Identification Of Sclerotinia Sclerotiorum Based On Spectral And Multi-spectral Image Technique

Oilseed rape is one of China's four major oil-bearing crops, with adaptability, having high economic value. The growth status of rape determines the yield and quality. As one of the major diseases of oilseed rape, the incidence rate of sclerotinia sclerotiorum can achieve to 10-30%, even to 80% annually. The yield of infected oilseed rape can be cut over 70% and the yiled and quality of rapeseed are affected seriously. At present, the prediction of sclerotinia sclerotiorum mainly depends on the human eye. But it is difficult to predicte disease in time, missing the best prevention and treatment time. At the same time, the human eye is subjective and demand more time and energy. It is unable to meet the modern agricultural production and management requirements. Therefore, a fast and accurate method for detection of Sclerotinia sclerotiorum and technology is needed. The early prediction and identification of sclerotinia sclerotiorum is the key technology in variable spraying system.In this paper, spectral and multi-spectral imaging technology for sclerotinia sclerotiorum identification is discussed. The main contents are as follows:1. The pathogenesis of Sclerotinia sclerotiorum was researched in this paper. Combined with digital agriculture techniques, the feasibility of early diagnosis of Sclerotinia sclerotiorum was discussed. Spectroscopy technology was applied in early diagnosis sclerotinia sclerotiorum. The partial least squares model, BP neural network model and least squares support model were established. The recognition rate of the BP neural network and least squares support vector machine model in which second-order differential treatment based on the extraction of characteristic values by partial least-squares analysis reached 100%. In order to improve the model computing speed, 8 effective Wavelengths were extracted based on the PLS model. The recognition rate of PLS recognition model base on the effective wavelengths can reach 90% in the threshold of 0.5. The effective wavelengths represent full-band information to some extent, and they are the most important wavelengths. It made a good foundation for the development of instrument.2. The red, green and near-infrared channel multi-spectral images was used to find the method of identifying sclerotinia sclerotiorum. The multi-spectral image was preprocessed by the denoising method, and then was extracted 2 color features and 5 texture features, then the partial least-squares regression analysis, BP neural network and least-squares - support vector machine model were established for the identification of sclerotinia sclerotiorum. Comparing the above several models, the recognition rate of the BP neural network model based on the MSC texture features reached 100% in the threshold of 0.5. Therefore, the use of multi-spectral imaging technology can quickly and accurately identify sclerotinia sclerotiorum.