基于深度学习的检测方法已经广泛应用在输电线路的绝缘子自爆缺陷巡检中，但在无人机航拍场景中自爆缺陷的小目标检测依旧面临着检测精度低以及大模型部署难度高等问题。并且现有的知识蒸馏体系在绝缘子自爆缺陷检测领域的应用较少，存在着不合理的样本匹配机制和温度选定机制严重限制绝缘子自爆缺陷小目标检测中大模型和小模型之间知识传递的问题。针对上述问题，本文做出的主要工作如下：

（1）针对绝缘子自爆缺陷小目标检测精度不佳和检测网络中冗余的通道导致模型成本增加的问题，本文提出了一种用于绝缘子自爆缺陷小目标检测的轻量级自适应YOLO（Lightweight Adaptive You Only Look Once，LA-YOLO）方法。首先，具有小目标检测层的双向自适应特征金字塔网络利用自适应融合策略强化了小尺寸的绝缘子自爆缺陷的特征传递，以提高自爆缺陷小目标检测的平均精度。其次，基于部分卷积的轻量级梯度分流模块Faster-C2f在骨干网络中降低冗余的特征提取，以减少参数量和浮点运算量（Floating-point Operations，FLOPs）。最后，任务和结构双解耦头在回归分支中引入空间感知卷积提高空间信息提取能力，在分类分支中减少冗余的卷积操作。实验结果表明，LA-YOLO在绝缘子自爆缺陷小目标检测上以更加轻量化的模型结构和更快的检测速度取得了优于现有方法（如InsuDet、ID-YOLO、FINet和BiFusion-YOLOv3）的检测性能。

（2）针对轻量化的结构设计、模型剪枝和模型量化等方法带来的绝缘子自爆缺陷小目标检测的精度下降问题，本文提出了动态聚焦知识蒸馏（Dynamic Focused Knowledge Distillation，DFKD）方法，用以构建绝缘子自爆缺陷检测的大模型向小模型的知识迁移路径。一方面，在蒸馏训练的不同时期，重要样本聚焦机制主动学习不同阶段学生模型的样本特点，为蒸馏学习匹配合适的蒸馏样本。并且还使用解耦样本质量的权重因子，将学生模型的学习精力集中在高质量的困难样本上。另一方面，随着由易到难的蒸馏过程，温度动态学习机制控制蒸馏样本的平滑度，为学生模型提供满足当前阶段学习难度的软标签。这个机制将引导学生模型学习教师模型的高阶特征分布，以提高模型的泛化能力。实验结果表明，DFKD在绝缘子自爆缺陷小目标检测的模型训练上，优于现有的知识蒸馏方法。本文也探索了DFKD与LA-YOLO、模型剪枝（Model Pruning，MP）的联合训练框架，以构建DFKD-LA-YOLO和DFKD-MP-YOLO。实验结果表明，DFKD-LA-YOLO在不改变LA-YOLO的网络结构的前提下，其在绝缘子自爆缺陷小目标上检测精度得到了再次提升。与现有的方法相比，DFKD-MP-YOLO兼具MP和DFKD的优点，利用更小的模型成本和更快的检测速度取得了更高的检测精度，可以依据不同设备的需求建立不同体量的模型。

本文所提方法在绝缘子自爆缺陷的小目标检测和模型轻量化部署上做出了贡献，也为电力系统的无人机巡线任务提供了可行的深度学习框架。未来，绝缘子缺陷数据集的多样化探索和模型剪枝在绝缘子自爆缺陷检测的应用研究将进一步提高无人机智能化巡检的效率和精确度。

The detection method based on deep learning has been widely applied in the inspection of insulator self-explosion defects (ISDs) in transmission lines. However, the small object detection of ISDs in unmanned aerial vehicle (UAV) photography scenes still faces problems such as low detection accuracy and high difficulty in deploying large models. Moreover, the existing knowledge distillation system has limited application in the field of ISD detection, and there are unreasonable sample matching mechanisms and temperature selection mechanisms that seriously limit the knowledge transfer between large and small models in the small object detection of ISDs. In response to the above issues, this article mainly focuses on the following work.

