矿产资源长期以来都是社会生产发展的重要基础，作为现代工业的“粮食”和“血液”，其在能源供给、基础设施建设、制造业发展等领域发挥着不可替代的作用。据国际能源署统计，全球90%以上的基础能源和80%的工业原材料都直接来源于矿产资源。然而，随着工业化进程的加速，矿产资源的无序开发导致了一系列生态环境问题，如何实现矿产资源开发与环境保护的平衡已成为全球性课题。在此背景下，利用遥感技术对矿区土地损毁情况进行全方位监测，不仅能为规范采矿活动提供科学依据，更能为评估矿区生态修复效果建立量化指标，具有重要的现实意义。

当前矿区遥感监测面临的主要技术挑战源于矿区复杂的地表特征。露天开采形成的阶梯状矿坑、尾矿库的液态/固态混合表面、矿区固体废弃物的分散分布等要素在遥感影像上呈现出显著的光谱异质性和空间破碎性。传统基于像素的监督分类方法和基于面向对象的图像分析方法在处理此类复杂场景时，往往存在边界模糊、小目标漏检等问题。针对这一技术瓶颈，本文创新性地构建了一个多尺度分支解耦网络（Multi-Scale Branch Decoupling Network，MSBDN），通过层次化特征解耦与融合策略，显著提升了遥感影像中矿区土地损毁的识别精度。本文主要开展了以下工作：

（1）通过对矿区遥感影像的光谱、纹理特性和矿区固废的空间分布模式的研究，引入深度学习关键技术，为土地损毁区域的高精度提取提供了技术支撑。本文选取了陕西省中南部秦巴山区核心区县的典型矿区作为实验区域，建立了遥感影像矿区解译体系，覆盖土地损毁整体区域及其固废：排土场、尾矿库、碎石堆等典型地物。并通过人工解译方式，通过像素级标注构建了矿区地物数据集。

（2）提出了一种基于解耦注意力与多尺度融合的矿区土地损毁识别与提取模型MSBDN，该模型创新性地采用了多分支并行特征提取机制。在编码器部分，全局分支解译模块（Global Branch Decoding Block ，GBDB）基于改进的Swin Transformer架构，通过移位窗口注意力机制捕捉矿区的全局空间关联；局部分支解耦模块（Local Branch Decoupling Block ，LBDB）则采用空间和通道解耦的注意力机制，重点提取矿区土地损毁边界和碎石堆等小尺度目标的局部细节特征。特别设计的分支级联模块（Branch Cascade Block ，BCB）在三个不同尺度上实现特征融合。在解码器部分，多尺度特征融合模块（Multi-Scale Feature Fusion Block ，MSFFB）实现了实现多尺度特征的语义对齐，该设计有效解决了传统网络架构中因简单跳跃连接导致的特征失配问题。

（3）使用标注的矿区土地损毁区域遥感数据集对MSBDN进行训练和测试，与经典的图像目标提取网络进行对比，并进行不同分支的消融实验。实验结果表明，MSBDN在矿区土地损毁遥感解译任务中展现出显著优势。在测试集上，总体准确率为96.6%，精度为91.1%，召回率为91.1%，较Vison Transformer、Deeplabv3+、Unet皆有提升。消融实验证实，双分支结构使目标检测准确率提升0.5%，精度提高6.7%，召回率提高2.45%。通过以上两种实验测试，验证了本方法对于遥感影像的矿区土地损毁区域可以实现有效提取。

（4）基于矿区固废的遥感影像特征，运用迁移学习的策略，对预训练的MSBDN实行冻结-微调方式，并引入CRF后处理，对矿区尺寸较小的固废进行目标提取。实验结果表明，改进后的MSBDN对小目标的整体提取效果较好，但是由于特征学习不足，对碎石堆仍出现漏判问题。实验验证了方法的可行性与改进潜力，为复杂矿区场景下的小目标智能识别提供了重要技术支撑。

