带式输送机作为煤矿主运输设备，运煤过程中夹杂的矸石和锚杆、铁丝等尖锐异物容易划伤皮带，损伤逐渐累积后造成皮带撕裂，不仅影响煤矿安全高效生产，还会给煤炭企业带来巨大的经济损失。若能在事故发生前，及时分拣出煤流异物，在空载或小煤流量情况下跟踪检查带面早期损伤状况，保证带式输送机连续平稳运行，对煤矿高效生产及智能化发展具有重大意义。因此，本文以带式输送机带面早期损伤与煤流异物为研究对象，利用深度学习和机器视觉进行目标检测研究。

针对低照度尘雾下带式输送机带面早期损伤小目标检测精度差的问题，提出一种考虑带面早期损伤的非遮挡煤流异物智能检测方法。首先使用限制对比度自适应直方图均衡化实现低照度尘雾图像增强；其次改进YOLOv5s检测层结构，增加浅层检测层以丰富小目标信息，同时删除大目标检测层及其相应模块以降低模型复杂度；接着使用Partial Conv与Res2Net对主干网络中瓶颈结构进行改进以提升模型特征提取能力；最后使用高效通道注意力机制提升通道特征利用率。改进后模型的带面早期损伤检测精度AP_0.5为91.30%，提升了10.00%，模型总体检测精度mAP_0.5高达98.30%，参数量为4.58×10⁶，计算量为14.80G，模型大小为10.10MB。结果表明考虑带面早期损伤的非遮挡煤流异物检测方法能在低照度尘雾环境实现带面早期损伤及煤流异物高精度检测。

考虑实际煤矿运煤过程中煤与煤流异物是互相堆叠、相互遮挡状态，提出一种多层遮挡煤流异物快速智能检测方法。首先使用软非极大值抑制对YOLOv5s后端处理算法进行优化以降低密集因素影响；其次对YOLOv5s使用SimOTA标签分配策略降低密集遮挡场景下歧义样本的影响；接着利用Slide Loss挖掘难样本，并使用Inner-SIoU优化边界框回归损失函数，增加遮挡目标和小目标检测精度；最后使用Group-Taylor剪枝方法对模型进行剪枝获得参数量仅0.42×10⁶、计算量1G、模型大小1.20MB的mAP_0.5为91.30%的高精度检测模型，在GTX 1050Ti、GT 1030及CPU设备上检测速度高达66.31、41.90、33.03FPS，远超YOLOv5s。实验结果表明，多层遮挡煤流异物智能检测方法能大幅度降低模型复杂度，在中低端GPU和CPU设备上满足实时处理要求，且能保证模型具有较高检测精度。

同时，针对煤流异物遮挡及带面早期损伤小目标检测问题，考虑带面损伤主要是由煤流异物造成的，两者在一定程度上属于因果关系，煤流异物检测优先度高于皮带早期损伤，故以多层遮挡煤流异物快速智能检测方法为主，改进Inner-SIoU边界框损失函数为Inner-MPDIoU，同样对模型使用Group-Taylor剪枝，最终获得参数量仅0.49×10⁶、计算量1.20G、模型大小1.30MB的mAP_0.5为90.30%的高精度检测模型，在GPU、CPU上检测速度能达到61.23、40.23、29.21FPS，远超YOLOv5s模型。实验结果表明，考虑带面早期损伤的多层遮挡煤流异物快速智能检测方法能以较低复杂度、较快速度实现带面早期损伤及多层遮挡煤流异物的高精度检测。

最后，通过PyQt5设计了带式输送机带面早期损伤及煤流异物检测系统，并将带面早期损伤及煤流异物检测模型及程序移植到嵌入式设备，在Raspberry Pi上完成带面早期损伤及煤流异物检测模型验证与系统功能验证。实验结果表明，本文所提模型在Raspberry Pi上检测效果较好，使用Group-Taylor剪枝的多层遮挡煤流异物快速智能检测模型在遮挡煤流异物图像上推理耗时1.60s，远低于YOLOv5s模型的17.31s，使用Group-Taylor剪枝的考虑带面早期损伤的多层遮挡煤流异物快速智能检测模型推理耗时为1.50s，远低于YOLOv5s模型的16.97s。

