Early Monitoring Of Diseade And Insect Pest Of Brassica Napus Based On Hyperspectral And Image Processing Technology

Digital agriculture is an indispensible technology in the field of agricultural development and fusion of morden information technology, artificial intellegence and agriculture. During the analysis of plant disease and insect pest, to obtain the information of plants suffering from the disease and insect pest in time and accurately is required by the digital agriculture. Accurate dose of pesticide spraying is required by digital agriculture plant diseases and insect pest management, locating the healthy part and injured part and the degree of injury severity by the intelligentized plant diseases and insect pests real-time monitoring is the premise of the accurate dose pesticide spraying and the key of the digital agriculture fine management. This research mainly focused on the Brassica napus L., which is a widely planted, high economic valued and alternative energy resource plant. The analysis is practiced on the spectral dimension data and spatial dimension data. A key information acquisition and identification model of rape plant diseases and insect pests, a rape classification model according to how long the rape was infected by sclerotinia, an location and reconstruction model of holes on rape leave corsed by cabbage caterpillar and an optimized algorithm of vein extraction of leave of Brassica napus L. which were injured by insect pests were established by technologies of hyperspectral imaging, spectral data acquisition, data modeling, intelligent computing and digital image analysis. The real time monitoring of Brassica napus L. which was suffering from disease and insect pest at early stage was performed. The main innovation results were achieved as follows:(1) A novel hyperspectral waveband selection algorithm based on between-classes instability index and mean impact value parameters selection is proposed and applied to establish a fast classifing model used to classify sclerotinia sclerotiorum scab on Brassica napus L. leave according to how long it is infected. The hyperspectral data of Brassica napus L. leave were classified into6classes according to the period the leaf infected with sclerotinia sclerotiorum. During the analysis of the spectral data in hyperspectral images, the algorithm of hyperspectral wavebands selection based on the between-class instability index and the algorithm of mean impact value parameters selection were applied to reduce dimension and extract feature. Fourty five optimized wavebands were selected and the correlation index and root-mean-square error are0.9and3.35675. After mean impact value variable selection,30optimized wavebands were selected.(2) The core idea of the between-class instability index is to choose the wavebands which are the best illustrations of difference among classes. The algorithms of swarm intelligence were applied to optimize topological structure, Weights and threshold values of neural network in order to establish an optimized classification model and classify the pixels of leaves accurately according to the period the leaf infected with sclerotinia sclerotiorum. To analyse the data of the leaves which were24hours,48hours,72hours,96hours,120hours and144hours after infected with sclerotinia sclerotiorum,30optimization wavebands were selected and used to establish neural network models optimazed by ant colony algorithm and particle swarm algorithm with prediction correlation coefficient of0.5770and0.8527, mean-root-square error of2.33268and1.02670, respectively.(3) A novel algorithm of wormholes on Brassica napus L. leaf caused by cabbage caterpillar detection and reconstruction was proposed and applied to perform wormhole on the edge of leaf reconstructing automatically. The algorithm makes it possible to detect the degree of injury severity caused by pests based on damaged leaf area computation. During the analysis of the space dimension data in hyperspectral image, the accuracy outline of rape leaf damaged by cabbage caterpillar was extracted. The wormholes on leaves were classified into two categories, closed wormholes and unclosed wormholes. For closed wormholes, holes filling function was applied; for unclosed wormholes, a novel method was proposed which including two steps, location and reconstruction. Locating factor and test function based on first-order derivative of inverse function of parameter equation of edge curve were established and the size of function pulse was used to locate the wormhole.(4) A novel G-WNNRA reconstruction algorithm were proposed. Wavelet transform and genetic algorithm were introduced into G-WNNRA prediction model. The input vectors of training dataset were composed of the polar angles and polar diameters of discrete points of unbitten edge of leaf and the input vectors of predicting dataset were composed of the polar angles and polar diameters of discrete points of bitten edge of leaf. The output vectors were the constraint values of discrete points. The experimental results show that the correlation indeces for training set and prediction set are0.998and0.953, mean-root-square error are0.00681and0.02714. The model of G-WNNRA can be used to identify and reconstruct the wormholes on leaf effectively and the performance was the best.(5) A novel optimized rape vein of leaf damaged not serious by pest detection algorithm was proposed. In addition to inherent color difference between mesophyll and vein pixels, color difference between injured pixels and healthy pixels because of the loss of nutrient and moisture will generate false edges during the vein edge detection. And the edge of wormholes will cause false edges too. These non vein edges should be regarded as noise which will disturb vein extraction. In this study, methods based on principal component analysis and derivative spectrophotometry were applied to extract vein of leaves which were damaged by cabbage caterpillar. The dimensions of spectral data were reduced and the space dimension data was processed by principal component analysis and derivative spectrophotometry combined with digital image processing technology such as methods of spatial filtering and image morphology. A method of leaf vein extraction was established ultimately.The results of the study realized the key information retrieval of rape suffering from main disease and insect pests and provided the foundation for further prevention and management, which are significative for science research and practice.