Research On Complex Planting Structure Extraction Based On Time-series Multi-featured Remote Sensing Data

With the comprehensive promotion of structural reform on the supply side of agriculture,the crop planting structure has been changing in recent years throughout China.Efficient and accurate acquisition of crop cultivation structure is of great significance for agricultural production,water resources management and food security.In this paper,we take Hohhot City in Inner Mongolia as the research area and carry out research on complex cropping structure extraction based on time-series multi-featured remote sensing data using Sentinel-2remote sensing images.The main research results are as follows:(1)In this paper,we use Sentinel-2 remote sensing images to draw NDVI curves of major crops’ weathering periods,and use S-G filtering for multi-temporal smoothing.Combining the main crops and their weathering data collected in the study area,we analyse the characteristics of maize,wheat,soybean and sunflower crops and typical weathering period images,propose the best crop identification window period,and construct multi-feature remote sensing images based on spectral feature images and NDVI time series.(2)The accuracy of the classification method based on temporal multi-features is significantly higher than that of single spectral features or NDVI features,which can better complement each other to improve the accuracy of crop remote sensing recognition.The optimal segmentation scale and the selection of main parameters are the prerequisites for object-oriented classification image segmentation.In this paper,the maximum area method is used to optimize the segmentation parameters for extracting feature images of cropland and planting structure,and the optimal segmentation scale as well as the optimal parameters of shape factor and tightness for cropland and planting structure are determined.Based on the image segmentation,a decision tree classification rule set was constructed using the classification samples for cropland and planting structure extraction.(3)The arable land area in Hohhot in 2021 and 2022 was 5531.97km2 and 5537.41km2 respectively,and the extraction accuracy of the arable land in the study area was over 90%;the overall accuracy of crop extraction in 2021 and 2022 was 92.59% and 91.83%,and the kappa coefficients were 0.9 and 0.89 respectively.There is a significant adjustment in the cropping structure in 2022 compared to 2021,with the most significant increase in soybean area,which reaches 197.03km2;vegetables and other areas have slightly increased by 4.4%;wheat,maize and sunflower cropping areas have all significantly decreased compared to 2021,with reductions of 29.03%,20.88% and 14.85% respectively.The results show that the method can provide a reference for high-precision extraction studies of complex cropping structures.