【摘要】 【目的】针对传统人工绘制方法提取地貌坎线及判定其方向过程中存在的效率低、主观性强且精度有限等问题，提出一种高效、准确的地貌坎线自动提取与方向判定方法，实现对地貌特征的自动化识别与分析。【方法】文章基于无人机技术获取的高分辨率正射影像，通过优化的Mask R-CNN模型进行耕地数据识别，并对分割识别后的数据进行矢量化以及方向判断，最终绘制出坎线数据，实现自动提取坎线数据的应用。【结果】该文提出了一种高效的坎线自动化提取技术，构建了从影像预处理、深度学习识别到矢量化与方向判定的端到端流程；通过优化的Mask R-CNN模型，在整体精度、高IoU阈值下的检测性能及对复杂地块的分割能力上均表现优异，并结合方向判定算法有效克服了图像噪声与地形干扰，实现了从遥感影像中自动、精准提取坎线信息，显著提升了测绘成图效率。【结论】未来将进一步优化模型的功能和体系，并应用于自动成图领域之中，实现对耕地数据的快速获取和分析。更多还原

【Abstract】 [Purpose] Traditional manual mapping methods for extracting geomorphic terrace edges and determining their orientation suffer from inefficiency, strong subjectivity, and limited precision.To address these issues,this study proposes an efficient and accurate method for the automatic extraction of geomorphic terrace edges and orientation determination by integrating deep learning and digital image processing techniques, enabling automated identification and analysis of geomorphic features. [Method] Utilizing high-resolution orthoimagery obtained via UAV technology, the study employed an optimized Mask R-CNN model to identify cultivated land data. The segmented and recognized data were subsequently vectorized,and orientation judgments were performed to ultimately plot the terrace edge data,achieving the application of automatic terrace line extraction. [Result] This paper proposed an efficient automated extraction technique for terrace lines,constructing an end-to-end pipeline from image preprocessing and deep learning recognition to vectorization and direction determination. Through an optimized Mask R-CNN model, it demonstrated excellent performance in overall accuracy, detection capability under high IoU thresholds, and segmentation ability for complex land parcels. Combined with a direction determination algorithm that effectively overcame interference from image noise and local terrain fluctuations, this method achieved automatic and accurate extraction of terrace line information from remote sensing imagery, significantly improving mapping efficiency for surveying personnel. [Conclusion]Future work will focus on further optimizing the model’s functionality and architecture and applying it in the field of automated mapping to facilitate the rapid acquisition and analysis of cultivated land data.更多还原