A lightweight forest fire detection algorithm based on YOLOv5s

doi:10.16265/j.cnki.issn1003-3033.2025.01.0127

Abstract

Abstract:

In order to solve the problems of complex structure, large scale and difficulty in balancing detection accuracy and efficiency of the current forest fire detection algorithm based on deep learning, a lightweight forest fire detection algorithm based on YOLOv5s was proposed. Firstly, an optimized background difference technique was used to eliminate the interference of fire-like objects in the background image, thus reducing the time required for image analysis. Secondly, a group blending strategy was designed to optimize the conventional convolution, and an efficient channel attention (ECA) mechanism and depthwise separable convolution were incorporated into the C3 module of feature extraction, which enhanced the ability of image feature extraction and fusion and at the same time effectively reduces the number of model parameters. Then, a dynamic non-monotonic focusing mechanism was used to optimize the WIOU loss function, reducing the harmful gradients generated by low-quality samples. Finally, sufficient experimental comparisons between the proposed algorithm and other algorithms on the constructed forest fire dataset. The results show that the proposed algorithm shows good generalization in various scenarios, and the detection accuracy of the flame target can reach 86.1%, which is 2.7% higher than that of the standard YOLOv5s, and the detection speed is increased by 11.4%, which effectively reduces the fire false alarm rate and enhances the detection performance of the model.

Key words: YOLOv5s, lightweighting, forest fire detection, depthwise separable convolution, attention, wise intersection over union(WIOU)

CLC Number:

X928.7火灾与爆炸事故

LIU Huilin, FANG Qiong, JIANG Yu, WEI Huazhang, WANG Tao, ZHANG Shuchuan. A lightweight forest fire detection algorithm based on YOLOv5s[J]. China Safety Science Journal, 2025, 35(1): 75-83.