(1) This paper proposes a lightweight adaptive you only look once (LA-YOLO) method for detecting small objects of ISDs, which solves the problem of poor accuracy in detecting small objects and increased model costs due to redundant channels in the detection network. Firstly, the bidirectional adaptive feature pyramid network with a small object detection layer utilizes an adaptive fusion strategy to enhance the feature transfer of small-sized ISDs, in order to improve the average accuracy of ISDs small object detection. Secondly, the Faster-C2f module (lightweight gradient diversion) based on the partial convolution reduces redundant feature extraction in the backbone network to reduce parameters and floating-point operations (FLOPs). Finally, the task and structure decoupling heads introduce spatial aware convolution in the regression branch to improve spatial information extraction capability, and reduce redundant convolution operations in the classification branch. The experimental results show that the LA-YOLO achieves better detection performance than the existing methods (e.g., InsuDet, ID-YOLO, FINet, and BiFusion-YOLOv3) in detecting small objects of ISDs, while having lower parameters and FLOPs as well as faster detection speed.

(2) Aiming at the problem of decreased accuracy in the small object detection of ISDs caused by lightweight structural design, model pruning (MP), and model quantization methods, this paper proposes the dynamic focused knowledge distillation (DFKD) method to construct a knowledge transfer path from large models to small models for ISD detection. On the one hand, at different stages of distillation training, the important sample focusing mechanism actively learns the sample characteristics of student models at different stages, matching appropriate distillation samples for distillation learning. And it also uses weight factors that decouple sample quality to focus the learning energy of the student model on high-quality difficult samples. On the other hand, as the distillation process progresses from easy to difficult, the temperature dynamic learning mechanism controls the smoothness of the distillation samples, providing the student model with soft labels that meet the current learning difficulty. This mechanism will guide the student model to learn the high-order feature distribution of the teacher model, in order to improve the model’s generalization ability. The experimental results show that the DFKD outperforms existing knowledge distillation methods in training small models for the small object detection of ISDs. This paper also explores the joint training framework of DFKD with LA-YOLO and MP to construct DFKD-LA-YOLO and DFKD-MP-YOLO. The experimental results show that the detection accuracy of DFKD-LA-YOLO on small objects of ISDs is improved again without changing the network structure of LA-YOLO. Compared with the existing methods, DFKD-MP-YOLO combines the advantages of the MP and the DFKD, achieving higher detection accuracy with lower model cost and faster detection speed. It can establish models of different volumes according to the needs of different devices.

The method proposed in this article has made contributions to the small target detection and lightweight deployment of ISDs, and also provides a feasible deep learning framework for UAV line inspection tasks in power systems. In the future, the diversified exploration of insulator defect datasets and the application research of model pruning in ISD detection will further improve the efficiency and accuracy of UAV intelligent inspection.

[1] 徐聪. 基于深度学习的航拍图像中绝缘子缺陷检测的研究[D]. 华北电力大学, 2021.

[2] 殷振晓. 基于深度学习的电力巡检图像中绝缘子识别与缺陷检测研究[D]. 华北电力大学(北京), 2022.

[3] 王建烨, 续欣莹, 谢刚等. 改进SSD模型的绝缘子自爆故障检测[J]. 现代电子技术, 2022, 45(14): 115-121.

[4] Fu Q, Liu J, Zhang X, et al. A small-sized defect detection method for overhead transmission lines based on convolutional neural networks[J]. IEEE Transactions on Instrumentation and Measurement, 2023, 72: 1-12.

[5] Wu Q, An J. An active contour model based on texture distribution for extracting inhomogeneous insulators from aerial images[J]. IEEE Transactions on Geoscience and Remote Sensing, 2014, 52(6): 3613-3626.

