Statistical Analysis of NDVI of Agricultural Land Cover Aerial Images

Abstract

AbstractMonitoring the state of agricultural land using aerial photography is an urgent problem. Monitoring includes such tasks as classification, segmentation, regression analysis, etc. All these tasks are related to the field of machine learning. However, machine learning methods require large representative datasets to perform effectively. The key issue in collecting of the representative datasets is to determine appropriate significant features that represent relevant objects. One of such features is Normalized Difference Vegetation Index (NDVI). To determine significant features, it is necessary to identify the statistically significant differences in these features between integral land cover surface classes. Nevertheless, in modern scientific literature, insufficient attention has been paid to this problem. This paper presents such a study on a large and representative Agricultural Vision Dataset, which contains images with the following classes of agricultural land cover surfaces: Double Plant, Drydown, Endrow, Nutrient Deficiency, Planter Skip, Water, Waterway, Weed Cluster. To test for differences in NDVI values across classes, Kruskal–Wallis and Mann–Whitney U tests were used, which showed statistically significant differences (p < 0.001). The results obtained in the course of the study allow us to conclude that the considered indices are significant as features of classes in the Agricultural Vision Dataset.KeywordsAerial imagesAgricultural land analysisNDVIKruskal–Wallis testMann–Whitney U test