煤炭作为我国能源结构的主导资源，其高效清洁利用对实现“双碳”目标至关重要。然而，煤炭开采过程中混杂的煤矸石不仅降低原煤品质，还造成资源浪费与环境污染。传统分选方法依赖人工或物理特性差异，存在效率低、适应性差等问题。基于深度学习的检测技术虽取得进展，但洗煤厂复杂场景下图像质量不高、目标模糊、小样本数据等挑战仍制约检测精度。针对上述问题，本文研究融合超分辨率重建与深度学习的煤矸石检测方法。

针对煤矸石检测任务中因场景图像导致的细节丢失问题，提出了一种融合超分辨率重建优化的改进方法。对于场景图像导致的细节丢失，构建基于增强型超分辨率生成对抗网络的重建优化。采用密集残差连接并去除批量归一化层，减少伪影生成，增强模型泛化能力，改进生成器网络结构。引入锐化损失模块，构建生成图像、真实图像与锐化图像的三方对抗机制，提升目标边缘清晰度。经超分辨率重建后，图像热力图中目标区域特征聚焦度明显增强，为后续检测奠定高质量数据基础。采用YOLOv7目标识别算法，通过多分支堆叠模块实现多尺度特征提取，结合空间金字塔池化与图像锐化模块，增强复杂背景下煤矸石特征的表征能力。实验结果表明，融合超分辨率与图像锐化模块的改进YOLOv7模型在煤、矸石检测任务中平均精度分别达97.51%与98.29%，较基线YOLOv7提升3.39%与6.82%；mAP@50为97.90%，F1-Score达0.96，较YOLOv5、Efficientnet等主流算法平均提升7.3%。且系统在非均匀光照、粉尘干扰场景保持识别稳定性，漏检率降低18%，优于对比算法。

针对小样本场景下的煤矸石检测任务，研究小样本分选场景的优化策略，提出了一种基于迁移学习的优化策略。该方法复用预训练模型的特征，并结合卷积与注意力机制混合模块，增强模型对煤矸石目标的特征表达能力。迁移学习在源任务上预训练模型，并将其迁移到目标任务中进行微调，使模型能够更快适应目标数据分布，从而加速收敛并提升检测准确率。基于大规模数据集的预训练模型在煤矸石检测任务中进行迁移学习，使其能够充分利用已有知识，提高检测的稳定性和准确性。卷积与注意力机制混合模块结合了卷积操作的局部特征提取能力和自注意力机制的长距离依赖建模能力，使模型在识别小目标时更加敏感，同时提升在复杂背景下的特征学习能力。实验结果表明，所提方法在煤和矸石检测任务中平均精度分别达94.40%和87.64%，mAP@50为91.03%，模型的检测性能和泛化能力得到提升。

基于上述研究，设计并实现了一种基于图像处理和深度学习技术的煤矸石识别系统。在系统中，集成了前文提出的融合超分辨率重建的煤矸石检测方法，支持用户上传煤矸石图片并进行自动识别分类。系统包含用户管理、图像识别等核心模块，不仅实现了对煤矸石的高精度检测，对检测结果进行可视化展现，还提供了后台管理功能，便于数据存储与分析，以适应实际应用需求。

Coal is the dominant resource in China’s energy structure, and its efficient and clean utilization is crucial for achieving the "dual carbon" goals. However, the coal gangue mixed with coal during mining not only lowers the quality of the raw coal but also leads to resource waste and environmental pollution. Traditional sorting methods rely on manual labor or differences in physical characteristics, which suffer from low efficiency and poor adaptability. While deep learning-based detection technologies have made progress, challenges like low image quality, blurred targets, and insufficient sample data in complex coal washing plant environments still limit detection accuracy. This paper investigates a coal gangue detection method that integrates super-resolution reconstruction and deep learning.