Figures/Tables 13

Fig.1

Fig.2

Table 1

Fig.3

Fig.4

Table 2

Fig.5

Fig.6

Table 3

Fig.7

Fig.8

Fig.9

Table 4

References 18

[1]	CHOI M, KIM C, OH H. A video-based SlowFastMTB model for detection of small amounts of smoke from incipient forest fires[J]. Journal of Computational Design and Engineering, 2022, 9(2):793-804.
[2]	熊昊, 李伟. 基于SVM的视频火焰检测算法[J]. 传感器与微系统, 2020, 39(1):143-145, 149.
	XIONG Hao, LI Wei. Video flame detection algorithm based on svm[J]. Sensors and Microsystems, 2020, 39(1):143-145,149.
[3]	GUAN Zhenwei, MIN Feng, HE Wei, et al. Forest fire detection via feature entropy guided neural network[J]. Entropy, 2022, 24(1):DOI:10.3390/e24010128.
[4]	李海, 孙鹏. 基于知识图谱的火灾图像识别研究[J]. 中国安全科学学报, 2023, 33(10):147-159. doi: 10.16265/j.cnki.issn1003-3033.2023.10.2088
	LI Hai, SUN Peng. Research on fire image recognition based on scientific knowledge graph[J]. China Safety Science Journal, 2023, 33(10):147-159. doi: 10.16265/j.cnki.issn1003-3033.2023.10.2088
[5]	REN Shaoqing, HE Kaiming, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6):1137-1149. doi: 10.1109/TPAMI.2016.2577031 pmid: 27295650
[6]	张为, 魏晶晶. 嵌入DenseNet结构和空洞卷积模块的改进YOLOv3火灾检测算法[J]. 天津大学学报, 2020, 53(9):976-983.
	ZHANG Wei, WEI Jingjing. Improved YOLOv3 fire detection-algorithm with embedded DenseNet structure and cavity convolution module[J]. Journal of Tianjin University, 2020, 53(9):976-983.
[7]	LIU Wei, ANGUELOV D, ERHAN D, et al. SSD: single shot multibox detector[C]. European Conference on Computer Vision (ECCV), 2016:21-37.
[8]	SHARMA J, GRANMO O, GOODWIN M, et al. Deep convolutional neural networks for fire detection in images[C]. Engineering Applications of Neural Networks:18^th International Conference, 2017:183-193.
[9]	邱思远, 王孝兰, 张伟伟, 等. 基于YOLOX的森林火灾早期预警研究[J]. 农业装备与车辆工程, 2023, 61(11):106-110.
	QIU Siyuan, WANG Xiaolan, ZHANG Weiwei, et al. Research on early of forest fire based on YOLOX[J]. Agricultural Equipment and Vehicle Engineering, 2023, 61(11):106-110.
[10]	王喆, 李享, 杨栋梁, 等. 基于YOLOv5s的城镇森林交界域火灾探测模型[J]. 中国安全科学学报, 2023, 33(6):152-158. doi: 10.16265/j.cnki.issn1003-3033.2023.06.1532
	WANG Zhe, LI Xiang, YANG Dongliang, et al. Fire detection model of wildland-urban interface based on YOLOv5s[J]. China Safety Science Journal, 2023, 33(6):152-158. doi: 10.16265/j.cnki.issn1003-3033.2023.06.1532
[11]	HOSSEINI A, HASHEMZADEH M, FARAJZADEH N. UFS-Net: a unified flame and smoke detection method for early detection of fire in video surveillance applications using cnns[J]. Journal of Computational Science, 2022, 61: DOI:10.1016/j.jocs.2022.101638.
[12]	高源, 杨望, 曹磊, 等. 基于改进YOLOv5的特征抑制多尺度林火检测算法[J]. 林业工程学报, 2023, 8(6):145-153.
	GAO Yuan, YANG Wang, CAO Lei, et al. Feature suppression multi-scale forest fire detection algorithm based on improved YOLOv5[J]. Journal of Forestry Engineering, 2023, 8(6):145-153.
[13]	张全, 张伟, 杨先凤, 等. 融合YOLOv5-ResNet级联网络的烟火检测方法[J]. 安全与环境学报, 2023, 23(2):397-405.
	ZHANG Quan, ZHANG Wei, YANG Xianfeng, et al. Smoke and fire detection method fusing YOLOv5-ResNet cascade network[J]. Journal of Safety and Environment, 2023, 23(2):397-405.
[14]	LIU Huilin, WEI Huazhang, YANG Gaoming, et al. An improved vibe algorithm based on adaptive thresholding and the deep learning-driven frame difference method[J]. Electronics, 2023, 12(16):DOI:10.3390/electronics12163481.
[15]	WANG Qilong, WU Banggu, ZHU Pengfei, et al. ECA-Net: efficient channel attention for deep convolutional neural networks[C]. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020:11 531-11 539.
[16]	CHO Y. Weighted intersection over union (WIOU) for evaluating image segmentation[J]. Pattern Recognition Letters, 2024, 185: 101-107.
[17]	REDDY E, THALE S. Pedestrian detection using YOLOv5 and complete-IoU loss for autonomous driving applications[C]. Proceedings of International Conference on Paradigms of Communication, Computing and Data Analytics, 2023:287-296.
[18]	章曙光, 唐锐, 邵政瑞, 等. 一种基于小波变换的YOLOv5火灾检测改进算法[J]. 无线电工程, 2023, 53(10):2303-2310.
	ZHANG Shuguang, TANG Rui, SHAO Zhengrui, et al. A wavelet transform-based improved algorithm for YOLOv5 fire detection[J]. Radio Engineering, 2023, 53(10):2303-2310.

模型计算	参数量	计算量	占比
常规卷积	P₁=294 912	F₁=301 989 888	0.559
分组混洗策略优化下的常规卷积	P₂=164 992	F₂=186 908 672	0.619

算法模型	AP₇₅	AP₅₀	AP
WIOUv1	52.24	62.37	45.16
WIOUv2 (r = 0.5)	53.34	63.86	45.56
本文WIOU (α = 1.7, δ = 4)	53.61	64.04	45.58

算法模型	P	R	FPR	FNR	m	模型大小/MB	检测用时/ms
Faster-RCNN	0.794	0.748	0.194	0.252	0.812	209.3	52.5
SSD	0.787	0.743	0.201	0.257	0.804	180.3	60.7
YOLOv4-tiny	0.785	0.739	0.202	0.261	0.811	27.3	8.4
YOLOv5s	0.798	0.745	0.189	0.255	0.834	14.3	9.9
YOLOv5s+小波变换	0.803	0.748	0.184	0.252	0.843	22.5	12.4
YOLOv5s+CIOU	0.814	0.842	0.192	0.158	0.846	20.2	11.7
本文算法	0.826	0.857	0.181	0.143	0.861	9.2	4.6

模型	m	速度/ (帧·s^-1)
YOLOv5s	0.834	35
YOLOv5s-ViBe	0.838	38
YOLOv5s-ViBe-分组混洗	0.839	42
YOLOv5s-ViBe-分组混洗-(ECA-C3)	0.847	39
YOLOv5s-ViBe-分组混洗- (ECA-C3)-WIOU	0.861	39