Extraction of Rape Planting Information in Cultivated Land Fragmentation Region Based on HJ-2 CCD Imagery

ZHANG Meng; XU Xiang; WANG Zhuang; WANG Jie

doi:10.15933/j.cnki.1004-3268.2025.07.018

Abstract：Owing to the diverse landscape pattern of farmland,complex crop planting structure,and scattered as well as irregular plots in the cultivated land fragmentation region,it is prone to have a severe mixed pixel phenomenon when applying medium and low spatial resolution remote sensing imagery,which makes it difficult to obtain high-precision crop planting information.The popularization and use of high spatial resolution remote sensing imagery has improved this situation and increased the accuracy of crop identification in cultivated land fragmentation region,but the research on planting extraction based on high spatial resolution remote sensing imagery for rape is still relatively scarce at present,and the existing identification methods suffer from the problems of large amount of data processing and complicated process.To address this,based on the HJ-2 CCD imagery with both spatial resolution and spectral band enhancement in Zongyang County,Anhui Province,a rapid and effective rape identification model for fragmented regions of cultivated land was proposed by utilizing the derived spectral feature set constructed by vegetation indices of key phenological stage image of rape blossoming as well as combining the Bagging ensemble strategy and minimum Mahalanobis distance classifier.The results of the study showed that,compared with five other typical rape identification modeling methods,RAW+SVM,RAW+RF,RAW+Bagging-MMDC,VI+SVM,VI+RF,the proposed model VI+Bagging-MMDC had the best results in land cover classification and rape extraction,with overall accuracy,Kappa coefficient,user accuracy of rape,and producer accuracy of rape being 94.27%,0.93,98.59%and 97.21%,respectively;Compared with the statistical data,the total area accuracy of the proposed model reached 92.80%,the relative error was 7.20%,and the planting distribution was highly consistent with the actual rape plots,thereby meeting the practical application requirements.The introduction of red-edge information increased the overall accuracy,Kappa coefficient and user accuracy of rape by 2.40 percentage points,0.03 and 4.94 percentage points,respectively.This study proves that the combined use of key vegetation index of rape blossoming stage,Bagging ensemble strategy,minimum Mahalanobis distance classifier and red-edge information can effectively improve the extraction accuracy of crops in the cultivated land fragmentation region.