【摘要】 【目的】提出一种基于改进Mask R-CNN的农村建筑物实例分割方法，以推动人工智能技术在农业信息领域的深度融合，有效提升农村建筑物的识别效率。【方法】文章以无人机平台获取的高分辨率正射影像为数据源，经裁剪与筛选后共得1 548张建筑物图像，其像素值为512 px×512 px，并采用Labelme工具完成图像标注工作。在此基础上，通过替换Mask R-CNN模型的骨干网络（Backbone），实现了对该实例分割算法的结构改进，有效提升农村建筑物目标的检测精度。【结果】（1）在测试集上，改进后的Mask R-CNN模型在整体精度、高IoU阈值下的检测性能以及对不同尺度农村建筑物的分割能力均取得提升。（2）改进后的Mask R-CNN模型实现从遥感影像中自动提取农村建筑物轮廓，有效减轻人工绘图负担，大幅提升测绘成图效率。【结论】改进的Mask R-CNN模型不仅提升对农村建筑物的检测精度，并且在细节信息的捕捉能力上得到改善，对智能成图的发展具有重要意义。更多还原

【Abstract】 [Purpose]This study proposes an improved Mask R-CNN-based method for instance segmentation of rural buildings, aiming to enhance identification efficiency and promote the deep integration of artificial intelligence in agricultural informatization. [Method]Employing high-resolution orthophotos captured by UAV platforms as the data source, a total of 1 548 building images were obtained after cropping and screening,with annotations completed using the Labelme tool.The structure of the instance segmentation algorithm was enhanced by replacing the backbone network in the Mask R-CNN model. This modification effectively improved the detection accuracy for rural building targets. [Result]On the test set,the improved Mask RCNN model exhibited enhanced overall accuracy,improved detection performance at high IoU thresholds, and superior segmentation capability for rural buildings across different scales. It successfully enabled the automatic extraction of building outlines from remote sensing imagery,thereby effectively reducing the reliance on manual mapping and significantly boosting surveying efficiency. [Conclusion]The enhanced Mask R-CNN model improves both the detection accuracy and the fine-grained detail-capturing capability for rural buildings, contributing significantly to the advancement of intelligent mapping.更多还原