| 摘 要: 火情检测的作用和意义在于保护人民群众的人身财产安全。为了提升火情检测的精度,提出一种基于改进YOLOv8的火情智能检测算法。首先,通过添加CBAM(Convolutional Block Attention Module)注意力模块帮助模型更加准确地检测和定位火情;其次,使用空洞空间卷积池化金字塔模块获取多尺度信息,提升火情检测精度;最后,引入BiFPN(Bi-directional Feature Pyramid Network)模块帮助模型针对性地学习特征。将该算法应用于火情数据集,实验结果表明,较之其他YOLO系列检测算法,改进YOLOv8的火情智能检测算法的火情检测效果最优;相较于YOLOv8n算法,检测精确率提升了1.4百分点,召回率提升了4.3百分点,mAP@0.5提升了5.1百分点,mAP@0.5-0.95提升了1.6百分点,可以有效地检测火情。 |
| 关键词: YOLOv8;目标检测;火情检测;深度学习;注意力机制 |
|
中图分类号: TP391
文献标识码: A
|
|
| An Intelligent Fire Detection Algorithm Based on Improved YOLOv8 |
|
YE Weihua1, LIU Haixiong2, HU Rong2
|
(1.Fuzhou Sof tware Park Technology Innovation Development Co., Ltd., Fuzhou 350101, China;
2.College of Computer Science and Mathematics, Fujian University of Technology, Fuzhou 350118, China)
ywh@fzrjy.com; Lmfwda@163.com; hurong@fjut.edu.cn
|
| Abstract: The role and significance of fire detection lie in protecting the personal and property safety of the people. In order to improve the accuracy of fire detection, an algorithm for intelligent fire detection based on improved YOLOv8 is proposed. Firstly, CBAM (Convolutional Block Attention Module) attention module is added to help the model more accurately detect and locate fires. Secondly, Atrous Spatial Pyramid Pooling (ASPP) module is used to obtain multi-scale information and improve the accuracy of fire detection. Finally, BiFPN (Bi-directional Feature Pyramid Network) module is introduced to help the model learn features more effectively. Experimental results on fire datasets show that, compared to other YOLO series detection algorithms, the improved YOLOv8 algorithm for intelligent fire detection performs the best. Compared to the YOLOv8n algorithm, the detection accuracy of the improved YOLOv8 algorithm has improved by 1.4 percentage points, the recall rate has improved by 4.3 percentage points, mAP@0.5 has improved by 5.1 percentage points, and mAP@ 0.5-0.95 has improved by 1.6 percentage points, which verifies its effectiveness in fire detection. |
| Keywords: YOLOv8; object detection; fire detection; Deep Learning; Attention Mechanism |