To tackle the problem of detail loss caused by scene images in coal gangue detection tasks, an improved method integrating super-resolution reconstruction optimization is proposed. For the detail loss caused by scene images, a reconstruction optimization based on an enhanced super-resolution generative adversarial network (SRGAN) is constructed. Dense residual connections are used, and batch normalization layers are removed to reduce the generation of artifacts and enhance the model’s generalization ability, improving the generator network structure. A sharpening loss module is introduced to establish a tripartite adversarial mechanism between generated images, real images, and sharpened images, thus enhancing the clarity of target edges. After super-resolution reconstruction, the focus on the target region’s features in the image heatmap is significantly enhanced, laying the foundation for high-quality data for subsequent detection. The YOLOv7 target recognition algorithm is employed, using multi-branch stacking modules to achieve multi-scale feature extraction. This is combined with spatial pyramid pooling and image sharpening modules to improve the representation capability of coal gangue features in complex backgrounds. Experimental results show that the improved YOLOv7 model with integrated super-resolution and image sharpening modules achieves average accuracies of 97.51% and 98.29% for coal and gangue detection tasks, respectively, an improvement of 3.39% and 6.82% over the baseline YOLOv7. The mAP@50 is 97.90%, and the F1-Score reaches 0.96, representing a 7.3% average improvement over mainstream algorithms such as YOLOv5 and EfficientNet. Moreover, the system maintains recognition stability under non-uniform lighting and dust-interfered environments, with an 18% reduction in missed detection rates, outperforming comparison algorithms.

For the coal gangue detection task in the small-sample scenario, an optimization strategy for the small-sample sorting scenario was studied, and an optimization strategy based on transfer learning was proposed. This method reuses the features of the pre-trained model and combines a hybrid module of convolution and attention mechanism to enhance the model's feature expression ability for coal gangue targets.Transfer learning involves pre-training the model on a source task and fine-tuning it on the target task, enabling the model to adapt to the target data distribution faster, thereby accelerating convergence and improving detection accuracy. Transfer learning using pre-trained models on large-scale datasets allows the model to make full use of existing knowledge, improving detection stability and accuracy. The convolution and attention mechanism hybrid module combines the local feature extraction ability of convolution operations with the long-range dependency modeling ability of self-attention mechanisms, making the model more sensitive when recognizing small targets and enhancing feature learning in complex backgrounds. Experimental results show that the proposed method achieves average accuracies of 94.40% and 87.64% for coal and gangue detection tasks, respectively, with an mAP@50 of 91.03%, outperforming traditional YOLO series algorithms.

Based on the above research, a coal gangue recognition system based on image processing and deep learning techniques has been designed and implemented. The system integrates the coal gangue detection method that combines super-resolution reconstruction, allowing users to upload coal gangue images for automatic recognition and classification. The system includes core modules such as user management and image recognition, achieving high-precision detection of coal gangue, visualizing detection results, and providing backend management functions for data storage and analysis to meet practical application requirements.

[1] 张强, 张润鑫, 刘峻铭, 等. 煤矿智能化开采煤岩识别技术综述[J]. 煤炭科学技术, 2022, 50(02): 1-26.

[2] 梅培林. 基于深度学习的煤矸石识别检测技术研究[D]. 中国矿业大学, 2023.

[3] 李娜, 秦昆德, 李想, 等. 融合超分辨率重建的YOLOv7煤矸石识别模型[J/OL]. 计算机工程与应用, 1-11.

[4] Zeng Q, Zhou G, Wan L, et al. Detection of coal and Gangue based on improved YOLOv8[J]. Sensors, 2024, 24(4): 1246.

[5] Bessinger S L, Nelson M G. Remnant roof coal thickness measurement with passive gamma ray instruments in coal mines[J]. IEEE Transactions on Industry Applications, 1993, 29(3): 562-565.

[6] 毋林林. 粉煤灰充填材料的井下安全性研究[D]. 太原理工大学, 2015.

[7] 宋忠广. 七台河矿区煤矸石特征及综合利用研究[D]. 辽宁工程技术大学, 2007.

[8] 董长双, 姚平喜, 刘志河. 井下煤和矸石液压式自动分选技术[J]. 煤炭科学技术, 2007, 35(3): 54-56.

[9] 郑克洪, 杜长龙, 刘飞. 煤和矸石选择性破碎分选理论研究[J]. 选煤技术, 2012 (1): 23-25.

[10] 刘瑜. 井下冲撞式煤矸分离中颗粒动力学行为研究[D]. 中国矿业大学,2011.

