Abstract：Rice smut is known as the cancer of rice. The pathogen not only affects rice yield, but also causes health risks if it attaches to food and enters the body. In this study, the normalized vegetation index (NDVI) of hyperspectral images was used to obtain feature points, and the incidence of rice smut disease was detected by the method of rank sum test. Firstly, 28 adjacent rice test areas with the same area were selected from the research base of the China National Rice Research Institute. Four farmland management methods were adopted in the area, namely natural growth and spraying with 3 different pesticides. Each management method had 7 different planting dates. The sowing dates of the plots in the adjacent experimental areas differed by 1 week before and after the plots, successively decreasing, each area planted about 500 rice plants. In the peak period of rice smut disease, the incidence of rice was first investigated on the spot, and the incidence index was obtained according to the number of incidences of rice ears per unit area. Then use the UAV-borne hyperspectral camera to shoot the test field according to the corresponding trajectory. In order to facilitate the subsequent hyperspectral image stitching, it is necessary to ensure that the aerial photography path covers the test field. According to the aerial photo coordinates, elevation information and similarity, multiple hyperspectral samples are sorted, and each hyperspectral image is stitched with high quality by the corresponding algorithm. Finally a complete hyperspectral image covering the entire test area is obtained. The normalized vegetation index that best reflects the incidence of rice smut disease is extracted from the hyperspectral image, and the feature points in the corresponding spectrum are obtained according to the index to achieve the purpose of feature dimensionality reduction. The data is cleaned with box plots to remove the feature points. Then use the cleaned feature points to perform rank sum test on the disease feature values of different rice test areas. The rank sum test is divided into two steps. The first step is to perform a rank sum test on the total sample to verify whether there is a significant difference, determine which set of samples the difference comes from; the second step is to arrange and combine the 4 sets of samples to obtain a total of 11 sets of samples to be tested in different combinations, and perform rank sum tests on the 11 sets of sample data. The significance level obtained by each group is much less than 0.01, indicating that there are extremely significant differences in sample data between different groups, and it also reflects the rationality of this method for detecting the incidence of rice smut disease. In order to show the different incidence areas, the planting areas with different incidences of rice smut are marked with different colors. Finally, the field rice incidence index was used as the control group to compare with the rank sum test results. The results showed that the rank sum test was feasible to detect the incidence of rice smut disease.
桑佳茂，陈丰农. 基于光谱特征点秩和检验的稻曲病发病程度检测[J]. 光谱学与光谱分析, 2021, 41(10): 3214-3219.
SANG Jia-mao, CHEN Feng-nong. Detection of Rice False Smut Grade Degree Based on the Rank Sum Test of Spectral Feature Points. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2021, 41(10): 3214-3219.