本研究聚焦于矿区遥感影像的智能解译，围绕遥感影像矿区土地损毁区域识别复杂度高、小目标易漏检等问题，提出了一套完整的技术路线并取得显著成果。首先，通过对遥感影像中不同矿区的光谱、纹理和空间分布特征进行系统分析，构建了涵盖典型地物的土地损毁区解译体系，并基于人工像素级标注构建了遥感数据集，为模型训练提供了坚实数据基础。在方法上，创新性地设计了融合解耦注意力机制与多尺度特征融合的MSBDN模型，引入双分支结构分别提取全局与局部特征，显著提升了土地损毁区的识别和提取效果。同时，结合迁移学习与CRF后处理策略，进一步提升了土地损毁区中固体废弃物的解译精度。研究成果不仅推动了遥感影像智能解译技术在矿区场景中的应用落地，也为资源环境监测、生态评估与智能化管理提供了重要的技术支撑，具有较强的工程推广价值与现实意义。

Mineral resources have long been a vital foundation for the development of social production. As the "grain" and "lifeblood" of modern industry, they play an irreplaceable role in energy supply, infrastructure construction, and the development of manufacturing. According to statistics from the International Energy Agency, more than 90% of global primary energy and 80% of industrial raw materials are directly derived from mineral resources. However, with the acceleration of industrialization, the unregulated exploitation of mineral resources has led to a series of ecological and environmental issues. How to balance resource development and environmental protection has become a global challenge. Against this backdrop, the application of remote sensing technology for comprehensive monitoring of land degradation in mining areas not only provides scientific support for regulating mining activities but also establishes quantitative metrics for evaluating ecological restoration effectiveness, thus holding great practical significance.

Currently, the main technical challenges in remote sensing monitoring of mining areas arise from the complex surface features typical of these zones. Elements such as terraced open-pit mines, the liquid/solid mixed surfaces of tailings ponds, and the scattered distribution of solid waste in mining areas exhibit significant spectral heterogeneity and spatial fragmentation in remote sensing imagery. Traditional pixel-based supervised classification methods and object-oriented image analysis methods often encounter issues such as blurred boundaries and missed detection of small targets in such complex scenarios. To address this technical bottleneck, this study innovatively proposes a Multi-Scale Branch Decoupling Network (MSBDN), which significantly enhances the recognition accuracy of land degradation targets in mining areas through a hierarchical feature decoupling and fusion strategy. The main work carried out includes:

（1）By analyzing the spectral and textural characteristics of mining area remote sensing imagery and the spatial distribution patterns of solid waste, this study introduces key deep learning techniques to support the high-precision extraction of land degradation zones. Typical mining areas in the core counties of the Qinba Mountain region in southern-central Shaanxi Province were selected as the study area. A remote sensing interpretation system for mining areas was established, covering the entire land degradation zone and representative solid waste features such as dump sites, tailings ponds, and gravel piles. A manually interpreted, pixel-level annotated dataset of mining area features was also constructed.

（2）A novel model for extracting land degradation in mining areas—MSBDN—was proposed, based on decoupled attention and multi-scale feature fusion. This model employs a multi-branch parallel feature extraction mechanism. In the encoder section, the Global Branch Decoding Block (GBDB), based on an improved Swin Transformer architecture, utilizes a shifted window attention mechanism to capture global spatial relationships in mining areas. The Local Branch Decoupling Block (LBDB) applies spatial and channel decoupled attention mechanisms to focus on local details, such as degradation boundaries and small gravel piles. A specially designed Branch Cascade Block (BCB) fuses features at three different scales. In the decoder, the Multi-Scale Feature Fusion Block (MSFFB) aligns multi-scale features semantically, effectively addressing the feature mismatch problems common in traditional architectures due to simplistic skip connections.

（3）The MSBDN was trained and tested using the annotated mining area remote sensing dataset. It was benchmarked against classic image segmentation models, and ablation studies were conducted on different branches of the model. Experimental results show that MSBDN demonstrates significant advantages in the task of interpreting land degradation in mining areas. On the test set, it achieved an overall accuracy of 96.6%, precision of 91.1%, and recall of 91.1%, outperforming models such as Vision Transformer, DeepLabv3+, and UNet. Ablation experiments further confirmed that the dual-branch structure improved small object detection accuracy by 0.5%, precision by 6.7%, and recall by 2.45%. These experiments validated the model’s effectiveness in extracting land degradation areas from remote sensing imagery.

