在煤炭开采、运输和加工过程中，常常混入矸石、木块、铁器等异物，这些异物不 仅影响煤炭质量，还可能对输送设备造成损害，甚至引发安全事故。因此，煤矿皮带异 物检测的研究对保障煤矿安全生产至关重要。然而，在煤矿场景中异物形态多样，背景 复杂，同时受行业安全限制，数据获取不易，导致传统基于大数据训练的目标检测方法 受阻。为了解决在少量标注数据下实现高效，精准的煤矿皮带异物检测问题，本文改进 了基于小样本学习的目标检测方法，旨在提升模型在有限数据条件下的检测和泛化能力。 （1）针对现有煤矿皮带异物检测方法普遍依赖大量标注样本，难以适应小样本检 测场景的问题，本文提出了一种基于迁移学习的小样本煤矿皮带异物检测方法，有效解 决了少样本情况下背景遮挡，异物区分困难的问题。首先集成多维协作注意力模块于特 征提取网络中，通过改进Bottleneck 结构设计，强化了模型在多维特征空间中的提取能 力，提升模型对遮挡异物的检测精度。其次引入渐进式特征金字塔网络，实现了高效的 跨尺度特征融合，增强了模型对不同尺寸异物的识别能力。最后，在少量样本的微调阶 段，设计判别损失函数，引导分类器学习更加具有区分度的特征表示，从而能够更加准 确地辨别出煤矿复杂环境中的各类异物，提高了分类的准确性和鲁棒性。实验结果表明， 改进后的算法在公共数据集PASCAL VOC和自建小样本煤矿皮带异物检测数据集上均 展示出良好的检测性能，在不同 shot 划分下，其检测精度均高于基线算法。在自建数 据集的 30-shot 设置中，新类别异物的平均检测精度达到了 83.6%，较两阶段微调方法 提升了7%。 （2）针对小样本煤矿皮带异物检测中的细长异物定位困难和类别特征表示不足的 问题，本文提出了一种基于特征增强和多模态融合的小样本煤矿皮带异物检测方法。该 方法基于元学习的检测 Transformer（Meta-learning-based Detection Transformer, Meta DETR）进行改进。首先设计了特征增强模块，通过抑制冗余信息和噪声，提升支持特 征的表达能力，使模型在少样本情况下仍能更好的提取关键特征信息。其次，在网络结 构中引入类别的文本信息，通过构建多模态融合模块，将文本信息与支持查询特征融合， 生成更具语义表达能力的特征，克服了单一模态的局限性，从而提升模型的泛化能力。 最后，为了提高细长异物的检测准确性，采用形状感知交并比损失函数(Shape-aware Intersection over Union, Shape-IoU)优化边界框回归过程，从而增强模型对细长异物的定 位能力。实验结果表明，所提方法在公共数据集PASCAL VOC上表现出良好的有效性 和可靠性。在小样本煤矿皮带检测任务中，该模型在检测精度方面优于近年来主流的小 样本目标检测模型，尤其是在极低1-shot 情况下，其检测精度较 Meta-DETR 模型提升 了5.8%。

