Research On Maize Disease Recognition Algorithm

As the main part of agriculture production in Heilongjiang Province, maize has a greatimportance on grain security in China and the national corn market. Because of the vast ofHeilongjiang Province area and the corn production areas, many different maize diseases mayoccur every year. In recent years, many researchers apply manifold learning algorithm to plantleaf image recognition. Compared with the traditional methods, pattern recognition methodbased on manifold learning algorithms, image recognition rate and efficiency is high, andinterfered by external factors is hardly. This paper sampled in the experiment field ofHeilongjiang Bayi Agricultural University, and preprocessed the data of sample images firstly.Then according to the basic principle of manifold learning algorithm, systematically studied themethod of feature extraction and lesion identification of maize leaf disease images.(1)Using the manifold learning algorithm of Principal Component Analysis(PCA), IsometricMapping(ISOMAP), Locally Linear Embedding(LLE) and Laplacian Eigenmaps(LE) algorithmto reduce the dimension of corn disease images and extract image intrinsic information. Afterthat, using the K-means, K-medoids, FCM, GK and GG algorithm to cluster analysis.(2)Diagnosis of maize leaf is sick or free. Experiments show that combined with LLEalgorithm and GK algorithm to recognize gray images of maize disease, and reduced to5--10dimensions(extract5--10features), the recognition rate is best. And extracted the feature after thethree-dimensional matrix is converted into one-dimensional matrix to recognize color images ofcorn disease, any combination of the algorithm can accurately recognize the disease leaf, therecognition rate is more than99.5%.(3)Classification of the maize disease categories. Experiments show that, for grayscale images,used the identification method in this paper to recognize the type of maize disease, therecognition rate is very low. This kind of identification method is not suitable. For color images,combine with LLE algorithm and FCM algorithm to recognize the type of maize disease, therecognition rate is fairly good.Algorithms used in this paper are realized by Matlab7.1. Experimental results show that usingmanifold learning algorithm for recognition of maize disease methods are feasible. They canreal-time and accurately identify corn disease, timely treatment, and gain better corn yields.