黑龙江国土资源

耕地边界精准提取是自然资源管理与粮食安全评估的关键技术。针对传统方法在复杂地形中精度不足的问题，提出基于高分辨率遥感与多源数据融合的耕地边界提取方法。通过PCA/ICA联合降维消除数据冗余，构建“深度学习初步分割—多源特征边界优化—形态学后处理”技术链路，提升复杂场景下的边界连续性与定位精度。实验选取江苏平原、云南山地、黑龙江黑土区验证，结果表明：融合多源数据后，模型平均IoU达0.92,较单一深度学习模型提升6.9%-9.5%,边界误差从8.7m降至3.2m以下，Kappa系数超0.85,显著优于随机森林、阈值分割等传统方法。研究构建的多源数据融合框架与优化技术为耕地动态监测提供了高效解决方案。

Precise extraction of cultivated land boundaries is a key technology in natural resource management and food security assessment.Considering the problem of insufficient accuracy of traditional methods in complex terrains, this study proposes a method for extracting cultivated land boundaries based on high-resolution remote sensing and multi-source data fusion.Data redundancy is eliminated through the joint dimension reduction of PCA/ICA,and the technical link of “deep learning preliminary segmentation-multi-source feature boundary optimization-morphological post-processing” is constructed to improve the boundary continuity and positioning accuracy in complex scenarios.The experiment selects the plain of Jiangsu, the mountainous area of Yunnan, and the black soil area of Heilongjiang for verification.The results show that after integrating multi-source data, the average IoU of the model reaches 0.92,which is 6.9%-9.5% higher than that of a single deep learning model.The boundary error decreases from 8.7 meters to less than 3.2 meters, and the Kappa coefficient exceeds 0.85,significantly superior to traditional methods, such as random forest and threshold segmentation.The multi-source data fusion framework and optimization technology constructed in the research provide an efficient solution for the dynamic monitoring of cultivated land.