基于深度学习的森林火灾烟雾监测

doi:10.13466/j.cnki.lyzygl.2023.04.018

林业资源管理 ›› 2023›› Issue (4): 150-160.doi: 10.13466/j.cnki.lyzygl.2023.04.018

基于深度学习的森林火灾烟雾监测

郑琰睿¹(), 杨林剑², 李曙光¹(), 张永炬³

1.浙江科技学院自动化与电气工程学院,杭州 310000
2.浙江伟星新型建材股份有限公司智能制造中心,浙江台州 318000
3.台州学院智能制造学院,浙江台州 318000

收稿日期:2023-05-09 修回日期:2023-06-20 出版日期:2023-08-28 发布日期:2023-10-16
通讯作者: 李曙光(1982-),男,浙江兰溪人,副教授,博士,主要研究方向:自动化检测技术与精密测量。Email:lishuguang@zust.edu.cn
作者简介:郑琰睿(1998-),男,山东烟台人,硕士研究生,主要研究方向:深度学习与图像处理。Email:yr-zheng@foxmail.com
基金资助:
浙江省基础公益研究计划项目“基于高灵敏光学弱磁探测的管道缺陷无损云检测系统应用基础研究”(LGF22F010009)

Deep Learning-Based Forest Fire Smoke Detection

ZHENG Yanrui¹(), YANG Linjian², LI Shuguang¹(), ZHANG Yongju³

1. School of Automation and Electrical Engineering,Zhejiang University of Science and Technology,Hangzhou 310000,China
2. Intelligent Manufacturing Center,Zhejiang Weixing New Building Materials Co.,Ltd.,Taizhou,Zhejiang 318000,China
3. Intelligent Manufacturing Academy,Taizhou University,Taizhou,Zhejiang 318000,China

Received:2023-05-09 Revised:2023-06-20 Online:2023-08-28 Published:2023-10-16

摘要/Abstract

摘要：

为了第一时间发现森林火灾,避免因森林火灾造成严重后果,提出以森林火灾烟雾为主要目标的检测模型YOLO-SCW,在YOLOv7的基础上,引入SPD-Conv层,以减小特征提取过程中小目标特征缺失的问题。之后在检测头金字塔池化部分增加坐标准意力机制模块,通过将位置信息编码到通道中,增加了模型对目标的关注度,并且减少了背景对检测效果的干扰。最后,通过WIoU矩形框损失函数,提高了预测框的回归速度与精度。在测试过程中,改进后的YOLO-SCW较YOLOv7模型mAP提高了9.1%,并减少了误检与漏检现象,证明YOLO-SCW有着更好的特征提取能力与泛化能力,对森林火灾烟雾监测任务表现出色。

关键词: YOLO-SCW, 森林火灾烟雾监测, 目标检测, 深度学习, 损失函数

Abstract:

In order to detect forest fires in the first time and avoid serious consequences caused by forest fires,a detection model YOLO-SCW with forest fire smoke as the main target is proposed,and the SPD-Conv layer is introduced based on YOLOv7 to reduce the problem of missing features of small targets in the feature extraction process.Then,the Coordinate Pay module is added in the pooling part of the detection head pyramid,and the location information is encoded into the channel,which increases the attention of the modelto the target and reduces the interference of the background on the detection effect.Finally,the WIoU rectangular box loss function is used to improve the regression speed and accuracy of the prediction box.During the test,the improved YOLO-SCW increased by 9.1% compared with the mAP of the YOLOv7 model,and reduced the false detection and missed detection,which proved that YOLO-SCW has better feature extraction and generalization ability,and has excellent performance for forest fire smoke detection tasks.

Key words: YOLO-SCW, forest fire smoke detection, object detection, deep learning, loss function

中图分类号:

S762.2

郑琰睿, 杨林剑, 李曙光, 张永炬. 基于深度学习的森林火灾烟雾监测[J]. 林业资源管理, 2023,(4): 150-160.

ZHENG Yanrui, YANG Linjian, LI Shuguang, ZHANG Yongju. Deep Learning-Based Forest Fire Smoke Detection[J]. FOREST RESOURCES WANAGEMENT, 2023,(4): 150-160.