In the process of coal mining, transportation, and processing, foreign objects such as gangue, wood, and metal debris are often mixed in. These foreign objects not only affect the quality of coal but may also damage conveying equipment and even lead to safety incidents. Therefore, research on foreign object detection on coal mine conveyor belts is crucial for ensuring safe production in coal mines. However, in coal mine scenarios, foreign objects exhibit diverse forms against complex backgrounds, and due to industry safety restrictions, data collection is challenging, hindering traditional big-data-trained object detection methods. To address the problem of achieving efficient and accurate foreign object detection on coal mine conveyor belts with a small amount of annotated data, the target detection methods based on few-shot learning are improved in this thesis, aiming to enhance the detection and generalization capabilities of models under constrained data conditions. (1) Given that existing methods for foreign object detection on coal mine conveyor belts generally rely on large amounts of annotated samples and struggle to adapt to few-shot detection scenarios, this thesis proposes a few-shot foreign object detection method based on transfer learning, effectively addressing issues such as background occlusion and difficulty in distinguishing foreign objects with limited samples. First, a multi-dimensional collaborative attention module is integrated into the feature extraction network. By improving the Bottleneck structure design, the model's feature extraction capability in multi-dimensional feature spaces is enhanced, thereby improving detection accuracy for occluded foreign objects. Second, a progressive feature pyramid network is introduced to achieve efficient cross-scale feature fusion, strengthening the model's ability to recognize foreign objects of varying sizes. Finally, during the fine-tuning stage with limited samples, a discriminative loss function is designed to guide the classifier in learning more distinctive feature representations, enabling more accurate identification of various foreign objects in the complex coal mine environment and improving classification accuracy and robustness. Experimental results demonstrate that the improved algorithm exhibits strong detection performance on both the public PASCAL VOC dataset and the self-constructed few-shot coal mine conveyor belt foreign object detection dataset. Under different shot settings, its detection accuracy surpasses that of the baseline algorithm. In the 30-shot setting of the self-built dataset, the average detection accuracy for new category foreign objects reached 83.6%, which is 7% higher than that of the two-stage fine-tuning method. (2) To address the challenges of locating slender foreign objects and insufficient category feature representation in few-shot foreign object detection on coal mine conveyor belts, this thesis proposes a few-shot detection method based on feature enhancement and multi-modal fusion. This method improves the Meta-DETR baseline network in three aspects. First, a feature enhancement module is designed to suppress redundant information and noise, enhancing the expressive power of support features, enabling the model to better extract key feature information even with limited samples. Second, textual information of categories is incorporated into the network structure. By constructing a multi-modal fusion module, textual information is fused with support and query features to generate more semantically expressive features, overcoming the limitations of a single modality and thereby improving the model's generalization ability. Finally, to enhance detection accuracy for slender foreign objects, the Shape-IoU loss function is adopted to optimize the bounding box regression process, strengthening the model's localization capability for slender foreign objects. Experimental results show that the proposed method exhibits strong effectiveness and reliability on the public PASCAL VOC dataset. In the few-shot coal mine conveyor belt detection task, the model outperforms mainstream few-shot object detection models in terms of detection accuracy, particularly in the extremely low 1-Shot scenario, where its detection accuracy improves by 5.8% compared to the Meta-DETR model.

[1] 王国法,富佳兴,王忠鑫.煤矿智能化重要进展与高质量发展方向[J].智能矿

山,2025,6(01):2-12.

[2] 杨梓,杨沐岩.我国煤矿安全生产跨入新阶段[N].中国能源报,2024-02-26(005).

[3] 张硕.基于 AI 技术的矿用带式输送机驱动装置故障诊断预警研究[J].煤矿机

电,2023,44(05):59-63.

[4] 郭强.煤矿井下带式输送机的应用及运行问题探析[J].西部探矿工程,2024,36(03):43

45.

[5] 阴栋栋.带式输送机输送带撕裂原因分析及其防护技术研究[J].机械管理开

发,2023,38(12):231-233.

[6] Zhang X, Ning Y, Lu C. Evaluation of coal supply and demand security in China and

associated obstacle factors[J]. Sustainability, 2022, 14(17): 10605.

[7] 张强,张润鑫.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.

[8] Song Y, Wang T, Cai P, et al. A comprehensive survey of few-shot learning: Evolution,

applications, challenges, and opportunities[J]. ACM Computing Surveys, 2023, 55(13s):

1-40.

[9] 李济军.以自动化信息化融合为基础的智慧矿山建设探析[J].当代化工研

究,2023,(16):194-196.

[10] 王锐,桂志国,刘祎,等.基于 X 射线和结构光相机的煤矸石分拣方法研究[J].中北大学

学报(自然科学版),2021,42(02):123-128+134.

[11] 郭永存,何磊,刘普壮,等.煤矸双能 X 射线图像多维度分析识别方法[J].煤炭学

报,2021,46(01):300-309.

[12] 李素环,夏云凯.X 射线块煤智能分选机在脏杂煤分选中的应用[J].洁净煤技

术,2021,27(S2):47-52.

[13] 毕东月.基于深度学习的输煤皮带故障视觉检测方法研究[J].中国安全生产科学技

术,2021,17(08):84-90.

[14] 刘富强,钱建生,王新红,等.基于图像处理与识别技术的煤矿矸石自动分选[J].煤炭学

报,2000,(05):534-537.

[15] 李曼,段雍,曹现刚,等.煤矸分选机器人图像识别方法和系统[J].煤炭学

报,2020,45(10):3636-3644.

[16] 薛旭升,杨星云,齐广浩,等.煤矿带式输送机分拣机器人异物识别与定位系统设计[J].工矿自动化,2022,48(12):33-41.