（4）Based on the remote sensing features of solid waste in mining areas, a transfer learning strategy was applied. The pre-trained MSBDN was fine-tuned using a freeze-unfreeze approach, and Conditional Random Fields (CRF) post-processing was introduced for extracting smaller-sized solid waste objects. Results show that the improved MSBDN performs well overall in small object extraction. However, due to insufficient feature learning, some misdetections of gravel piles still occurred. The experiments verified both the feasibility and potential for improvement of this approach, providing crucial technical support for intelligent small-object recognition in complex mining scenarios.

This study focuses on the intelligent interpretation of remote sensing imagery in mining areas. To address challenges such as the high complexity of degradation zone identification and frequent omission of small objects, a comprehensive technical approach was developed with remarkable outcomes. First, through systematic analysis of spectral, textural, and spatial distribution features of land degradation elements, an interpretation framework covering typical features was established, and a pixel-level annotated dataset was created to support model training. Methodologically, an innovative MSBDN model integrating decoupled attention mechanisms and multi-scale feature fusion was designed, incorporating a dual-branch structure to extract global and local features separately, thereby significantly improving recognition and extraction performance. Furthermore, combining transfer learning with CRF post-processing enhanced the accuracy of solid waste interpretation within degraded zones. These research achievements not only advance the application of intelligent remote sensing interpretation in mining areas but also offer critical technical support for resource-environment monitoring, ecological evaluation, and intelligent management, with strong engineering value and practical significance.

[1] 柳晓娟,侯华丽,孙映祥,等.关于中国绿色矿业内涵与实现路径的思考[J].矿业研究与开发,2021,41(10):180-186.

[2] 朱迪,吴泽斌.失调与调适:矿产资源开发利用引发的社会问题及对策分析[J].中国矿业,2020,29(07):1-8.

[3] 张绍良,朱立军,侯湖平,等.“五位一体”视域下的矿山生态修复[J].环境保护,2014,42(Z1):72-74.

[4] 《全国重要生态系统保护和修复重大工程总体规划（2021—2035年）》发布[J].腐植酸,2020,(04):79.

[5] 2030年前碳达峰行动方案[J].中国科技奖励,2021,(12):7-15.

[6] 朱武婧.矿产资源开发与生态环境保护协调发展探究[J].黑龙江环境通报,2024,37(12):116-119.

[7] 高成虎.榆神矿区某煤矿矿山地质环境影响评估与恢复治理方案研究[D].长安大学,2018.

[8] 杨金中,聂洪峰,荆青青.初论全国矿山地质环境现状与存在问题[J].国土资源遥感,2017,29(02):1-7.

[9] 张成业,李军,雷少刚,等.矿区生态环境定量遥感监测研究进展与展望[J].金属矿山,2022,(03):1-27.

[10] 黄登冕,张聪,姚晓军,等.矿山环境遥感监测研究进展[J].遥感技术与应用,2022,37(05):1043-1055.

[11] Li J, Cai Y, Li Q, et al. A review of remote sensing image segmentation by deep learning methods[J]. International Journal of Digital Earth, 2024, 17(1): 2328827.

[12] Kussul N, Lavreniuk M, Skakun S, et al. Deep learning classification of land cover and crop types using remote sensing data[J]. IEEE Geoscience and Remote Sensing Letters, 2017, 14(5): 778-782.

[13] Ji C Y, Liu Q, Sun D, et al. Monitoring urban expansion with remote sensing in China[J]. International Journal of Remote Sensing, 2001, 22(8): 1441-1455.

[14] Yang L, Cervone G. Analysis of remote sensing imagery for disaster assessment using deep learning: a case study of flooding event[J]. Soft Computing, 2019, 23(24): 13393-13408.

