关键词: 杭州湾, 土地覆盖分类, 中低空间分辨率, 训练决策树, 感兴趣区修正

Abstract: Due to the reliance on medium and low-resolution historical remote sensing images,the classification results of long-term land cover sequences often have low accuracy or are not fully applicable to certain areas.Therefore,this study proposes an improved land cover classification method based on training decision trees to reveal the drastic changes in land cover in the Hangzhou bay under the interweaving of intense human activities and natural succession.The study firstly uses the synthetic image classification based on the preprocessed bands and multi-index inversion results,modifies the region of interest of the image to be classified on the basis of the classification results of the pre-trained decision tree model,and then obtains the classification results through supervised classification methods and verifies them.The 30 meter resolution land cover classification results of the Hangzhou bay from 2000 to 2024 show that the overall average accuracy of this method reaches 91.72%,and the Kappa coefficient is 0.91.Further analysis indicates that the land structure evolution in the study area can be divided into three stages: the stage dominated by grain production (2000—2007),the dynamic development stage (2007—2015),and the ecological restoration stage (2015—2024).

Key words: Hangzhou bay, land cover classification, medium and low-resolution spatial resolution, training decision trees, region of interest refinement