Research On Extraction Method Of Crop Planting Structure Based On Machine Learning Algorithms

Crop planting structure presents the spatial distribution patterns and proportions of cultivated area of different crops in a production unit,region or country.It is a specific manifestation of land use by agricultural production activities.The timely and accurate extraction of crop planting structure using remote sensing satellite images is not only the basis for subsequent agricultural applications such as crop growth monitoring,yield estimation,irrigation guidance,and disaster assessment,but also a macro grasp of food reserves,population capacity,and agricultural trade.Therefore,the extraction of crop planting structure is of urgent demands and has great application potential.However,current methods of crop planting structure extraction do not pay enough attention to new data sources,and the redundant feature variables in machine learning-based methods reduce the computational efficiency of the data and cause accumulate errors.In addition,timely and quantitative operational framework for large scale areas need to be developed.In view of this,we intend to combine the machine learning theory to develop fast and robust methods for extracting crop planting structures based on the new Sentinel-2 data with the support of Google Earth Engine cloud platform.The research achievements of this dissertation are listed as follows:1.The research progress of crop planting structure extraction is summarized from the aspect of remote sensing data and technical methods,respectably.The analyst results show that the open access Sentinel-2 data with high spatial and temporal resolution has great potential in agricultural applications.From the technical methods aspect,traditional machine learning based methods can automatically construct decision rules,but they have excessive dependence on labeled samples.Even so,it is still the mainstream method for crop planting structures extraction in large areas up to now.2.The relative importance of different spectral and temporal features of the new Sentinel-2 remote sensing data which are used in the extraction of crop planting structure is analyzed.At the same time,the effectiveness of the newly added red edge band of Sentinel-2 is verified.The results show that the images of the vigorous growth period are the best choice to distinguish different crops,since the stems and leaves of the crops in this period are fully developed thus easy to distinguish.Short-wave infrared bands are extremely important in crop planting structure extraction,since these bands can well characterize the water content and residual coverage of vegetation.In addition,the red edge bands of Sentinel-2 are also very important in fine crop classification and presents better discrimination ability than the visible bands,since the red edge band being able to effectively capture the physiological information of vegetation pairs.3.With the support of the Google Earth Engine cloud platform,the spatial distribution patterns of the main crop types of Heilongjiang Province in 2017 and 2018 are extracted.Then,the mechanism of the spatial pattern change between these two years is analyzed.The results show that during 2017 and 2018,soybean planting area decreased while corn planting area increased,while rice the planting area and location did not change much.The planting change from soybean to corn mainly occurred in the northern and central part of Daqing and the central part of Jiamusi.This change made the difference in benefits between corn and soybeans smaller and at the same time brought higher income to farmers.