[15] Song W, Song W, Gu H, et al. Progress in the remote sensing monitoring of the ecological environment in mining areas[J]. International journal of environmental research and public health, 2020, 17(6): 1846.

[16] Chen Y, Lin Z, Zhao X, et al. Deep learning-based classification of hyperspectral data[J]. IEEE Journal of Selected topics in applied earth observations and remote sensing, 2014, 7(6): 2094-2107.

[17] Kotaridis I, Lazaridou M. Remote sensing image segmentation advances: A meta-analysis[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2021, 173: 309-322.

[18] 白方铄,马骁,杨永均,等.基于三参数阈值分割的露天矿山土地遥感分类方法[J].测绘与空间地理信息,2024,47(05):27-30.

[19] 龙亦凡,乔雯钰,孙静.基于SVM的大屯矿区遥感影像变化检测[J].测绘与空间地理信息,2020,43(12):107-110+115.

[20] Huangfu W, Qiu H, Cui P, et al. Automated extraction of mining‐induced ground fissures using deep learning and object‐based image classification[J]. Earth Surface Processes and Landforms, 2024, 49(7): 2189-2204.

[21] Kotaridis I, Lazaridou M. Delineation of Open-Pit Mining Boundaries on Multispectral Imagery[J]. Kotaridis, I. and Lazaridou, M.(2020).‘Delineation of Open-Pit Mining Boundaries on Multispectral Imagery’. in Remote Sensing. IntechOpen. doi, 2020, 10.

[22] Hinton GE, Salakhutdinov RR. Reducing the dimensionality of data with neural networks[J]. Science, 2006, 313(5786): 504-507.

[23] Long J, Shelhamer E, Darrell T. Fully convolutional networks for semantic segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 39(4):640-651.

[24] Ronneberger O, Fischer P, Brox T. U-net: Convolutional networks for biomedical image segmentation[C]. Medical Image Computing and Computer-Assisted Intervention–MICCAI 2015: 18th International Conference, 2015: 234-241.

[25] Badrinarayanan V, Kendall A, Cipolla R. Segnet: A deep convolutional encoder-decoder architecture for image segmentation[J]. IEEE transactions on pattern analysis and machine intelligence, 2017, 39(12): 2481-2495.

[26] Chen L C, Papandreou G, Kokkinos I, et al. Semantic image segmentation with deep convolutional nets and fully connected crfs[J]. arXiv preprint arXiv:1412.7062, 2014.

[27] Chen L C, Papandreou G, Kokkinos I, et al. Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs[J]. IEEE transactions on pattern analysis and machine intelligence, 2017, 40(4): 834-848.

[28] Chen L C, Papandreou G, Schroff F, et al. Rethinking atrous convolution for semantic image segmentation[J]. arXiv preprint arXiv:1706.05587, 2017.

[29] Chen L C, Zhu Y, Papandreou G, et al. Encoder-decoder with atrous separable convolution for semantic image segmentation[C]. Proceedings of the European conference on computer vision (ECCV). 2018: 801-818.

[30] Lin G, Milan A, Shen C, et al. Refinenet: Multi-path refinement networks for high-resolution semantic segmentation[C]. Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 1925-1934.

[31] Zhao H, Shi J, Qi X, et al. Pyramid scene parsing network[C]. Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 2881-2890.

[32] Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need[J]. Advances in neural information processing systems, 2017, 30.

[33] Dosovitskiy A, Beyer L, Kolesnikov A, et al. An image is worth 16x16 words: Transformers for image recognition at scale[J]. arXiv preprint arXiv:2010.11929, 2020.

[34] Liu Z, Lin Y, Cao Y, et al. Swin transformer: Hierarchical vision transformer using shifted windows[C]. Proceedings of the IEEE/CVF international conference on computer vision. 2021: 10012-10022.

[35] Zheng S, Lu J, Zhao H, et al. Rethinking semantic segmentation from a sequence-to-sequence perspective with transformers[C]. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2021: 6881-6890.