As the main transportation equipment in coal mine, the belt conveyor is easy to scratch the belt due to the sharp foreign objects such as gangue, anchor rod and wire in the process of coal transportation. The damage gradually accumulates and causes the belt to tear, which not only affects the safe and efficient production of coal mine, but also brings huge economic losses to coal enterprises. It is of great significance for the efficient production and intelligent development of coal mine to timely sort out the foreign objects of coal flow before the accident, track and check the early damage of the belt surface under the condition of no-load or small coal flow, and ensure the continuous and stable operation of the belt conveyor. Therefore, this thesis takes the early damage of belt surface and coal flow foreign objects as the research object, and uses deep learning and machine vision to carry out target detection research.

Aiming at the problem of poor detection effect of small target of early damage on belt surface under low illumination and dust fog condition, an intelligent detection method of non-occluded coal flow foreign objects considering belt surface damage is proposed. Firstly, the low illumination and dust fog image enhancement is realized by using the contrast limited adaptive histogram equalization. Secondly, the YOLOv5s detection layer structure is improved, the shallow detection layer structure is added to enrich the small target information, the large target detection layer and its corresponding modules are deleted to reduce the complexity of the model at the same time. Then, Partial Conv and Res2Net are used to improve the bottleneck structure in the backbone network to improve the feature extraction ability of the model. Finally, the efficient channel attention mechanism is used to improve the channel feature utilization. The early damage of belt surface detection accuracy AP_0.5 of the improved model is 91.30%, which is increased by 10.00%. The overall detection accuracy mAP_0.5 of the model is as high as 98.30%, the parameter quantity is 4.58×10⁶, the calculation amount is 14.80G, and the model size is 10.10MB. The results show that the non-occluded coal flow foreign objects detection method considering the surface damage can realize high-accuracy detection of the surface damage and the coal flow foreign objects in the low-illumination dust and fog environment.

Considering that the coal and the foreign objects are stacked and occluded with each other in the process of coal transportation in the actual coal mine, a fast intelligent detection method of multi-layer occluded coal flow foreign objects is proposed. Firstly, Soft Non-Maximum Suppression is used to optimize the YOLOv5s back-end processing algorithm to reduce the influence of dense factor. Secondly, the SimOTA tag allocation strategy is used to reduce the influence of ambiguous samples in dense occlusion scenes for YOLOv5s. Then, Slide Loss is used to excavate difficult samples, and Inner-SIoU is used to optimize the bounding box regression loss function to increase the detection accuracy of occlusion target and small target. Finally, the Group-Taylor pruning method is used to prune the model to obtain a high-accuracy detection model with a parameter number of only 0.42×10⁶, a calculation amount of 1G, and a mAP_0.5 of 91.30% with a model size of 1.20MB. The detection speed on GTX 1050Ti, GT 1030 and CPU devices is as high as 66.31, 41.90, 33.03FPS, far exceeding YOLOv5s. The experimental results show that the intelligent detection method of multi-layer occluded coal flow foreign objects can greatly reduce the complexity of the model, meet the real-time processing requirement on the low-medium end GPU and CPU devices, and ensure that the model has high detection accuracy.

At the same time, aiming at the detection problems of coal flow foreign objects occlusion target and small target of early damage of belt surface, it is considered that the belt damage is mainly caused by coal flow foreign objects, and the two belong to causality to a certain extent. The priority of coal flow foreign objects are higher than the early damage of belt, so the multi-layer occlusion coal flow foreign objects fast intelligent detection method is the main method, the Inner-SIoU bounding box loss function is changed to Inner-MPDIoU in this method, and Group-Taylor pruning is also used for the model. Finally, a high-accuracy detection model with a parameter of only 0.49×10⁶, a calculation amount of 1.20G, a model size of 1.30 MB, and a mAP_0.5 of 90.30 % is obtained. The detection speed on GPU and CPU can reach 61.23, 40.23, and 29.21FPS, far exceeding the YOLOv5s model. The experimental results show that the fast intelligent detection method of multi-layer occluded coal flow foreign objects considering the early damage of conveyor belt surface can realize the high-accuracy detection of the early damage of belt surface and the multi-layer occluded coal flow foreign objects with low complexity and fast speed.

