为及时、准确地识别出煤矿区采动地裂缝，避免次生地质灾害发生和恢复煤矿区土地生态环境，以陕西省榆林市神木县西北部柠条塔煤矿采煤工作面裂缝发育区为研究区，基于低空无人机遥感影像，结合实地查勘，构建面向对象监督分类模型方法，开展地表采动裂缝提取方法研究。首先，借助ESP（Estimation of Scale Parameter）最优尺度评价工具得到候选分割参数，结合目视解译快速确定最优分割参数，得到裂缝、植被等影像对象；利用特征空间优化工具从24个初始特征集中确定15个优化特征参数构建优化特征集；在此基础上结合支持向量机(SVM, Support Vector Machine)、K最近邻(KNN, K Nearest Neighbor)、随机森林(RF, Radom Forest)、朴素贝叶斯(NB, Naive Bayes) 多种机器学习分类器模型，试验分析得出：地物的分类效果和分类精度具有一致性，SVM分类方法总体效果最好，在4个易错分区域中表现最好，误分小斑块数量最少，总体分类精度达到88.97%，Kappa系数到达0.849，裂缝提取精度F₁值达到87.87%，Kappa系数达到0.848。4种分类方法整体分类精度均在80%以上，优选模型方法准确提取到研究区10条主要裂缝，相比传统人工矢量化更加高效，借助低空无人机遥感影像和面向对象方法可有效提取矿区地表采动裂缝，研究可为采煤沉陷地质灾害调查监测及土地生态修复提供技术支持。

In order to promptly and exactly identify the mining ground fissures in coal mining areas, and avoid the secondary geological disasters, as well as restore the land ecological environment in the coal mining areas, this study focused on the extraction method of surface mining induced fissures, with the fissure development zone of coal mining face of Ningtiaota Coal Mine as the study area, which was located in the northwest of Shenmu County, Yulin City, Shaanxi Province. Meanwhile, the smooth execution of this research was based on low-altitude UAV remote sensing images, field surveys, and the construction of an object-oriented supervision classified model method. The images acquisition process was shown as follows: Firstly, the candidate segmentation parameters were obtained utilizing the ESP（Estimation of scale parameter）optimal segmentation scale evaluation tool, and then the optimal segmentation parameters were determined immediately combining visual interpretation, finally the image objects such as fissures and vegetation were obtained. 15 optimized feature parameters were determined from 24 initial feature sets to construct the optimized feature set with the feature space optimization tool. On this basis, a variety of machine learning classifier models were combined, such as Support Vector Machine, K Nearest Neighbor, Random Forest, Naive Bayes, etc. The experimental analysis results presented that the classification effect and accuracy of the land features were consistent. The SVM classification method had the best overall effect, performing best in the four erroneously partitioned domains, with the least number of misclassified small patches. The overall classification accuracy achieved 88.97%, and the Kappa coefficient attained 0.849. In addition, the F1 value of crack extraction accuracy reached 87.87%, with the Kappa coefficient amount to 0.848. The overall classification accuracy of the four classification methods was above 80%. The optimal model method accurately extracted 10 main fissures in the research area, which was more efficient than traditional manual vectorization. The surface mining fissures could be effectively extracted by the aid of low-altitude drone remote sensing images and object-oriented methods. This research could provide technical support for the investigation and monitoring of geological disasters caused by coal mining subsidence and land ecological restoration.