[36] Strudel R, Garcia R, Laptev I, et al. Segmenter: Transformer for semantic segmentation[C]. Proceedings of the IEEE/CVF international conference on computer vision. 2021: 7262-7272.

[37] Xie E, Wang W, Yu Z, et al. SegFormer: Simple and efficient design for semantic segmentation with transformers[J]. Advances in Neural Information Processing Systems, 2021, 34: 12077-12090.

[38] 郭新, 张斌, 程坤. 面向小目标提取的改进DeepLabV3+模型遥感图像分割[J]. 遥感信息, 2022, 37(02): 34-44.

[39] Li R, Liu W, Yang L, et al. DeepUNet: A deep fully convolutional network for pixel-level sea-land segmentation[J]. IEEE journal of selected topics in applied earth observations and remote sensing, 2018, 11(11): 3954-3962.

[40] Xu Z, Su C, Zhang X. A semantic segmentation method with category boundary for Land Use and Land Cover (LULC) mapping of Very-High Resolution (VHR) remote sensing image[J]. International Journal of Remote Sensing, 2021, 42(8): 3146-3165.

[41] Shi W, Zhang M, Ke H, et al. Landslide recognition by deep convolutional neural network and change detection[J]. IEEE Transactions on Geoscience and Remote Sensing, 2020, 59(6): 4654-4672.

[42] Liu X, Peng Y, Lu Z, et al. Feature-Fusion Segmentation Network for Landslide Detection Using High-Resolution Remote Sensing Images and Digital Elevation Model Data[J]. IEEE Transactions on Geoscience and Remote Sensing, 2023, 61: 1-14.

[43] Sun W, Wang R. Fully convolutional networks for semantic segmentation of very high resolution remotely sensed images combined with DSM[J]. IEEE Geoscience and Remote Sensing Letters, 2018, 15(3): 474-478.

[44] 杨飞霞. 遥感图像特征提取与融合方法的研究[D]. 北京邮电大学, 2021.

[45] Hu J, Shen L, Sun G. Squeeze-and-excitation networks[C]. Proceedings of the IEEE conference on computer vision and pattern recognition. 2018: 7132-7141.

[46] Guo M H, Xu T X, Liu J J, et al. Attention mechanisms in computer vision: A survey[J]. Computational visual media, 2022, 8(3): 331-368.

[47] Li H, Qiu K, Chen L, et al. SCAttNet: Semantic segmentation network with spatial and channel attention mechanism for high-resolution remote sensing images[J]. IEEE Geoscience and Remote Sensing Letters, 2020, 18(5): 905-909.

[48] Yang X, Fan X, Peng M, et al. Semantic segmentation for remote sensing images based on an AD-HRNet model[J]. International Journal of Digital Earth, 2022, 15(1): 2376-2399.

[49] Sun K, Xiao B, Liu D, et al. Deep high-resolution representation learning for human pose estimation[C]. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2019: 5693-5703.

[50] 谭琨, 王雪, 杜培军. 结合深度学习和半监督学习的遥感影像分类进展[J]. 中国图象图形学报, 2019, 24(11): 1823-1841.

[51] Goodfellow I J,Pouget-Abadie J,Mirza M,et al.Generative adversarial nets[C]. Proceedings of the 27th International Conference on Neural Information Processing Systems. Montreal:MITPress,2014

[52] Kingma D P,Welling M.Auto-encoding variational bayes[EB/OL].

[53] 黄聪. 基于生成对抗网络的遥感图像公路沿线地物分割研究[D]. 兰州交通大学, 2023.

[54] Zhu J Y, Park T, Isola P, et al. Unpaired image-to-image translation using cycle-consistent adversarial networks[C]. Proceedings of the IEEE international conference on computer vision. 2017: 2223-2232.

[55] 曾发明, 杨波, 吴德文等. 基于Canny边缘检测算子的矿区道路提取[J]. 国土资源遥感, 2013, 25(04): 72-78.

[56] 范莹琳, 杜松, 赵岳, 等. 基于随机森林算法的煤矸石山信息提取[J]. 自然资源遥感.