[11] 孔力, 李红, 徐恕宏, 等. 双能γ射线透射法煤矸石在线识别与分选系统[J]. 华中理工大学学报, 1997, 25(10): 107-108.

[12] 袁华昕. 基于X射线图像的煤矸石智能分选控制系统研究[D]. 东北大学,2014.

[13] Zhao Y, Wang S, Guo Y, et al. The identification of coal and gangue and the prediction of the degree of coal metamorphism based on the EDXRD principle and the PSO-SVM model[J]. Gospodarka Surowcami Mineralnymi-Mineral Resources Management, 2022: 113-129.

[14] 马宪民, 蒋勇, 卜祥莉. 基于图像处理的煤矸石自动分选系统的研究[C]//中国自动化学会智能自动化专业委员会. 2003年中国智能自动化会议论文集（下册）. 西安科技大学电控学院. 2003:403-406.

[15] 薛祯也.基于深度学习的煤矸石识别方法研究[D]. 西安科技大学,2019.

[16] 王莉, 于国防, 沈慧宇, 等. 基于CNN卷积神经网络的煤矸石自动分选研究[J]. 江苏建筑职业技术学院学报, 2019, 19(4): 35-39.

[17] Lv Z, Wang W, Xu Z, et al. Cascade network for detection of coal and gangue in the production context[J]. Powder Technology, 2021, 377: 361-371.

[18] Ding Z, Chen G, Wang Z, et al. A Real-Time Multilevel Fusion Recognition System for Coal and Gangue Based on Near-Infrared Sensing[J]. IEEE Access, 2020, 8: 178722-178732.

[19] 吕旻姝. 基于深度学习的煤矸石图像识别研究[D]. 安徽理工大学, 2021.

[20] 苏玲玲. 基于深度学习的煤矸石识别方法研究[D]. 西安科技大学, 2018.

[21] 赵学军, 李建. 一种基于深度学习的煤矸石检测方法[J]. 矿业科学学报, 2021, 6(06): 730-736.

[22] Li M, He X, Yuan Y, et al. Multiple factors influence coal and gangue image recognition method and experimental research based on deep learning[J]. International Journal of Coal Preparation and Utilization, 2023, 43(8): 1411-1427.

[23] Xu S, Zhou Y, Huang Y, et al. YOLOv4-tiny-based coal gangue image recognition and FPGA implementation[J]. Micromachines, 2022, 13(11): 1983.

[24] 汝洪芳, 张冬冬. YOLOv5检测煤矸石的改进方法[J]. 黑龙江科技大学学报, 2021, 31(06): 818-823.

[25] Dong C, Loy C C, He K, et al. Learning a deep convolutional network for image super-resolution[C]//Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part IV 13. Springer International Publishing, 2014: 184-199.

[26] Shi W, Caballero J, Huszár F, et al. Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 1874-1883.

[27] Kim J, Lee J K, Lee K M. Accurate image super-resolution using very deep convolutional networks[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 1646-1654.

[28] Kim S, Kang I, Kwak N. Semantic sentence matching with densely-connected recurrent and co-attentive information[C]//Proceedings of the AAAI conference on artificial intelligence. 2019, 33(01): 6586-6593.

[29] He K, Zhang X, Ren S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 770-778.

[30] Zhang Y, Tian Y, Kong Y, et al. Residual dense network for image super-resolution[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2018: 2472-2481.

[31] Ledig C, Theis L, Huszár F, et al. Photo-realistic single image super-resolution using a generative adversarial network[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 4681-4690.

[32] Liu Y, Wu Y H, Sun G, et al. Vision transformers with hierarchical attention[J]. Machine Intelligence Research, 2024, 21(4): 670-683.

[33] Wang X, Yu K, Wu S, et al. ESRGAN: enhanced super-resolution generative adversarial networks[C]//European Conference on Computer Vision,2019:63-79.

[34] Ji H, Gao Z, Mei T, et al. Vehicle detection in remote sensing images leveraging on simultaneous super-resolution[J]. IEEE Geoscience and Remote Sensing Letters, 2019, 17(4): 676-680.

