Image Recognition Of Millet Leaf Disease Based On Machine Vision

Millet is one of the main multigrain crops in China.Its sown area occupies 80% of the world's sown area,and its output accounts for about 85% of the world's total millet output.On the premise of the same planting area,the yield investment ratio and economic benefits of millet are better than those of corn,wheat and other major crops.However,the serious loss of grain yield is caused by diseases and insect pests every year,and the control measures for different types of millet leaf diseases are also different.Based on the similarity and diversity of disease categories,In order to solve disease problem,it is an important guarantee to finding an effective automatic detection method.In this paper,we have selected millet blast disease,white hair disease,brown spot disease,rust disease and red leaf disease as the test objects,and the recognition method based on machine vision is used to realize the classification and discrimination of five cereal leaf diseases,which provides an effective scientific method for the automatic detection of cereal leaf diseases.The main contents of this paper are as follows:(1)Image collection and trend analysis of millet leaf diseases.The five disease types were photographed by mobile phone.The five kinds of diseases of millet leaves,such as blast disease,white hair disease and red leaf disease,the experimental background of the disease is complex.According to the trend analysis of the same disease plants,it was found that the transformation from early stage to late stage could be completed in only 13 days.(2)Research on image segmentation of millet leaf disease in complex background.This method combines the morphological operation of gray-scale image,and introduces the narrow-band fast method to complete the edge detection of leaf contour,the segmentation of diseases on millet leaves under complex background was basically realized.The three methods of maximum inter-class variance method based on super green feature,region growth method based on seed point selection and K-Mean clustering algorithm based on Lab space were used to compare the segmentation effect.The results show that the maximum inter-class variance method based on super-green features has the best effect,and can be used for the subsequent feature extraction.(3)Feature extraction and feature combination optimization of millet leaf diseases.Based on the gray level and gradient information of gray level co-occurrence matrix and gray level gradient co-occurrence matrix,the texture features of RGB and lab images of five diseases are extracted.Based on the binary image of disease spots,the morphological features are extracted and the original 34 dimensional feature space is constructed.The ant colony algorithm is selected to reduce the dimension of the feature space from the previous 34 dimension to 16 dimension,reducing unnecessary redundant information.(4)Parameter optimization and classification of SVM.In the experiment,the 16 optimized by ant colony is used as the input sample,and support vector machine is selected as the classification and recognition tool.The total number of samples in the experiment is 900,including six kinds of samples including disease of millet leaves and healthy millet leaves.The number of samples in each disease is 150,of which 600 are used as training sets and 300 are used as testing sets.Compared with the gray wolf optimization algorithm,the recognition rate based on the gray wolf optimization is 96.67%.Compared with the grid search method,the recognition rate is increased by 5.34%,and the recognition time is shortened by 233.4s.The experimental results show that the classification and recognition of millet leaf diseases based on machine vision is effective and can realize the automatic detection of five kinds of millet leaf diseases.