Finally, the software platform of belt surface damage and coal flow foreign objects detection system is designed by PyQt5, and the early damage of belt surface and coal flow foreign objects detection model and program are transplanted to embedded equipment. The early damage of belt surface and coal flow foreign objects detection model verification and system function verification are completed on Raspberry Pi. The experimental results show that the proposed models have better detection effect on Raspberry Pi. The reasoning time of the multi-layer occlusion coal flow foreign objects fast intelligent detection model using Group-Taylor pruning is 1.60s on the occlusion coal flow foreign objects image, which is much lower than 17.31s of the YOLOv5s model. The reasoning time of the multi-layer occlusion coal flow foreign objects fast intelligent detection model considering early damage of belt surface with Group-Taylor pruning is 1.50s, which is much lower than 16.97s of the YOLOv5s model.

[1]国家统计局. 中华人民共和国2023年国民经济和社会发展统计公报[N]. 人民日报, 2024-03-01(010).

[2]刘峰, 曹文君, 张建明, 等. 我国煤炭工业科技创新进展及“十四五”发展方向[J]. 煤炭学报, 2021, 46(01): 1-15.

[3]陈晓晶. 井工煤矿运输系统智能化技术现状及发展趋势[J]. 工矿自动化, 2022, 48(06): 6-14+35.

[4]王国法. 煤矿智能化最新技术进展与问题探讨[J]. 煤炭科学技术, 2022, 50(01): 1-27.

[5]王国法, 庞义辉, 刘峰, 等. 智能化煤矿分类、分级评价指标体系[J]. 煤炭科学技术, 2020, 48(03): 1-13.

[6]王国法, 刘峰, 庞义辉, 等. 煤矿智能化-煤炭工业高质量发展的核心技术支撑[J]. 煤炭学报, 2019, 44(02): 349-357.

[7]毛清华, 毛金根, 马宏伟, 等. 矿用带式输送机智能监测系统研究[J]. 工矿自动化, 2020, 46(06): 48-52+58.

[8]段雍. 基于图像的煤矸识别和定位方法研究与实现[D]. 西安: 西安科技大学, 2020.

[9]高新波, 莫梦竟成, 汪海涛, 等. 小目标检测研究进展[J]. 数据采集与处理, 2021, 36(03): 391-417.

[10]孙海伦, 宋纯贺, 于诗矛, 等. 边缘计算对工业互联网产业发展的重要意义及研究现状[J]. 自动化博览, 2021, 38(02): 17-21.

[11]马宏伟, 杨文娟, 张旭辉. 基于红外热像的带式输送机监测与预警系统[J]. 激光与红外, 2017, 47(04): 448-452.

[12]高睿. 煤矿井下带式输送机输送带纵向撕裂保护系统研究[J]. 内蒙古煤炭经济, 2020(15): 29-30.

[13]李瑶, 王义涵. 带式输送机煤流量自适应检测方法[J]. 工矿自动化, 2020, 46(06): 98-102.

[14]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.

[15]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.

[16]吕志强. 复杂环境下煤矿皮带运输异物图像识别研究[D]. 徐州: 中国矿业大学, 2021.

[17]胡璟皓, 高妍, 张红娟, 等. 基于深度学习的带式输送机非煤异物识别方法[J]. 工矿自动化, 2021, 47(6): 57-62+90.

[18]Wang Y, Wang Y, Dang L. Video detection of foreign objects on the surface of belt conveyor underground coal mine based on improved SSD[J]. Journal of Ambient Intelligence and Humanized Computing, 2020: 1-10.

[19]李曼, 何仙利, 杨茂林, 等. 外在水分对煤矸图像特征影响的试验研究[J]. 煤炭科学技术, 2022,50(10): 179-187.

[20]Li M, He X L, Duan Y, et al. Experimental study on the influence of external factors on image features of coal and gangue[J]. International Journal of Coal Preparation and Utilization, 2022, 42(9): 2770-2787.

[21]任志玲, 朱彦存. 改进CenterNet算法的煤矿皮带运输异物识别研究[J]. 控制工程, 2023, 30(04):703-711.

[22]郝帅, 张旭, 马旭, 等. 基于CBAM-YOLOv5的煤矿输送带异物检测[J]. 煤炭学报, 2022, 47(11): 4147-4156.

