YOLOv8原始模型采用CIoU (complete intersection over union)损失函数[17-18],该函数在处理不同形状目标的相关性时,未考虑样本难易程度的区分问题。GIoU(generalized intersection over union)损失函数在保留IoU的尺度不变性的同时,进一步考虑了目标的非重叠区域,可以更准确描述不同形状目标之间的关系。此外,GIoU损失函数可以反映目标的难易程度,弥补CIoU损失函数在相关性描述和样本区分上的不足。因此,本文采用GIoU损失函数替换CIoU损失函数。
采用不同的损失函数,对改进YOLOV8模型进行实验,结果见表3。由表3可知,DIoU(distance intersection over union)损失函数通过最小化预测框和真实框中心点距离,加快模型的收敛速度,但未能有效处理长宽比的差异,因此其mAP_0.5:0.95值略小于GIoU损失函数。CIoU损失函数在DIoU损失函数的基础上进一步引入长宽比一致性作为惩罚项,提升了回归精度,但仍未能在非重叠情况下提供足够的优化信息,限制了其性能的进一步提升。EIoU(efficient intersection over union)损失函数和SIoU(scaled intersection over union)损失函数均细化了长宽比和方向惩罚项,但对于树线接地故障的特征复杂性较高的特定任务,没有使模型的mAP值提高。Shape IoU(shape intersection over union)损失函数强调预测框与真实框形状的相似性,但在目标边界复杂的场景下,其形状优化的作用相对有限。
LUOChen, FENGYu, WUKai, et al. A lightweight outage perception model for power grids based on prompt learning from multi-source outage data [J/OL].Modern Electric Power,1-11.(2023-12-18) [2024-05-14].
ZHANGMeijin, KUAIYu, CAIZhijun,et al.Multi-criteria fusion fault line selection for improved zero-sequence current component[J]. Journal of Liaoning Technical University (Natural Science),2020,39 (1):71-77.
NINGXin, HUXinyue, ZHANGHua,et al.Characteristic analysis of tree-contact single-phase-to-ground fault in power distribution lines [J]. Proceedings of the CSU-EPSA,2023,35(7):137-143.
ZHAOShenyuan, CHENTianxiang, XUHuikai,et al Experimental research on transient impedance variation characteristics of 10 kV tree line faulted trees[J].Journal of Guangxi University (Natural Science Edition), 2023,48(2):393-406.
YANGSenlin, YANGChangqing, MEIJiming,et al.Research and application status on forest fire risk assessment and monitoring for overhead transmission lines[J].Journal of Sichuan Forestry Science and Technology,2021,42(6):126-130.
LAIQiupin, YANGJun, TANBendong,et al.An automatic recognition and defect diagnosis model of transmission line insulator based on YOLOv2 network[J].Electric Power,2019,52(7):31-39.
HAOShuai, MARuize, ZHAOXinsheng, et al.Fault detection of YOLOv3 transmission line based on convolutional block attention model[J].Power System Technology,2021,45(8):2979-2987.
ZHENGWei, YANGXiaohui, ZhongbinLYU,et al.Real-time inspection model for key components of transmission lines based on improved YOLOv4[J].Science Technology and Engineering,2021,21 (24):10393-10400.
HAOShuai, YANGLei, MAXu,et al.YOLOv5 transmission line fault detection based on attention mechanism and cross-scale feature fusion[J].Proceedings of the CSEE,2023,43(6):2319-2331.
ZHOUFei, GUODudu, WANGYang,et al.Vehicle detection algorithm based on improved YOLOv8 in traffic surveillance[J].Computer Engineering and Applications,2024,60(6):110-120.
[25]
YUZ F.Sim-YOLO:a real-time Chinese scene text detection method[C]//2023 IEEE 7th Information Technology and Mechatronics Engineering Conference.September 15-17,2023,Chongqing,China. IEEE,2023:2305-2309.
BAOCongwang, ZHUGuangyong, ZOUWang,et al.Rearing fault transfer diagnosis model based on SimAM attention mechanism[J]. Journal of Mechanical & Electrical Engineering,2024,41(5):862-869, 893.
TIANTian, CHENGZhiyou, JUWei,et al.Small sample classification of tea diseases based on SimAM-ConvNeXt-FL[J].Transactions of the Chinese Society for Agricultural Machinery,2024,55(3):275-281.
LIUXiangju, LIUYang, JIANGShexiang. DCN-YOLOv5 underwater target detection based on SimAM attention[J/OL].Journal of Chongqing Technology and Business University (Natural Science Edition),1-9.(2023-10-23)[2024-12-19].
WANGHaiyong, WANGZhiqing.Improved SSD object detection algorithm based on attention and feature fusion[J].Software,2023,44 (4):1-5.
[38]
REZATOFIGHIH, TSOIN, GWAKJ,et al. Generalized intersection over union: a metric and a loss for bounding box regression[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. June 15-20,2019,Long Beach,CA,USA.IEEE,2019:658-666.