Research Of Crop-disease Classification And Recognition Technology Based On Evolutionary Algorithm And Improved Bow Model

The outbreaks of crops diseases and insect pests can result in serious decline of yield and quality. Besides, blindly application of chemical pesticides can lead pesticides remnant, environmental damage and so on. With the development of computing science, research and application of machine vision extends to agriculture engineering. In order to rational using pesticide application and increasing quality and production, crops pets detection and recognition can provide precise diagnosis and preventive suggestions by using the machine vision. Besides, it's significant for the development of the precision and intelligent agriculture. This paper takes maize as the study object and designs a classification and recognition method applying to the fields of corps diseases and insect pests. It fine extract the disease spots of images and integrates improved bag-of-words model to achieve classification and recognition of maize diseases and insect pests. The research work and innovative points in this paper are as follows:(1) Collecting and analyzing maize diseases and insect pests image datasets. This paper collects maize leaf diseases and insect pests images to conduct tests of common diseases in maize including northern leaf blight, southern leaf blight, anthracnose disease and so on. Besides, the image datasets are normalized. By analyzing the maize disease and insect pests images, the differences in the colors and textures between disease spots and the normal one are used as a basis for image segmentation and extraction of disease spots.(2) Considering the features of maize disease spots, this paper presents a method of histogram quadric segmentation based on evolutionary algorithm. Combining with analysis of maize pests image, it fully consider the disease spots colors and texture features and so on. And a tuple consisting of grey and chromaticity of image constructs a two-dimension histogram. It can deal with the problem of one-dimension which cannot distinguish the bimodal distribution of the object and background and improve the traditional two-dimension histogram applying to disease spots extraction. It also uses the OTSU's histogram threshold method based on evolutionary algorithm to design chromosome code which is suited data features of pests images. And it selects initial population with the image analysis to improve optimization efficiency. Meanwhile, it continues the search within the fluctuation range of optimal threshold by setting fluctuation threshold which achieves the combination of global and local searching.(3) This paper introduces guided filtering to achieve precise extraction of maize disease spots. In the extraction process, image segmentation results are filtered by using gray guided path which uses the idea of guided filtering. Setting weights matrix to recovers edges, fuzzy textures and rough spots of segmented images. It optimizes the disease spots extraction results and better keep the edges and texture features of disease spots.(4) Based on research of each modular of tradition Bag-of-Words algorithms, this paper presents an improved Bag-of-Words algorithms combining with disease spots extraction algorithm which is used in identification of crops diseases and insect pests. It uses evolutionary algorithm for an object region extraction of crops pests images. Then, extracted features of images are mapped to the high- space by using spatial pyramid. And then the data set is classified by LIB-SVM.Through the experiments of disease spots extraction, this paper presents disease spots extraction algorithm which can be effective to determine the disease spots of maize images and extract disease spots which have a detail edges and textures. The disease spots extraction algorithm is compared with OTSU algorithm based on threshold, EM algorithm based on clustering, G-MRF algorithm and SRG algorithm. The results of experiment demonstrate the effectiveness and advantage of this algorithm. On this basis, through the experiments of classification images for disease and insect pests, the improved bag-of-words model algorithm designed in this paper can be applied in identification of crops diseases and insect pests. It improve the recognition performance of traditional bag-of-words model and pays more attention to features of disease spots. Experiments shows that it is applicable for identification of disease and insect pests and achieved satisfactory results.