[23]程德强, 徐进洋, 寇旗旗, 等. 融合残差信息轻量级网络的运煤皮带异物分类[J]. 煤炭学报, 2022, 47(3): 1361-1369.

[24]曹现刚, 郝朋英, 王鹏, 等. 多因素光照条件下高质量煤矸图像获取方法研究[J]. 煤炭科学技术, 2023, 51(01): 455-463.

[25]李博, 王学文, 庞尚钟, 等. 煤与矸石图像特征分析及试验研究[J]. 煤炭科学技术, 2022, 50(8): 236-246.

[26]郜亚松, 张步勤, 郎利影. 基于深度学习的煤矸石识别技术与实现[J]. 煤炭科学技术, 2021, 49(12): 202-208.

[27]徐志强, 吕子奇, 王卫东, 等. 煤矸智能分选的机器视觉识别方法与优化[J]. 煤炭学报, 2020, 45(6): 2207-2216.

[28]郭永存, 王希, 何磊， 等. 基于TW-RN优化CNN的煤矸识别方法研究[J]. 煤炭科学技术, 2022, 50(1): 228-236.

[29]毛清华, 李世坤, 胡鑫, 等. 基于改进YOLOv7的煤矿带式输送机异物识别[J]. 工矿自动化, 2022, 48(12): 26-32.

[30]陈卓. 基于深度学习的矿用带式输送机异物检测方法研究[D]. 徐州: 中国矿业大学, 2023.

[31]江峰. 基于深度学习的带式输送机煤流异物监测系统研究[D]. 徐州: 中国矿业大学, 2023.

[32]郭启皇, 乔铁柱. 基于Otsu算法的输送带撕裂视觉检测系统研究[J]. 煤炭技术, 2017, 36(4): 279-282.

[33]刘晓阳, 刘晶, 张向阳, 等. 基于二维Gabor滤波器的胶带撕裂检测[J]. 工矿自动化, 2021, 47(4): 103-107.

[34]Zhang M C, Shi H, Zhang Y, et al. Deep learning-based damage detection of mining conveyor belt[J]. Measurement, 2021, 175(99): 109-130.

[35]Zhang M C, Zhang Y, Zhou M S, et al. Application of lightweight convolutional neural network for damage detection of conveyor belt[J]. Applied Sciences, 2021, 11(16): 7282.

[36]Qu D R, Qiao T Z, Pang Y S, et al. Research on ADCN method for damage detection of mining conveyor belt[J]. IEEE Sensors Journal, 2021, 6(21): 8662-8669.

[37]Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[J]. IEEE Computer Society, 2014: 580-587.

[38]Girshick R. Fast r-cnn[C]//Proceedings of the IEEE International Conference on Computer Vision. 2015: 1440-1448.

[39]He K M, Gkioxari G, Dollár P, et al. Mask r-cnn[C]//Proceedings of the IEEE International Conference on Computer Vision. 2017: 2961-2969.

[40]Jocher, Glenn, Chaurasia, et al. Ultralytics[CP]. (2022-9-12)[2024-4-8]. https://githu-b.com/ultralytics/ultralytics.

[41]Lim J S, Astrid M, Yoon H J, et al. Small object detection using context and attention[C]//2021 international Conference on Artificial intelligence in information and Communication (ICAIIC). 2021: 181-186.

[42]张馨月, 降爱莲. 融合特征增强和自注意力的SSD小目标检测算法[J]. 计算机工程与应用, 2022, 58(05): 247-255.

[43]Ying X ,Wang Q, Li X W, et al. Multi-attention object detection model in remote sensing images bascd on multi-scale[J]. IEEE Access, 2019(07): 94508-94519.

[44]Luo Y H, Cao X, Zhang J T, et al. CE-FPN: Enhancing channel information for object detection[J]. Multimedia Tools and Applications, 2022, 81(21): 30685-30704.

[45]Kim S W, Kook H K, Sun J Y, et al. Parallel feature pyramid network for object detection[C]//Proceedings of the European conference on computer vision (ECCV). 2018: 234-250.

[46]Zhao B J, Zhao B Y, Tang L B, et al. Multi-scale object detection by top-down and bottom-upfeature pyramid network[J]. Journal of Systems Engineering and Electronics, 2019, 30(1): 1-12.

