Experimental Research And Algorithm For Plant Leaf Disease Based On Hyperspectral Imaging Technology

Plant pests and diseases is not only a major obstacle to agricultural production which seriously restricts the plant production, but also one of the dominant factors that affects the high-quality and high-efficiency of the sustainable agriculture development. Early detection and control of pests and diseases characteristics during the development plays an important role in improving yield, reducing the economic losses caused by pets and disease on agricultural production.Leaf lesions are the most direct expression of the varieties and stages of the diseases. At the present stage, it is mainly depended on the visual estimation of the plant protection experts. By the limits of the experts' subjectivity, it's difficult and inefficient to give a quantitative expression of the disease severity. The lag of the access to the information affects the accuracy of the prediction of plants diseases which leads to an incalculable damage.In order to obtain a more efficient and objective evaluation of the disease, some researchers are applying digital image processing technology to extract and process plant leaf lesion image information.Currently the researches on the extraction of the crop lesion images are based on color photos, while the shortage of the spectral feature limits the extraction of the available spectral feature of the disease area. Hyperspectral remote sensing imaging, also known as remote sensing, using the electromagnetic spectrum ultraviolet, visible, near-infrared and mid-infrared region to obtain a lot of very narrow spectrum and continuous image data, and taking fully advantage of the spectral information to analyze the plant leaves lesions.Based on this, after a comprehensive consideration of the existing research results, this paper focus on the segmentation of the disease leaf spots, and we select some leaves from a plant as the research subjects for a systematically study. Meanwhile we make a texture analysis on the extracted area and construct a plant lesion visualization blade profile in order to lay a foundation for large-scale remote monitoring and identification of plant leaves lesion later.This paper's main work is shown as the following four aspects:(1) We apply spectral angle, the minimum distance, Mahalanobis distance and the maximum likelihood algorithm to extract disease area and evaluate the accuracy of the extracted results. The experimental comparison shows that the Mahalanobis classification algorithm preforms best. However, due to the same spectrum and the same thing with different spectral phenomena, the segmentation algorithm based on spectral characteristics may produce an incorrect classification. In order to further improve the accuracy of the segmentation, we proposed a combining objectoriented classification algorithm to achieve an efficient segmentation.(2) The extraction of the characteristic parameters of the plant leaf disease texture vector. We apply GLCM algorithms to extract the entropy, energy, inertia, contrast, correlation and other texture features of the disease image to identify the basis for post-lesion.(3) Combined with partial least squares regression and vegetation index algorithm, we apply image processing technology to get a visualized disease leaf profile, at the same time it can show the distribution of blade fungal infections and then evaluate the infection epidemic of the leaves.