Identification of Sclerotinia Sclerotiorum Based on Spectral and Multi-spectral Image Technique
|Course||Agricultural Mechanization Engineering|
|Keywords||Precision Agriculture Rape leaf Sclerotinia Spectral and multi-spectral imaging technology Image Processing|
Rapeseed is one of China's four major oil crops, canola rapeseed growth conditions determine the yield and quality. Sclerotinia rapeseed production as one of the important diseases, diseased plants generally cut more than 70%, seriously affecting the yield and quality of rapeseed. Currently, the diagnosis of Sclerotinia rely on the human eye to discriminate more, the drawback is predicted not timely, subjectivity, low efficiency, therefore, need for a quick, accurate detection method for Sclerotinia and techniques. In this paper, spectral and multi-spectral imaging technology Sclerotinia for rapid identification and early diagnosis. The main research contents and innovations are as follows: 1. Combined Sclerotinia pathogenesis, the use of visible / near-infrared spectroscopy techniques for early diagnosis of Sclerotinia. ASD portable spectrometer using rape leaves samples collected spectra, the establishment of Sclerotinia identify partial least squares model (PLS), BP neural network model (BPNN), least squares support vector machine identification model (LS-SVM), and comparing recognition results of these models. Where second-order differential treatment based on partial least squares analysis and after extracting characteristic values ??and the establishment of BP neural network and least squares support vector machine recognition model works best recognition rate has reached 100%. In order to improve the computing speed of the model, according to the established model extraction PLS identification information about Sclerotinia eight characteristic wavelengths. Based on these characteristic wavelength recognition model created to predict the correct rate in the determination threshold value is 0.5 and 0.4, respectively, in case 87.5% and 65%. Illustrate these characteristic wavelengths in a certain extent, on behalf of all wavelengths of the information is a lot of spectral data is the most important part of the wavelength, the wavelength for feature extraction model simplification and subsequent instrument development has laid a good foundation. (2) application of the red, green and near-infrared three-channel multispectral images were Sclerotinia Identification Method. The collected red, green and near-infrared spectral image consisting of the removal of the background noise processing and extract the RGB, HSI color space and color features 12 red, green and near-infrared image 15 of three channels texture characteristics, the main use of partial least squares regression analysis, BP neural network and least squares support vector machine to establish a model of Sclerotinia identification. Compare the established model of recognition effect. Found that pretreatment of MSC were established under the color and texture features based on BP neural network model of optimal effect, the judgment threshold value of 0.1 in the case of the recognition rate of 100%. In order to evaluate the applicability of BPNN model, this paper also chose different batches of 50 samples and 50 healthy infected sample input to this model, the range of allowable error of 0.1 discriminant to predict the correct rate reached 99%, 98%. This shows that this model has two models some versatility. Therefore, the use of multi-spectral imaging technology can quickly and accurately identify Sclerotinia, to achieve real-time, reliable monitoring and control of plant diseases provides a new method.