Based On Fusion Of Multiple Classifiers Remote Sensing Identification Of Winter Rape In Southwest Mountainous Area

Oilseed rape is the first oilseed crop in China.It occupies an irreplaceable position in domestic oil supply and has great ornamental value.Accurate and timely information on winter rape cultivation is essential to ensure the security of our edible oil supply and to achieve sustainable development goals.The bright yellow flowers are a unique feature of rape compared to other crops.Therefore,the yellow flower index at flowering has previously been used to detect rape on aerial imagery or medium and high-resolution satellites.However,the topography of the main areas where oilseed rape is grown in China is complex and the cropping structure is chaotic;at the same time,the yellow flower signal of oilseed rape is weak in the early stages of growth.Therefore,timely identification of planting information for winter rape in the mountainous regions of the south-west remains challenging.To better address the above challenges,this thesis takes the northeastern part of Hechuan District in Chongqing as the study area,firstly extracts the winter crop planting range in the study area with geographical stratification as the leading idea,then proposes an early winter oilseed rape index,and finally uses a multiclassifier fusion method to achieve timely and accurate identification of winter oilseed rape in complex environments.The main researches and achievements of this thesis are as follows:(1)The problem of extracting winter crop ranges based on synthetic Normalised Vegetation Indices in south-western mountainous areas was addressed.Based on the GEE and crop phenological characteristics,Sentinel-2 NDVI data were synthesized based on NDVI optimum,and the winter planting range of the study area was obtained by decision tree classification.(2)A winter oilseed rape identification model for early winter oilseed rape indices is constructed.The Early Season Winter Rape Index(ESWRI)was constructed by combining the fertility mechanism of early winter rape growth and the temporal spectral characteristics in the red band,green band,near-infrared band and short-wave infrared band of Sentinel-2 remote sensing images.The ESWRI was compared with the commonly used rape indices by Fisher function and FCI clustering.The ESWRI had the largest Fisher value and the smallest DBI value,which confirmed the separability of the ESWRI from winter wheat in the early season of winter rape growth and demonstrated the timeliness of the ESWRI for winter rape identification.(3)A multi-dimensional feature and multi-classifier fusion-based approach for winter oilseed rape identification is proposed.Combining ESWRI indices and other spectral features,as well as environmental features,the optimal temporal phase for performing classification was determined to be t4 by J-M(Jeffreys-Matusita)distance values.The Dempster-Shafer evidence theory was used to fuse several classifiers,including the K-nearest neighbor classifier(KNN),support vector machine classifier(SVM),and random forest classifier(RF),to classify winter oilseed rape and other features in the study area,and the overall classification accuracy reached about 93%.The results show that the multi-classifier integration method based on D-S evidence theory has high robustness high.It can be used for winter oilseed rape identification in complex environment.