[57] 袁定波, 刘成林, 汪国斌. 面向对象的矿区信息提取方法的应用与研究[J]. 遥感信息, 2013, 28(02): 110-115.

[58] 程璐. 面向对象结合支持向量机（SVM）在露天矿区信息提取中的应用研究[D]. 青海大学, 2017.

[59] 刘佳丽. 基于遥感的露天灰岩矿山开采信息提取[D]. 华北理工大学, 2019.

[60] 程国轩, 牛瑞卿, 张凯翔, 赵凌冉. 基于卷积神经网络的高分遥感影像露天采矿场识别[J]. 地球科学, 2018, 43(S2): 256-262.

[61] 张峰极, 吴艳兰, 姚雪东等. 基于改进DenseNet网络的多源遥感影像露天开采区智能提取方法[J]. 遥感技术与应用, 2020, 35(03): 673-684.

[62] Lyu J, Hu Y, Ren S, et al. Extracting the tailings ponds from high spatial resolution remote sensing images by integrating a deep learning-based model[J]. Remote Sensing, 2021, 13(4): 743.

[63] Xie H, Pan Y, Luan J, et al. Open-pit mining area segmentation of remote sensing images based on DUSegNet[J]. Journal of the Indian Society of Remote Sensing, 2021, 49(6): 1257-1270.

[64] 张月树,马杰,邵颖,等.陕南秦巴山区地质-生态环境评价与可持续发展[J].陕西地质,2000,(02):77-89.

[65] Hecker C, van Ruitenbeek F J A, van der Werff H M A, et al. Spectral absorption feature analysis for finding ore: A tutorial on using the method in geological remote sensing[J]. IEEE geoscience and remote sensing magazine, 2019, 7(2): 51-71.

[66] 陈彦.矿山地质环境影响和土地损毁预测评估研究[J].能源与环保,2022,44(08):19-26.

[67] 田伟学.基于Transformer的目标检测和语义分割模型在煤矿遥感影像的应用[D].中国矿业大学,2023.

[68] 格日乐,王丽,秦丽媛.“双碳”战略目标下煤基固废材料的高质化利用与矿山生态修复的战略思考[J].内蒙古煤炭经济,2024,(03):119-122.

[69] Woo S, Park J, Lee J Y, et al. Cbam: Convolutional block attention module[C]//Proceedings of the European conference on computer vision (ECCV). 2018: 3-19.

[70] Srivastava N, Hinton G, Krizhevsky A, et al. Dropout: a simple way to prevent neural networks from overfitting[J]. The journal of machine learning research, 2014, 15(1): 1929-1958.

[71] Fan J, Bocus M J, Hosking B, et al. Multi-scale feature fusion: Learning better semantic segmentation for road pothole detection[C]//2021 IEEE international conference on autonomous systems (ICAS). IEEE, 2021: 1-5.

[72] Jadon S. A survey of loss functions for semantic segmentation[C]//2020 IEEE conference on computational intelligence in bioinformatics and computational biology (CIBCB). IEEE, 2020: 1-7.

[73] Lin T Y, Goyal P, Girshick R, et al. Focal loss for dense object detection[C]//Proceedings of the IEEE international conference on computer vision. 2017: 2980-2988.

[74] Berman M, Triki A R, Blaschko M B. The lovász-softmax loss: A tractable surrogate for the optimization of the intersection-over-union measure in neural networks[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2018: 4413-4421.

[75] Torrey L, Shavlik J. Transfer learning[M]//Handbook of research on machine learning applications and trends: algorithms, methods, and techniques. IGI global, 2010: 242-264.

[76] Weiss K, Khoshgoftaar T M, Wang D D. A survey of transfer learning[J]. Journal of Big data, 2016, 3: 1-40.

[77] Yosinski J, Clune J, Bengio Y, et al. How transferable are features in deep neural networks?[J]. Advances in neural information processing systems, 2014, 27.

[78] Lafferty J, McCallum A, Pereira F. Conditional random fields: Probabilistic models for segmenting and labeling sequence data[C]//Icml. 2001, 1(2): 3.