[17] 王家臣,李良晖,杨胜利.不同照度下煤矸图像灰度及纹理特征提取的实验研究[J].煤

炭学报,2018,43(11):3051-3061.

[18] Alfarzaeai MS, Hu E, Peng W, Qiang N, Alkainaeai MMA. Coal Gangue Classification

Based on the Feature Extraction of the Volume Visual Perception ExM-SVM.Energies

2023; 16(4):2064.

[19] 陈立,杜文华,曾志强,等.基于小波变换的煤矸石自动分选方法[J].工矿自动

化,2018,44(12):60-64.

[20] 王燕,郭潇樯,刘新华.带式输送机大块异物视觉检测系统设计[J].机械科学与技

术,2021,40(12):1939-1943.

[21] Redmon J, Divvala S, Girshick R, et al. You only look once: Unified, real-time object

detection [C]//Proceedings of the IEEE Conference on Computer Vision and Pattern

Recognition. 2016: 779-788.

[22] Liu W, Anguelov D, Erhan D, et al. Ssd: Single shot multibox detector [C]//European

Conference on Computer Vision. Springer, Cham, 2016: 21-37.

[23] Ren S, He K, Girshick R, et al. Faster r-cnn: Towards real-time object detection with

region proposal networks[J]. Advances in Neural Information Processing Systems, 2015,

28: 91–99.

[24] Cai Z, Vasconcelos N. Cascade r-cnn: Delving into high quality object detection

[C]//Proceedings of the IEEE conference on computer vision and pattern recognition.

2018: 6154-6162.

[25] 吴守鹏,丁恩杰,俞啸. 基于改进 FPN 的输送带异物识别方法[J].煤矿安全，2019,

50(12): 127-130.

[26] 史凌凯,耿毅德,王宏伟,等.基于改进 Mask R-CNN 的刮板输送机铁质异物多目标检

测[J]. 工矿自动化, 2022, 48(10): 55-61.

[27] 赵健,王奕,王海峰,等.基于 TDConv 与统一注意力检测头的异物检测算法[J].矿业安

全与环保,2024,51(04):26-34.

[28] 陈世涛,张敏,栗超.基于 YOLOv5 的带式输送机煤堆异物检测[J].洁净煤技

术,2024,30(S2):12-18.

[29] 边铁山.基于 SE-YOLOv5 模型皮带异物检测算法研究[J].中国矿业,2024,33(07):127

134.

[30] 唐俊,李敬兆,石晴,等.基于 Faster-YOLOv7 的带式输送机异物实时检测[J].工矿自动

化,2023,49(11):46-52+66.

[31] 高涵,赵培培,于正,等.基于特征增强与 Transformer 的煤矿输送带异物检测[J].煤炭科学技术,2024,52(07):199-208.

[32] 洪炎,汪磊,苏静明,等.基于改进 YOLOv8 的煤矿输送带异物检测[J].工矿自动

化,2024,50(06):61-69.

[33] 曹现刚,李虎,王鹏,等.基于跨模态注意力融合的煤炭异物检测方法[J].工矿自动

化,2024,50(01):57-65.

[34] 李海军,孔繁程,魏嘉彧等.基于深度学习的小样本目标检测综述[J].兵工自

化,2024,43(01):35-42.

[35] 郭小萍,赵霄丰,李元.融入噪声的监督增强网络用于小样本数据增强[J].电子测量技

术,2024,47(20):109-116.

[36] 吴菁,杨邦勤,张银建,等.面向小样本苗绣图像的生成与识别研究[J].现代信息科

技,2025,9(02):24-32.

[37] 张子豪,赵德春,王子琼等.基于样本增强的帕金森病识别算法研究[J].生物医学工程

学杂志,2024,41(01):17-25+33.

[38] Wang X, Huang T E, Darrell T, et al. Frustratingly simple few-shot object detection[J].

arxiv preprint arxiv:2003.06957, 2020.

[39] Khandelwal S, Goyal R, Sigal L. Unit: Unified knowledge transfer for any-shot object

detection and segmentation[C]//Proceedings of the IEEE/CVF Conference on Computer

Vision and Pattern Recognition. 2021: 5951-5961.

[40] Yang Z, Zhang C, Li R, et al. Efficient few-shot object detection via knowledge

inheritance[J]. IEEE Transactions on Image Processing, 2022, 32: 321-334.

