Feature Selection and Analysis of Powdery Mildew of Winter Wheat based on Multi-Temporal Satellite Imagery

The powdery mildew of wheat cause visible and obvious image changes and lead to different spectral reflections, which can be detected by the environment and disaster reduction small satellites. The information of satellites can be transformed into many kinds of indices, some of which could be sensitive and effective in studying and detecting the powdery mildew in wheat, however, it is still unknown whether all of these features are useful in evaluating and classifying the severity of the infected plants. In this paper, we try to study and analyze the relationship and functions of the features to find a more effective combination of features in estimating and evaluating the plants and fields. We studied the feature selection and analysis in the process to discuss whether they are all necessary in the classification. Based on the multi-temporal moderate resolution images, it is found that SR_T4, NDVI_T4, MSR_T4 carry the same information and that MSR_T4 is more suitable in this study. The two-stage features are reconfirmed to be powerful in the study and enrich the results of the classification. But NDVI_T42 can be eliminated and hardly has an effect on the results. However the universality of the conclusion still needs further study due to the limitation of the data dimensionality and the analysis measure selection