[6] Wang W, Wang Y, Han J, et al. Recognition and drop-off detection of insulator based on aerial image[C]//Proceedings of the 2016 9th international symposium on computational intelligence and design, 2016, 1: 162-167.

[7] Zhai Y, Chen R, Yang Q, et al. Insulator fault detection based on spatial morphological features of aerial images[J]. IEEE Access, 2018, 6: 35316-35326.

[8] 尚方, 朱明泽, 刘生等. 基于颜色和距离的绝缘子航拍图像故障检测方法[J]. 黑龙江电力, 2020, 42(02): 100-103.

[9] 崔克彬. 基于图像的绝缘子缺陷检测中若干关键技术研究[D]. 华北电力大学(北京), 2016.

[10] 方挺, 韩家明. 航拍图像中绝缘子缺陷的检测与定位[J]. 计算机科学, 2016, 43(S1): 222-225.

[11] 赵振兵, 徐磊, 戚银城等. 基于Hough检测和C-V模型的航拍绝缘子自动协同分割方法[J]. 仪器仪表学报, 2016, 37(02): 395-403.

[12] 苗向鹏. 基于图像处理的接触网绝缘子识别与破损检测[D]. 西南交通大学, 2017.

[13] 商俊平, 李储欣, 陈亮. 基于视觉的绝缘子定位与自爆缺陷检测[J]. 电子测量与仪器学报, 2017, 31(06): 844-849.

[14] 韩正新, 乔耀华, 孙阳等. 基于图像识别的无人机输电线路绝缘子故障检测方法研究[J]. 现代电子技术, 2017, 40(22): 179-181+186.

[15] 廖圣龙. 航拍输电线图像中部件检测关键技术研究[D]. 大连海事大学, 2017.

[16] 王梦. 基于绝缘子图像的缺陷检测方法研究[D]. 华中科技大学, 2019.

[17] 杨蔺. 航拍绝缘子图像识别及故障检测方法设计与研究[D]. 重庆邮电大学, 2020.

[18] 仝卫国, 于浩进, 张秋实. 基于神经网络的绝缘子故障诊断[J]. 计算机仿真, 2013, 30(09): 310-313.

[19] Liao S, An J. A robust insulator detection algorithm based on local features and spatial orders for aerial images[J]. IEEE Geoscience and Remote Sensing Letters, 2017, 12(5): 963-967.

[20] 谢亮凯. 基于机器视觉的铁路接触网鸟巢与绝缘子识别检测方法研究[D]. 华东交通大学, 2017.

[21] Li B, Wu D, Cong Y, et al. A method of insulator detection from video sequence[C]//Proceedings of the 2012 4th International Symposium on Information Science and Engineering, 2012: 14-16.

[22] Xian W, Zhang Y. Insulator identification from aerial images using support vector machine with background suppression[C]//Proceedings of the 2016 International Conference on Unmanned Aircraft Systems, 2016.

[23] Reedy M J B, Chandra B K, Mohanta D K. A DOST based approach for the condition monitoring of 11kV distribution line insulators[J]. IEEE Transactions on Dielectrics and Electrical Insulation, 2011, 18(2): 588-95.

[24] 伍洋. 基于机器学习的航拍图像绝缘子识别方法研究[D]. 华北电力大学, 2016.

[25] Tao X, Zhang D, Wang Z, et al. Detection of power line insulator defects using aerial images analyzed with convolutional neural networks[J]. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2018, 50(4): 1486-1498.

[26] 汤踊, 韩军, 魏文力等. 深度学习在输电线路中部件识别与缺陷检测的研究[J]. 电子测量技术, 2018, 41(6): 60⁃65.

[27] 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.

[28] He K, Gkioxari G, Dollár P, et al. Mask R-CNN[C]//Proceedings of the IEEE International Conference on Computer Vision, 2017, 2961-2969.

[29] 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.

