| 摘 要: 针对化工火灾烟雾目标检测算法模型存在准确率低、参数多以及计算复杂度大等问题,提出了一种基于改进 YOLOv8s的化工火灾烟雾检测算法模型CIFM-YOLO。该模型首先在主干网络中引入高效神经网络架构FasterNetBlock和卷积门控线性单元CGLU(Convolutional Gate Linear Unit),并融入C2f(CSPDarknet53 to 2-stage FPN)模块,形成新的轻量化网络结构 Faster-C2f-CGLU;其次在颈部网络中融合轻量化颈部网络 Slim-Neck和ASF-YOLO架构的思想,形成新的轻量注意尺度序列融合颈部网络LightAsfNeck。实验结果显示,较原 YOLOv8s模型,CIFM-YOLO模型的准确率上升了7.7百分点,mAP@0.5和 mAP@0.5:0.95指标分别提高了2.1百分点和5.1百分点,浮点运算量减少了24.3%,参数量减少了22%,优化效果显著。 |
| 关键词: 化工火灾烟雾检测;YOLOv8s;CIFM-YOLO |
|
中图分类号: TP391
文献标识码: A
|
|
| Research on Chemical Fire Smoke Detection Algorithm Based on Improved YOLOv8s |
|
QIAN Hao, FU Lihui
|
(School Automation, Huaiyin Institute of Technology, Hua'i an 223003, China)
qian116972@163.com; flh3650326@163.com
|
| Abstract: To address issues of low accuracy, excessive parameters, and high computational complexity in chemical fire smoke detection algorithm models, this study proposes CIFM-YOLO, a chemical fire smoke detection algorithm model based on improved YOLOv8s. Firstly, the model introduces the efficient neural network architecture FasterNet Block and the Convolutional Gate Linear Unit (CGLU) into the backbone network, integrating them into the C2f (CSPDarknet53 to 2-stage FPN) module to form a new lightweight network architecture called Faste-r C2-f CGLU.Secondly, a new lightweight attention scale sequence fusion neck network, LightAsfNeck, is formed by integrating the lightweight neck network Slim-Neck and the concepts of the ASF-YOLO architecture in the neck network.Experimental results demonstrate that compared to the original YOLOv8s model, the CIFM-YOLO model achieves a 7.7 percentage points improvement in accuracy, with mAP @ 0. 5 and mAP @ 0. 5: 0. 95 metrics increasing by 2. 1 percentage points and 5.1 percentage points respectively. Additionally, floating-point operations (FLOPs) are reduced by 24.3% , and the number of parameters decreases by 22% , showing significant optimization. |
| Keywords: chemical fire smoke detection; YOLOv8s; CIFM-YOLO |