[35] Rabbi J, Ray N, Schubert M, et al. Small-object detection in remote sensing images with end-to-end edge-enhanced GAN and object detector network[J]. Remote Sensing, 2020, 12(9): 1432-1456.

[36] He S, Zou H, Wang Y, et al. ShipSRDet: An end-to-end remote sensing ship detector using super-resolved feature representation[C]//2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS. IEEE, 2021: 3541-3544.

[37] 张天霖, 逄征, 陈红珍, 等. 面向舰船目标识别的遥感图像超分辨率重建[J]. 计算机工程与应用,2024,60(13):190-199.

[38] 王文瑾, 游子绎, 邵历江, 等. 融合超分辨率重建的YOLOv5松枯死木识别模型[J]. 农业工程学报,2023,39(05):137-145.

[39] 周飞燕, 金林鹏, 董军. 卷积神经网络研究综述[J]. 计算机学报,2017,40(6):1229-1251.

[40] Hou C, Liu G, Tian Q, et al. Multisignal modulation classification using sliding window detection and complex convolutional network in frequency domain[J]. IEEE Internet of Things Journal, 2022, 9(19): 19438-19449.

[41] Hara K, Saito D, Shouno H. Analysis of function of rectified linear unit used in deep learning[C]//2015 international joint conference on neural networks (IJCNN). IEEE, 2015: 1-8.

[42] Jiang P, Ergu D, Liu F, et al. A Review of Yolo algorithm developments[J]. Procedia computer science, 2022, 199: 1066-1073.

[43] Liu W, Anguelov D, Erhan D, et al. Ssd: Single shot multibox detector[C]//Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11–14, 2016, Proceedings, Part I 14. Springer International Publishing, 2016: 21-37.

[44] Lienhart R, Maydt J. An extended set of haar-like features for rapid object detection[C]//Proceedings. international conference on image processing. IEEE, 2002, 1: I-I.

[45] Wang H, Hu D. Comparison of SVM and LS-SVM for regression[C]//2005 International conference on neural networks and brain. IEEE, 2005, 1: 279-283.

[46] He K, Gkioxari G, Dollár P, et al. Mask r-cnn[C]//Proceedings of the IEEE international conference on computer vision. 2017: 2961-2969.

[47] Girshick R. Fast r-cnn[C]//Proceedings of the IEEE international conference on computer vision. 2015: 1440-1448.

[48] Ren S, He K, Girshick R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[J]. IEEE transactions on pattern analysis and machine intelligence, 2016, 39(6): 1137-1149.

[49] Wang Y, Wang C, Zhang H, et al. Automatic ship detection based on RetinaNet using multi-resolution Gaofen-3 imagery[J]. Remote Sensing, 2019, 11(5): 531-545.

[50] Tian J, Ma K K. A survey on super-resolution imaging[J]. Signal, Image and Video Processing, 2011, 5: 329-342.

[51] Lee H, Battle A, Raina R, et al. Efficient sparse coding algorithms[J]. Advances in neural information processing systems, 2006, 19-27.

[52] Timofte R, De Smet V, Van Gool L. Anchored neighborhood regression for fast example-based super-resolution[C]//Proceedings of the IEEE international conference on computer vision. 2013: 1920-1927.

[53] Wang C Y, Bochkovskiy A, Liao H Y M. YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2023: 7464-7475.

[54] Wang C Y, Yeh I H, Mark Liao H Y. Yolov9: Learning what you want to learn using programmable gradient information[C]//European conference on computer vision. Cham: Springer Nature Switzerland, 2024: 1-21.

[55] 肖万新, 李华锋, 张亚飞, 等. 多尺度特征学习和边缘增强的医学图像融合[J]. Laser & Optoelectronics Progress, 2022, 59(6): 0617029-0617029-10.

[56] Bjorck N, Gomes C P, Selman B, et al. Understanding batch normalization[J]. Advances in neural information processing systems, 2018, 31.

[57] Wang X. Laplacian operator-based edge detectors[J]. IEEE transactions on pattern analysis and machine intelligence, 2007, 29(5): 886-890.

[58] Pan X, Ge C, Lu R, et al. On the integration of self-attention and convolution[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2022: 815-825.