[30] 何宁辉, 王世杰, 刘军福等. 基于深度学习的航拍图像绝缘子缺失检测方法研究[J]. 电力系统保护与控制, 2021, 49(12):9.

[31] Liang H, Zuo C, Wei W. Detection and evaluation method of transmission line defects based on deep learning[J]. IEEE Access, 2020, 8: 38448-38458.

[32] Wen Q, Luo Z, Chen R, et al. Deep learning approaches on defect detection in high resolution aerial images of insulators[J]. Sensors, 2021, 21(4): 1033.

[33] Zhao W, Xu M, Cheng X, et al. An insulator in transmission lines recognition and fault detection model based on improved Faster R-CNN[J]. IEEE Transactions on Instrumentation and Measurement, 2021, 70: 1-8.

[34] 严波, 任仰勋, 方登洲等. 基于串联Faster R-CNN的绝缘子自爆检测识别[J]. 计算机技术与发展, 2021, 31 (12): 175-179+186.

[35] Lin T Y, Dollár P, Girshick R, et al. Feature pyramid networks for object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, 2117-2125.

[36] 赵振兵, 李延旭, 戚银城等. 基于动态焦点损失函数和样本平衡方法的绝缘子缺陷检测方法[J]. 电力自动化设备, 2020, 40(10): 7.

[37] 翟永杰, 杨珂, 王乾铭等. 基于混合样本迁移学习的盘型绝缘子缺陷检测[J]. 中国电机工程学报, 2023, 43(07): 2867-2877.

[38] Dian S, Zhong X, Zhong Y. Faster R-Transformer: an efficient method for insulator detection in complex aerial environments[J]. Measurement, 2022, 199: 111238.

[39] Liu W, Anguelov D, Erhan D, et al. SSD: single shot multibox detector[C]// Proceedings of the European Conference on Computer Vision, 2016, 21-37.

[40] Tan M, Pang R, Le Q V. EfficientDet: scalable and efficient object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, 10781-10790.

[41] Zhou X, Wang D, Krähenbühl P. Objects as points[J]. arXiv:1904.07850, 2019.

[42] Tian Z, Shen C, Chen H, et al. FCOS: a simple and strong anchor-free object detector[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 44(4): 1922-1933.

[43] Redmon J, Divvvala 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.

[44] Redmon J, Farhadi A. YOLO9000: better, faster, stronger[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017.

[45] Redmon J, Farhadi A. YOLOv3: an incremental improvement[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018.

[46] Bochkovskiy A, Wang C Y, Liao H Y. YOLOv4: optimal speed and accuracy of object detection[J]. arXiv:2004.10934, 2020.

[47] Ultralytics. YOLOv5. 2020 [EB/OL]. https://github.com/ultralytics/yolov5.

[48] Ge Z, Liu S, Wang F, et al. YOLOx: exceeding yolo series in 2021[J]. arXiv:2107.08430, 2021.

[49] Li C, Li L, Jiang H, et al. YOLOv6: a single-stage object detection framework for industrial applications[J]. arXiv:2209.02976, 2022.

[50] Wang C Y, Bochkovskiy A, Liao H Y. 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.

[51] Ultralytics. YOLOv8. 2023 [EB/OL]. https://github.com/ultralytics/ultralytics.

[52] 陈文贺, 李彩林, 袁斌等. 有效的绝缘子自爆缺陷定位方法[J]. 计算机工程与设计, 2019, 40(08): 2346-2352.

[53] Miao X, Liu X, Chen J, et al. Insulator detection in aerial images for transmission line inspection using single shot multibox detector[J]. IEEE Access, 2019: 1-1.

[54] 刘业鹏. 基于深度学习的输电线路目标检测方法研究[D]. 华北电力大学, 2019.

[55] Zheng H, Sun Y, Liu X, et al. Infrared image detection of substation insulators using an improved fusion single shot multibox detector[J]. IEEE Transactions on Power Delivery, 2020, 36(6): 3351-3359.

