| 摘 要: 针对现有仪器设备检测算法在管廊复杂环境下检测精度低、模型复杂等问题,提出一种改进YOLOv8n的SLB-YOLOv8n仪器设备检测算法。首先,构建C2f-star模块并替换主干网络中的C2f,降低模型复杂度;其次,在SPPF模块添加LSKA注意力机制,增强低光环境识别能力,并将特征融合方式替换为BiFPN,提高识别精度;最后,将损失函数替换为 WIoUv3,加快模型收敛速度。使用管廊自建数据集进行训练,结果表明SLB-YOLOv8n相比YOLOv8n参数量减少了31.9%,而mAP 提升了0.9%。改进后的算法在轻量化的同时提高了识别精度。 |
| 关键词: 改进的YOLOv8n算法 StarNet 目标检测 注意力机制 轻量化 |
|
中图分类号: TP391
文献标识码: A
|
|
| Research on Target Detection Algorithm for Utility Tunnel Equipment Based on Deep Learning |
|
SHI Weimin1, RUAN Fangcao1, LI Zhiqiang2, SUN Lei1, QING Dong2
|
(1.School of Mechanical Engineering, Zhejiang SC-I TECH University, Hangzhou 3100182, China;
2.Suzhou TeYu Robot Technology Co., Ltd., Suzhou 215163, China)
swm@zstu.edu.cn; ruanfangcao@163.com; lzq_2002_0@163.com;; 202110601013@mails.zstu.edu.cn; qing1447228775@163.com
|
| Abstract: To address issues such as low detection accuracy and high model complexity of existing equipment detection algorithms in complex utility tunnel environments, this study proposes an improved SLB-YOLOv8n algorithm
for instrument and equipment detection. First, a C2f-star module is constructed to replace the C2f in the backbone network, reducing model complexity. Second, the LSKA attention mechanism is integrated into the SPPF module to enhance recognition capability in low-light conditions. The feature fusion method is replaced with BiFPN to improve detection accuracy. Finally, the loss function is substituted with WIoUv3 to accelerate model convergence. Trained on a self-built utility tunnel dataset, results show that SLB-YOLOv8n reduces parameters by 31.9% while increasing mAP by 0.9% compared to YOLOv8n. The enhanced algorithm achieves lightweight design while improving recognition precision. |
| Keywords: improved YOLOv8n algorithm StarNet target detection attention mechanism lightweight |
