paper proposes an improved deep neural network model by studying the traditional machine learning algorithm and deep learning algorithm,and explores the effective application of machine vision image recognition technology in maize leaf disease recognition.Compared with the traditional artificial judgment,the user only needs to input the image of maize leaves,and after the processing of the algorithm model,the diseased category of maize leaves in the image can be output.The research contents of this paper are as follows:(1)Collect a large number of images of maize leaves and manually label the category of the images;The dataset is augmented to increase data volume and prevent network underfitting.(2)A disease recognition method based on traditional machine learning was constructed,and the classification of disease images of maize leaves was realized by combining SURF,K-means,SVM and other algorithm modules,and the experimental results were analyzed in detail.(3)The single residual network Res Net-50 model algorithm was constructed,and the number of its classification was changed to 8 categories,so as to realize the identification of maize leaf disease category.(4)Combining the common classification networks Res Net,Dense Net and Inceptionv4,a multi-channel parallel neural network Group Net was designed,and the Soft Max function was used to classify multiple categories,so as to improve the classification accuracy of agricultural images such as maize leaves.In this paper,the multi-channel parallel classification network Group Net was trained and verified on the dataset of maize leaf disease,and the precision of top-1 and top-2 is 87.53% and 98.41% respectively.According to the experimental results,the classification algorithm of deep learning can judge maize diseases more efficiently and accurately than classification algorithm of traditional machine learning,and the Group Net proposed in this paper can further improve the accuracy of the disease identification of the corn leaf image by the algorithm,and is more suitable for the application in the actual production process. |