[56] 李斌, 屈璐瑶, 朱新山等. 基于多尺度特征融合的绝缘子缺陷检测[J]. 电工技术学报, 2023, 38 (01): 60-70.

[57] Liu J, Jia R, Li W, et al. High precision detection algorithm based on improved RetinaNet for defect recognition of transmission lines[J]. Energy Reports, 2020, 6: 2430-2440.

[58] Gao Z, Yang G, Li E, et al. Novel feature fusion module-based detector for small insulator defect detection[J]. IEEE Sensors Journal, 2021, 21(15): 16807-16814.

[59] Sadykova D, Pernebayeva D, Bagheri M, et al. IN-YOLO: real-time detection of outdoor high voltage insulators using UAV imaging[J]. IEEE Transactions on Power Delivery, 2019, 35(3): 1599-1601.

[60] Liu X, Miao X, Jiang H, et al. Box-point detector: a diagnosis method for insulator faults in power lines using aerial images and convolutional neural networks[J]. IEEE Transactions on Power Delivery, 2021, 36(6): 3765-3773.

[61] Zhang X, Zhang Y, Liu J, et al. InsuDet: a fault detection method for insulators of overhead transmission lines using convolutional neural networks[J]. IEEE Transactions on Instrumentation and Measurement, 2021, 70: 1-12.

[62] 罗潇, 於锋, 彭勇. 基于深度学习的无人机电网巡检缺陷检测研究[J]. 电力系统保护与控制, 2022, 50 (10): 132-139.

[63] Yang Z, Xu Z, Wang Y. Bidirection-fusion-YOLOv3: an improved method for insulator defect detection using UAV image[J]. IEEE Transactions on Instrumentation and Measurement, 2022, 71: 1-8.

[64] 高伟, 周宸, 郭谋发. 基于改进YOLOv4及SR-GAN的绝缘子缺陷辨识研究[J]. 电机与控制学报, 2021, 25 (11): 93-104.

[65] Hao K, Chen G, Zhao L, et al. An insulator defect detection model in aerial images based on multiscale feature pyramid network[J]. IEEE Transactions on Instrumentation and Measurement, 2022, 71: 1-12.

[66] Zhang Z D, Zhang B, Lan Z C, et al. FINet: an insulator dataset and detection benchmark based on synthetic fog and improved YOLOv5[J]. IEEE Transactions on Instrumentation and Measurement, 2022, 71: 1-8.

[67] 唐靓, 余明慧, 武明虎等. 基于改进YOLOv5的绝缘子缺陷检测算法[J]. 华中师范大学学报(自然科学版), 2022, 56 (05): 771-780.

[68] 肖粲俊, 潘睿志, 李超, 等. 基于改进YOLOv5s绝缘子缺陷检测技术研究[J]. 电子测量技术, 2022, 45 (24): 137-144.

[69] 胡祥, 李英娜. 基于YOLOv5和U-Net++改进的绝缘子自爆故障检测方法[J]. 陕西理工大学学报(自然科学版), 2022, 38(3): 22-29.

[70] Song Z, Huang X, Ji C, et al. Deformable YOLOx: detection and rust warning method of transmission line connection fittings based on image processing technology[J]. IEEE Transactions on Instrumentation and Measurement, 2023, 72: 1-21.

[71] 张剑锐, 魏霞, 张林鍹等. 改进YOLO v7的绝缘子检测与定位[J]. 计算机工程与应用, 2023, 1-11.

[72] He M, Qin L, Deng X, et al. MFI-YOLO: multi-fault insulator detection based on an improved YOLOv8[J]. IEEE Transactions on Power Delivery, 2023.

[73] Deng F, Xie Z, Mao W, et al. Research on edge intelligent recognition method oriented to transmission line insulator fault detection[J]. International Journal of Electrical Power & Energy Systems, 2022, 139: 108054.

