Research On Classification And Monitor Of Wheat Take-all Disease Land-based Imaging Spectrometer Based On SVM

Wheat take-all is a fungal disease that starts as a root rot, causing stunting and nutrient-deficiency symptoms in the tops. As the cross-region harvest and seed transportation in past few years, take-all disease has become one of the devastating diseases threatening wheat production. The use of remote sensing, especially the imaging spectrometer, to monitor crops has become widespread using, because it can provide real-time, accurate, and automatic assessments. As the development of the imaging spectral technology, the spectrometer can measure the spectrum characters in the same time of getting images, and their spectral resolution has advance to reflect the real information of the objectives.Support vector machine(SVM) is a kind of machine learning theory based on statistics, it has advantage on small sample, and it have been used to solve the practical agricultural issues with the non-linear or high-dimension. Take-all imaging spectrogram supervised classification method have benefit on raising the monitor accuracy, reduce the costing, and improving the usage rate of pesticide. However, the method to achieve the take-all spectrogram classified fast and precisely is still remains challenges. This study is focus on the SVM classifying algorithm and its kernel function, researching the advantage and disadvantage of SVM on the take-all spectrogram classification and monitor. The main works as follows:1. The traditional SVM is used to solve the dichotomy. How to expend the SVM classifier to multi-classified is a research focus. This study start on the analysis of several existing SVM algorithms, and using gray-cluster and principal component analysis(PCA) method to establish a spectral data pro-process system based on the take-all diseases characters. This system considered the noise sensitive problems, and it has improved the classified accuracy remarkably. 2. This study combine SVM with Grey-System, and proposing a new SVM based on Grey-Sigmoid kernel function. This algorithm used the Grey similarity degree between samples instead of the traditional dot product. The experiment shows that the SVM based on Grey-Sigmoid kernel function is able to cut a lot of training time consuming without influence the accuracy.