[41] Sun B, Li B, Cai S, et al. Fsce: Few-shot object detection via contrastive proposal

encoding[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern

recognition. 2021: 7352-7362.

[42] Lu Y, Chen X, Wu Z, et al. Decoupled metric network for single-stage few-shot object

detection[J]. IEEE Transactions on Cybernetics, 2022, 53(1): 514-525.

[43] 周志宇,王天一,左治江等.小样本条件下的公路建设项目场景识别与安全预警[J].江

汉大学学报(自然科学版),2024,52(01):80-90.

[44] 刘珂,林珊玲,师欣雨,等.基于多尺度上下文提取的小样本野生动物检测[J].液晶与显

示,2025,40(03):516-526.

[45] Fan Q, Zhuo W, Tang C K, et al. Few-shot object detection with attention-RPN and multi

relation detector[C]//Proceedings of the IEEE/CVF conference on computer vision and

pattern recognition. 2020: 4013-4022.

[46] Karlinsky L, Shtok J, Harary S, et al. Repmet: Representative-based metric learning for

classification and few-shot object detection[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2019: 5197-5206.

[47] 杜海顺,安文昊,张春海,等.基于渐进式学习和增强原型度量的小样本农作物病害识

别方法[J].农业机械学报,2024,55(12):344-353.

[48] Zhang G, Luo Z, Cui K, et al. Meta-detr: Image-level few-shot detection with inter-class

correlation exploitation[J]. IEEE transactions on pattern analysis and machine

intelligence, 2022, 45(11): 12832-12843.

[49] Yan X, Chen Z, Xu A, et al. Meta r-cnn: Towards general solver for instance-level low

shot learning[C]//Proceedings of the IEEE/CVF international conference on computer

vision. 2019: 9577-9586.

[50] Hu H, Bai S, Li A, et al. Dense relation distillation with context-aware aggregation for

few-shot object detection[C]//Proceedings of the IEEE/CVF conference on computer

vision and pattern recognition. 2021: 10185-10194.

[51] Kang B, Liu Z, Wang X, et al. Few-shot object detection via feature reweighting

[C]//Proceedings of the IEEE/CVF international conference on computer vision. 2019:

8420-8429.

[52] 赵宗扬,康杰虎,吴斌,等.基于 FRL-Net 的高鲁棒性多尺度小样本轨道入侵异物检测

方法研究[J].仪器仪表学报,2024,45(01):239-249.

[53] 付瑞玲,曹桂州,张洋洋,等.小样本元学习网络在海上船舶识别中的应用[J].电讯技

术,2024,64(08):1187-1194.

[54] Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need[J]. Advances in neural

information processing systems, 2017, 30.

[55] Yu Y, Zhang Y, Cheng Z, et al. MCA: Multidimensional collaborative attention in deep

convolutional neural networks for image recognition[J]. Engineering Applications of

Artificial Intelligence, 2023, 126: 107079.

[56] Yang G, Lei J, Zhu Z, et al. AFPN: Asymptotic feature pyramid network for object

detection[C]//2023 IEEE International Conference on Systems, Man, and Cybernetics

(SMC). IEEE, 2023: 2184-2189.

[57] Chollet F. Xception: Deep learning with depthwise separable convolutions

[C]//Proceedings of the IEEE conference on computer vision and pattern recognition.

2017: 1251-1258.

[58] Selvaraju R R, Cogswell M, Das A, et al. Grad-CAM: visual explanations from deep

networks via gradient-based localization[J]. International journal of computer vision,

2020, 128: 336-359.

[59] Radford A, Kim J W, Hallacy C, et al. Learning transferable visual models from natural language supervision[C]//International conference on machine learning. PmLR, 2021: 8748-8763.

[60] Zhang H, Zhang S. Shape-iou: More accurate metric considering bounding box shape and

scale[J]. arxiv preprint arxiv:2312.17663, 2023.

[61] Zhu X, Su W, Lu L, et al. Deformable detr: Deformable transformers for end-to-end

object detection[J]. arxiv preprint arxiv:2010.04159, 2020.

[62] Shangguan Z, Rostami M. Improved region proposal network for enhanced few-shot

object detection[J]. Neural Networks, 2024, 180: 106699.

[63] Han J, Ren Y Ding J,et al. Few-shot object detection viavariational feature aggregation

[C]//Proceedings of the AAAIConference on Artificial Intelligence.2023,37(1):755-763.