[74] 冯世凯, 陈千. 基于改进YOLOv4的轻量化玻璃绝缘子缺陷检测[J]. 计算机仿真, 2024, 41 (01): 116-122+153.

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

[76] Zhang T, Zhang Y, Xin M, et al. A light-weight network for small insulator and defect detection using UAV imaging based on improved YOLOv5[J]. Sensors, 2023, 23(11): 5249.

[77] 贾玉进, 张振程, 李浠铭等. 基于轻量化YOLOv5网络的输电线路绝缘子缺陷检测[J/OL]. 电力学报, 2024, 1-9.

[78] Han K, Wang Y, Tian Q, et al. GhostNet: more features from cheap operations[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, 1580-1589.

[79] 翟永杰, 赵晓瑜, 王璐瑶等. IDD-YOLOv7:一种用于输电线路绝缘子多缺陷的轻量化检测方法[J]. 图学学报, 2023, 1-12.

[80] Tang Y, Han K, Guo J, et al. GhostNetv2: enhance cheap operation with long-range attention[J]. Advances in Neural Information Processing Systems, 2022, 35: 9969-9982.

[81] 赵立杰,袁昌彪,黄明忠等. 基于通道剪枝的ACAM-YOLOv5s绝缘子缺陷检测 [J]. 电子测量技术, 2023, 46 (09): 108-116.

[82] Hinton G, Vinyals O, Dean J. Distilling the knowledge in a neural network[J]. arXiv:1503.02531, 2015.

[83] 杨珂浩, 于龙, 高仕斌等. 基于无监督知识蒸馏的高铁绝缘子缺陷检测算法[J]. 电气化铁道, 2023, 34 (01): 9-14.

[84] Zhao Z, Lv X, Li Y, et al. Defect Detection Method for Key Area Guided Transmission Line Components Based on Knowledge Distillation[J]. Frontiers in Energy Research, 2023, 11: 1287024.

[85] Wang T, Zhai Y, Li Y, et al. Insulator Defect Detection Based on ML-YOLOv5 Algorithm[J]. Sensors, 2023, 24 (1): 204.

[86] Liu X, Rao Z, Zhang Y, et al. UAVs Images Based Real-Time Insulator Defect Detection with Transformer Deep Learning[C]//2023 IEEE International Conference on Robotics and Biomimetics (ROBIO), 2023, 1-6.

[87] Zhou H, Song L, Chen J, et al. Rethinking the soft label of knowledge extraction: a bias-balance perspective[J]. arXiv:2102.00650, 2021.

[88] Zhao B, Cui Q, Song R, et al. Decoupled knowledge distillation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, 11953-11962.

[89] Yang Z, Zeng A, Li Z, et al. From knowledge distillation to self-knowledge distillation: a unified approach with normalized loss and customized soft labels[J]. arXiv:2303.13005, 2023.

[90] Li Z, Li X, Yang L, et al. Curriculum temperature for knowledge distillation[C]//Proceedings of the AAAI Conference on Artificial Intelligence, 2023, 37(2): 1504-1512.

[91] Yim J, Joo D, Bae J, et al. A gift from knowledge distillation: fast optimization, network minimization and transfer learning[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, 4133-4141.

[92] Ahn S, Hu S X, Damianou A, et al. Variational information distillation for knowledge transfer[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019, 9163-9171.

[93] Zheng Z, Ye R, Wang P, et al. Localization distillation for dense object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, 9407-9416.

[94] Yang L, Zhou X, Li X, et al. Bridging cross-task protocol inconsistency for distillation in dense object detection[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023, 17175-17184.

[95] Wang J, Chen Y, Zheng Z, et al. CrossKD: cross-head knowledge distillation for dense object detection[J]. arXiv:2306.11369, 2023.

[96] Yang Z, Li Z, Jiang X, et al. Focal and global knowledge distillation for detectors[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, 4643-4652.