[47]Tan M X, Pang R M, Le Q V. EfficientDet: Scalable and efficient object detection[J].Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2020, 10778-10787.

[48]张侣, 周博文, 吴亮红. 基于改进卷积注意力模块与残差结构的SSD网络[J]. 计算机科学, 2022, 49(03): 211-217.

[49]李子豪, 王正平, 贺云涛. 基于自适应协同注意力机制的航拍密集小目标检测算法[J]. 航空学报, 2023, 44(13): 244-254.

[50]韩强. 面向小目标检测的改进YOLOv8算法研究[D]. 长春: 吉林大学, 2024.

[51]Zhang S F, Wen L Y, Bian X, et al. Occlusion-aware R-CNN: Detecting pedestrians in a crowd[C]//Proceedings of the European Conference on Computer Vision (ECCV). 2018: 637-653.

[52]Zhou M, Zou Z X, Shi Z W, et al. Local attention networks for occluded airplane detection in remote sensing images[J]. IEEE Geoscience and Remote Sensing Letters, 2019, 17(3): 381-385.

[53]安珊, 林树宽, 乔建忠, 等. 基于生成对抗网络学习被遮挡特征的目标检测方法[J]. 控制与决策, 2021, 36(05): 1199-1205.

[54]阳珊, 王建, 胡莉, 等. 改进RetinaNet的遮挡目标检测算法研究[J]. 计算机工程与应用, 2022, 58(11): 209-214.

[55]Dong X P, Shen J B, Yu D J, et al. Occlusion-aware real-time object tracking[J]. IEEE Transactions on Multimedia, 2016, 19(4): 763-771.

[56]Ren Y, Zhu C R, Xiao S P. Deformable Faster R-CNN with aggregating multi-layer features for partially occluded object detection in optical remote sensing images[J]. Remote Sensing, 2018, 10(9): 1470.

[57]Wang X L, Xiao T, Jiang Y N, et al. Repulsion loss: Detecting pedestrians in a crowd[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recogni-tion. 2018: 7774-7783.

[58]Li D Y, Wang G F, Zhang Y, et al. Coal gangue detection and recognition algorithm based on deformable convolution YOLOv3[J]. IET Image Processing, 2021, 16(1): 134-144.

[59]Iandola F N, Han S, Moskewicz M W, et al. SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and < 0.5MB model size[EB/OL]. (2016-11-4)[2024-4-8]. htt-ps://arxiv.org/abs/1602.07360.

[60]Howard A, Sandler M, Chu G, et al. Searching for mobilenetv3[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. 2019: 1314-1324.

[61]Ma N N, Zhang X Y, Zheng H T, et al. Shufflenet v2: Practical guidelines for efficient cnn architecture design[C]//Proceedings of the European Conference on Computer Vision (ECCV). 2018: 116-131.

[62]Zhuang Z W, Tan M K, Zhuang B H, et al. Discrimination-aware channel pruning for deep neural networks[EB/OL]. (2019-1-14)[2024-4-8]. https://arxiv.org/abs/1810.11809.

[63]Xu X W, Lu Q, Yang L, et al. Quantization of fully convolutional networks for accurate biomedical image segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018: 8300-8308.

[64]Pham H, Guan M Y, Zoph B, et al. Efficient neural architecture search via parameters sharing[C]//International Conference on Machine Learning. PMLR, 2018: 4095-4104.

[65]Wang L, Yoon K J. Knowledge distillation and student-teacher learning for visual intelligence: A review and new outlooks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 44(6): 3048-3068.

[66]Tang Y H, You S, Xu C, et al. Reborn filters: Pruning convolutional neural networks with limited data[C]//Proceedings of the AAAI Conference on Artificial Intelligence. 2020, 34(04): 5972-5980.

[67]韦越, 陈世超, 朱凤华, 等. 基于稀疏正则化的卷积神经网络模型剪枝方法[J]. 计算机工程, 2021, 47(10): 61-66.

[68]Nagel M, Baalen M V, Blankevoort T, et al. Data-free quantization through weight equalization and bias correction[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. 2019: 1325-1334.

[69]Zhu F, Gong R H, Yu F W, et al. Towards unified int8 training for convolutional neural network[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2020: 1969-1979.