图/表 15

图1

图2

图3

图4

图5

图6

表1

图7

图8

图9

图10

图11

图12

图13

图14

参考文献 23

[1]	白夜, 王博, 武英达, 等. 2021年全球森林火灾综述[J]. 消防科学与技术, 2022, 41(5):705-709.
[2]	覃先林, 陈尔学, 李增元, 等. 基于MODIS数据的森林覆盖变化监测方法研究[J]. 遥感技术与应用, 2006(3):178-183.
[3]	张玲玲, 邱建文, 王文江. 气溶胶激光雷达在森林防火中的应用[J]. 焦作大学学报, 2021, 35(4):93-95.
[4]	陈曦, 刘和剑. 自动巡航森林火灾检测小车的设计[J]. 绿色科技, 2016(16):163-166.
[5]	Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[C]. International Conference on Learning Representations(ICLR), 2015,abs/1409.1556:1-14.
[6]	Szegedy C, Liu Wei, Jia Yangqing, et al. Going deeper with convolutions[C]. 2015 IEEE Conference on Computer Vision and Pattern Recognition(CVPR), 2015, 7298594:1-9.
[7]	He Kaiming, Zhang Xiongyu, Ren Shaoqing, et al. Deep residual learning for image recognition[C]. 2016 IEEE Conference on Computer Vision and Pattern Recognition(CVPR), 2016, 90:770-778.
[8]	Redmon J, Divvala S, Girshick R, et al. You only look once:Unified,real-time object detection[C]. 2016 IEEE Conference on Computer Vision and Pattern Recognition(CVPR), 2016, 91:779-788.
[9]	张倩, 周平平, 王公堂, 等. 基于合成图像的Faster R-CNN森林火灾烟雾检测[J]. 山东师范大学学报:自然科学版, 2019, 34(2):180-185.
[10]	Xue Zhenyang, Lin Haifeng, Wang Fang. A small target forest fire detection model based on YOLOv5 improvement[J]. Forests, 2022, 13(8):1332. doi: 10.3390/f13081332
[11]	皮骏, 刘宇恒, 李久昊. 基于YOLOv5s的轻量化森林火灾检测算法研究[J]. 图学学报, 2023, 44(1):26-32.
[12]	叶铭亮, 周慧英, 李建军. 基于改进Swin Transformer的森林火灾检测算法[J]. 中南林业科技大学学报, 2022, 42(8):101-110.
[13]	Lin Ji, Lin Haifeng, Wang Fang. STPM_SAHI:A small-target forest fire detection model based on swin transformer and slicing aided hyper inference[J]. Forests, 2022, 13(10):1603. doi: 10.3390/f13101603
[14]	Wang Chienyao, Bochkovskiy A, Liao Hongyuan. YOLOv7:Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[J/OL]. Computer Science(2022-07-06)[2022-11-21]. https://arxiv.org/abs/2207.02696.arXiv preprint,2022,arXiv:2207.02696.
[15]	Lee Youngwan, Hwang Joongwon, Lee Sangrok, et al. An energy and GPU-computation efficient backbone network for real-time object detection[J/OL]. Computer Science(2019-04-22)[2023-02-21]. https://arxiv.org/abs/1904.09730.arXiv preprint,2019,arXiv:1904.09730.
[16]	Wang Chienyao, Liao Hongyuan, Yeh I-Hau. Designing network design strategies through gradient path analysis[J/OL].(2022-11-09)[2023-2-15]. https://arxiv.org/abs/2211.04800.arXiv preprint,2022,arXiv:2211.04800.
[17]	Ding Xiaohan, Zhang Xiangyu, Ma Ningning. et al. RepVGG:Making GG-style convNets great again[C]. 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR), 2021, 01352:13728-13737.
[18]	Raja S, Luo T. No more strided convolutions or pooling:A new CNN building block for low-resolution images and small objects[J/OL]. Computer Science(2022-08-07)[2023-03-02]. https://arxiv.org/abs/2208.03641.arXiv preprint,2022,arXiv:2208.03641.
[19]	Zheng Zhaohui, Wang Ping, Ren Dongwei, et al. Enhancing geometric factors in model learning and inference for object detection and instance segmentation[J]. IEEE Transactions on Cybernetics, 2022, 52(8):8574-8586. doi: 10.1109/TCYB.2021.3095305
[20]	Tong Zanjia, Chen Yuhang, Xu Zewei, et al. Wise-IoU:Bounding box regression loss with dynamic focusing mechanism[J/OL]. Computer Science(2023-01-24)[2023-04-09]. https://arxiv.org/abs/2301.10051.arXiv preprint,2023,arXiv:2301.10051.
[21]	Hou Qibin, Zhou Daquan, Feng Jiashi, et al. Coordinate attention for efficient mobile network design[C]. 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR), 2021, 01350:13708-13717.
[22]	Du Shuangjiang, Zhang Pin, Xiang Pengan, et al. Improved bounding box regression loss function based on CIOU loss for multi-scale object detection[C]. 2021 IEEE 2nd International Conference on Pattern Recognition and Machine Learning(PRML), 2021, 9520717:92-98.
[23]	Hu Jie, Shen Li, Sun Gang, et al. Squeeze-and-Excitation networks[C]. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018, 00745:7132-7141.

序号	gamma	SPD	CA	WIoU	精度/%	召回率/%	mAP(IoU=0.5)
1	×	×	×	×	87.2	83.7	86.3
2	√	×	×	×	90.3	89.1	91.2
3	×	√	×	×	91.1	90.4	93.9
4	×	×	√	×	91.9	85.0	91.4
5	×	×	×	√	80.0	81.6	83.8
6	√	√	√	×	90.5	91.1	93.4
7	√	√	√	√	94.4	92.5	95.4

基于深度学习的森林火灾烟雾监测

Deep Learning-Based Forest Fire Smoke Detection

RichHTML

可视化

摘要/Abstract

引用本文

使用本文

图/表 15

参考文献 23

相关文章 4

编辑推荐

Metrics

本文评价

[1]	汪帅, 卢楠, 郑红, 李晖, 彭检贵, 张同, 魏彦华. 基于深度学习的野生动物视频目标检测[J]. 林草资源研究, 2024, 0(1): 34-40.
[2]	张殿岱, 王雪梅. 基于高分辨率遥感影像的植被分类方法比较[J]. 林业资源管理, 2021, 0(3): 108-113.
[3]	林双双, 钟九生, 何鑫, 江丽, 段纪维, 代仁丽, 何志远. 光谱信息支持下城区林地信息提取方法[J]. 林业资源管理, 2021, 0(3): 96-100.
[4]	黄彦晓, 方陆明, 黄思琪, 高海力, 杨来邦, 楼雄伟. 基于改进的Faster R-CNN模型的树冠提取研究[J]. 林业资源管理, 2021, 0(1): 173-179.