[97] Yang Z, Li Z, Shao M, et al. Masked generative distillation[C]//Proceedings of the European Conference on Computer Vision, 2022, 53-69.

[98] Cao W, Zhang Y, Gao J, et al. PKD: general distillation framework for object detectors via pearson correlation coefficient[J]. Advances in Neural Information Processing Systems, 2022, 35: 15394-15406.

[99] Shu C, Liu Y, Gao J, et al. Channel-wise knowledge distillation for dense prediction[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021, 5311-5320.

[100] Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[J]. arXiv:1409.1556, 2014.

[101] Szegedy C, Liu W, Jia Y, et al. Going deeper with convolutions[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2015, 1-9.

[102] 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.

[103] Liu S, Qi L, Qin H, et al. Path aggregation network for instance segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, 8759-8768.

[104] Li X, Wang W, Wu L, et al. Generalized focal loss: learning qualified and distributed bounding boxes for dense object detection[J]. Advances in Neural Information Processing Systems, 2020, 33: 21002-21012.

[105] Zheng Z, Wang P, Liu W, et al. Distance-IoU loss: faster and better learning for bounding box regression[C]//Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(07), 12993-13000.

[106] Dai Y, Gieseke F, Oehmcke S, et al. Attentional feature fusion[C]//Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2021, 3560-3569.

[107] Chen J, Kao S, 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.

[108] Kirillov A, Mintun E, Ravi N, et al. Segment anything[J]. arXiv:2304.02643, 2023.

[109] Chen G, Wang H, Chen K, et al. A survey of the four pillars for small object detection: Multiscale representation, contextual information, super-resolution, and region proposal[J]. IEEE Transactions on systems, man, and cybernetics: systems, 2020, 52(2): 936-953.

[110] Cheng G, Yuan X, Yao X, et al. Towards large-scale small object detection: survey and benchmarks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(11): 13467-13488.

[111] Woo S, Debnath S, Hu R, et al. Convnext v2: co-designing and scaling convnets with masked autoencoders[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, 16133-16142.

[112] Cai H, Gan C, Han S. EfficientVit: enhanced linear attention for high-resolution low-computation visual recognition[J]. arXiv:2205.14756, 2022.

[113] Nascimento M G， Fawcett R， Prisacariu V A. DSConv： efficient convolution operator[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2019, 5148-5157.

[114] Liu S, Huang D, Wang Y. Learning spatial fusion for single-shot object detection[J]. arXiv:1911.09516, 2019.

[115] Li H, Li J, Wei H, et al. Slim-neck by GSConv: a better design paradigm of detector architectures for autonomous vehicles[J]. arXiv:2206.02424, 2022.

[116] Yang G, Lei J, Zhu Z, et al. AFPN: asymptotic feature pyramid network for object detection[J]. arXiv:2306.15988, 2023.

[117] Dai X, Chen Y, Xiao B, et al. Dynamic Head: unifying object detection heads with attentions[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, 7373-7382.

[118] Zhuang J, Qin Z, Yu H, et al. Task-specific context decoupling for object detection[J]. arXiv:2303.01047, 2023.

[119] Zhang S, Chi C, Yao Y, et al. Bridging the gap between anchor-based and anchor-free detection via adaptive training sample selection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020, 9759-9768.

[120] Creswell A, White T, Dumoulin V, et al. Generative adversarial networks: an overview[J]. IEEE signal processing magazine, 2018, 35(1): 53-65.

[121] Lee J H, Park S, Mo S, et al. Layer-adaptive sparsity for the magnitude-based pruning[J]. arXiv:2010.07611, 2020.

[122] Bolya D, Foley S, Hays J, et al. TIDE: a general toolbox for identifying object detection errors[C]//Proceedings of the European Conference on Computer Vision, Glasgow, UK, 2020, 558-573.

[123] Lv W, Xu S, Zhao Y, et al. Detrs beat yolos on real-time object detection[J]. arXiv:2304.08069, 2023.