为了解决可见光与红外图像融合导致颜色不连续、目标分布碎片化的问题,文中提出了一种面向电力设备的多维交互注意力增强YOLOv8n 多模态检测算法。针对融合图像颜色不连续、缺乏视觉一致性现象,通过增强通道、高度、宽度三个维度之间特征相互作用,设计并引入了多维交互注意力模块;为了避免过拟合,使用平滑技术对 Slide Loss 损失函数进行改进,替换了YOLOv8n 的完全交并比(complete intersection over union,CIoU)Loss 损失函数。实验结果表明,改进YOLOv8n 算法在可见光和红外融合图像目标检测任务中表现优异,在精确率、召回率和交并比(intersection over union,IoU)高阈值范围内的性能上均优于多种改进的YOLO 系列算法,能够有效提升目标检测的精度。
[1] 李淑慧,蔡伟,王鑫,等 .深度学习框架下的红外与可见光图像融 合方法综述[J/OL].计算机工程与应用,2025∶1-19[2025-04-22]. http ∶//kns. cnki. net/kcms/detail/11.2127. TP. 20250122.1601. 007.html.
[2] 朱自文,宋晓鸥,崔巍,等 . 可见光-红外图像融合的目标检 测综述[J/OL].计算机工程与应用,2025 ∶1-26[2025-04-22]. http ∶//kns. cnki. net/kcms/detail/11.2127. TP. 20250321.1453. 006.html.
[3] 贺强,刘洋,游鑫,等 .基于改进YOLOv9的电力设备红外目标检 测模型[J/OL].计算机技术与发展 ,2025∶1-9[2025-04-22]. https ∶//doi. org/10.20165/j. cnki. ISSN1673-629X. 2025.0112.
[4] 杨达伟,杨明圣,付博 . 改进 YOLOv7的电力设备红外图像 多目标检测[J]. 红外技术,2025,47(3)∶326-334.
[5] 王秋忆,周浩,郑婷婷 . 改进 RetinaNet的电力设备目标检测 方法[J]. 计算机与现代化,2024(1)∶47-52,58.
[6] 屈璐瑶 . 基于多尺度特征融合与特征细化的电力设备目标 检测研究[D]. 天津∶天津大学,2023.
[7] 赵杰,陈嘉晋 . 变电站低照度场景红外可见光图像融合[J]. 电气技术,2025,26(3)∶22-29.
[8] 陈海秀,卢海涛,房威志,等 . 基于多重特征的红外与可见 光图像融合算法[J/OL]. 电光与控制,2025∶1-8[2025-04-22]. http ∶//kns. cnki. net/kcms/detail/41.1227. TN. 20250320.0952. 002.html.
[9] Jiang Q,Liu Y,Yan Y,et al. A contour angle orientation for power equipment infrared and visible image registration[J]. IEEE Transactions on Power Delivery,2020,36(4)∶2559-2569.
[10] Yang Yu,Yi Zhang,Zeyu Cheng,et al. MCA ∶Multidimen sional collaborative attention in deep convolutional neural networks for image recognition[J]. Engineering Applications of Artificial Intelligence,2023,126(Part C)∶107079.
[11] Ziping Yu,Hongbo Huang,Weijun Chen,et al. YOLO- FaceV2∶A scale and occlusion aware face detector[J].Pattern Recognition,2024,(155)∶110714
[12] Yaseen M. What is YOLOv8∶An in-depth exploration of the internal features of the next-generation object detector[DB/ OL].(2024-08-28)[2025-04-22]. https ∶//arxiv. org/abs/2408. 15857.
[13] Zhang H,Xu C,Zhang S.Inner-IoU ∶More effective intersec- tion over union loss with auxiliary bounding box[DB/OL].
(2023-11-14)[2025-04-22]. https ∶//arxiv. org/abs/2311.02877.
[14] Wang J,Xu C,Yang W,et al. A normalized Gaussian Was serstein distance for tiny object detection[DB/OL].(2022-06- 14)[2025-04-22].https ∶//arxiv.org/abs/2110.13389.
[15] Kim D,Heo B,Han D. DenseNets reloaded ∶paradigm shift beyond ResNets and ViTs[C]//European Conference on Com puter Vision.Cham ∶Springer,2024∶395-415.
[16] Khaled A.BCN ∶batch channel normalization for image clas- sification[C]//International Conference on Pattern Recogni tion. ham∶Springer,2025∶295-308.