Extraction of compound rotation planting structure based on Sentinel-2A image

【Objective】Remote sensing technology was used to explore the extraction method of compound rotation planting structure in Erhai Lake basin in Dali, Yunnan, which aimed to provide theoretical reference for the adjustment and management of crop planting structure in the basin.【Method】Multi-scale segmentation of Sentinel-2A image of Erhai Lake basin was conducted, and position features were added in addition to common features such as spectrum and texture to build a classification feature set. Then C5.0 decision tree algorithm was used to study classification rules, and confusion matrix was used to evaluate classification accuracy.【Result】In crop classification, the spectral feature Mean_Green was the most important parameter, the position feature Y_Center was mainly used to distinguish between purple leaf lettuce and corn, and X_Center was one of the important features for broad bean extraction. According to C5.0 decision tree classification rules, fine plot maps of crops in the late spring and early spring were drawn. Late spring crops were divided into rice, corn, purple leaf lettuce and other crops, while early spring crops were divided into broad bean, oilseed rape, purple leaf lettuce and other crops. Then, a total of 16 kinds of compound rotation planting structure were generated, among which rice-broad bean(29.54%) and corn-broad bean rotation(23.66%) accounted for the largest proportion of the total cultivated land area in the basin, while rice-oilseed rape and corn-oilseed rape rotation occupied only 2.05% and 1.58% of the total cultivated land area, which were smaller. The area planted with purple leaf lettuce in both late spring and early spring accounted for 4.26% of the total cultivated land area, and the total area using the remaining 11 rotation methods was 7726.77 ha, accounting for 38.91% of the total cultivated land area. The overall classification accuracy of late spring crops was 80.00%, and the Kappa coefficient was 0.7142. The overall classification accuracy of early spring crops was 81.75%, and the Kappa coefficient was 0.6983.【Suggestion】When extracting the compound rotation planting structure, it is suggested to consider the C5.0 decision tree algorithm and add position features to extract crop information. In view of the maldistribution of the compound rotation planting structure, it is recommended that local agricultural departments should strengthen publicity and guidance, optimize and adjust the crop planting structure in Erhai Lake basin by following the principle of adaptation to local conditions, so as to improve the ecological, economic and social benefits of agriculture planting in the basin.