[70]Fan A, Stock P, Graham B. Training with Quantizatian Noise for Extreme Model Compression[J]. 2020, 13(9): 58-66.

[71]郭俊伦, 彭书华, 李俊杰. 基于特征重建的知识蒸馏方法[J]. 现代计算机, 2020(29): 43-47+73.

[72]辛丽. 基于低秩稀疏分解的深度卷积神经网络模型压缩方法研究[D]. 青岛: 山东科技大学, 2021.

[73]徐晓, 蒋志翔, 张杨. 基于通道剪枝和量化的卷积神经网络压缩方法[J]. 计算机工程与设计, 2021, 42(10): 2860-2866.

[74]Zhang Y F, Ding L, Sharma G. Hazerd: An outdoor scene dataset and benchmark for single image dehazing[C]//2017 IEEE International Conference on Image Processing (ICIP). IEEE, 2017: 3205-3209.

[75]Wang J W, Yang W, Guo H W, et al. Tiny object detection in aerial images[C]//2020 25th International Conference on Pattern Recognition (ICPR). IEEE, 2021: 3791-3798.

[76]Wang Q L, Wu B G, Zhu P F, et al. ECA-Net: efficient channel attention for deep convolutional neural networks[J]. Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2020, 11531-11539.

[77]单鹏飞, 李晨炜, 来兴平, 等. 模拟暗湿工况下煤矸混合体态势热敏图像精准辨识试验[J]. 煤炭学报, 2023, 48(S2): 802-812.

[78]Gao S H, Cheng M M, Zhao K, et al. Res2Net: A new multi-scale backbone architecture[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021,43(2): 652-662.

[79]Chen J R, Kao S H, He H, et al. Run, Don't walk: Chasing higher FLOPS for faster neural networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2023: 12021-12031.

[80]孙方伟, 李承阳, 谢永强, 等.深度学习应用于遮挡目标检测算法综述[J]. 计算机科学与探索, 2022, 16(06): 1243-1259.

[81]张胜虎, 马惠敏. 遮挡对于目标检测的影响分析[J]. 图学学报, 2020, 41(06): 891-896.

[82]Bodla N, Singh B, Chellappa R, et al. Soft-NMS improving object detection with one line of code[C]//Proceedings of the IEEE International Conference on Computer Vision. 2017: 5561-5569.

[83]Ge Z, Liu S T, Li Z M, et al. Ota: Optimal transport assignment for object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2021: 303-312.

[84]Ge Z, Liu S T, Wang F, et al. Yolox: Exceeding yolo series in 2021[EB/OL]. (2021-8-6)[2024-4-8]. https://arxiv.org/abs/2107.08430.

[85]周浩然. 基于神经网络的轻量化实时目标检测算法研究[D]. 成都: 西南交通大学, 2024.

[86]Yu Z P, Huang H B, Chen W J, et al. Yolo-facev2: A scale and occlusion aware face detector[EB/OL]. (2022-8-4)[2024-4-8]. https://arxiv.org/abs/2208.02019.

[87]Zhang H, Xu C, Zhang S J. Inner-iou: More effective intersection over union losswith auxiliary bounding box[EB/OL]. (2023-11-14)[2024-4-8]. https://arxiv.org/abs/2311.02877.

[88]Molchanov P, Mallya A, Tyree S, et al. Importance estimation for neural network pruning[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2019: 11264-11272.

[89]Fang G F, Ma X Y, Song M L, et al. Torch-Pruning[CP]. (2019-12-15)[2024-4-8]. https://github.com/VainF/Torch-Pruning.

[90]Lee J, Park S, Mo S, et al. Layer-adaptive sparsity for the magnitude-based pruning[EB/OL]. (2021-5-9)[2024-4-8]. https://arxiv.org/abs/2010.07611.

[91]Li H, Kadav A, Durdanovic I, et al. Pruning filters for efficient convnets[EB/OL]. (2017-3-10)[2024-4-8]. https://arxiv.org/abs/1608.08710.

[92]Yann L C, John S, Sara A. Optimal brain damage[J]. Advances in neural information processing systems, 1989, 2: 599-605.

[93]Ma S L, Xu Y. Mpdiou: a loss for efficient and accurate bounding box regression[EB/OL]. (2023-7-14)[2024-4-8]. https://arxiv.org/abs/2307.07662.