改进DeeplabV3+的火焰分割与火情分析方法

doi:10.19665/j.issn1001-2400.2021.05.006

摘要/Abstract

摘要：

火焰检测与火势发展分析对火情的控制十分重要。提出一种改进DeeplabV3+的火焰分割与火情分析方法:首先在DeeplabV3+的解码器部分增加低层特征来源,将其与高层特征融合后采用2倍上采样逐步恢复图像尺寸,保留更多的细节信息,实现更加准确的火焰分割;然后将火焰视频每帧分割得到的像素数组成火势发展序列,基于关键点对序列进行分段和线性拟合,获取火势发展的关键趋势。实验结果表明,所提方法可以在对火焰进行准确分割的基础上,有效地分析火情发展态势,为火情的检测与控制提供有效的帮助。

关键词: 深度学习, 火焰分割, DeeplabV3+, 火情分析

Abstract:

Fire detection and development analysis are significant for fire control.The fire segmentation based on the improved DeeplabV3+ and the analytical method for fire development are proposed:First,the low-level feature sources are added to the decoder of the DeeplabV3+,which is fused with high-level features,and the image size is gradually recovered by 2 times up sampling to retain more details and achieve more accurate fire segmentation.Then,the number of pixels obtained by each fire video frame is combined into a fire series,and key points are used to segment and linearly fit the series to obtain the key trend of fire development.Experimental results show that the proposed method can effectively analyze the fire development situation on the basis of accurate fire segmentation,and provide an effective help for fire detection and control.

Key words: deep learning, fire segmentation, deeplabV3+, fire analysis

中图分类号:

TP391

宁阳,杜建超,韩硕,杨传凯. 改进DeeplabV3+的火焰分割与火情分析方法[J]. 西安电子科技大学学报, 2021, 48(5): 38-46.

NING Yang,DU Jianchao,HAN Shuo,YANG Chuankai. Fire segmentation based on the improved DeeplabV3+ and the analytical method for fire development[J]. Journal of Xidian University, 2021, 48(5): 38-46.

图/表 12

表1

Resnet结构"

名称	输出大小	50层	101层
Conv 1	240×240	Conv 7×7,64,stride 2	Conv 7×7,64,stride 2
Conv 2_x	120×120	3×3 max pool $1 × 1,64 3 × 3,64 1 × 1,256$ ×3	3×3 max pool $1 × 1,64 3 × 3,64 1 × 1,256$ ×3
Conv 3_x	60×60	$1 × 1,128 3 × 3,128 1 × 1,512$ ×4	$1 × 1,128 3 × 3,128 1 × 1,512$ ×4
Conv 4_x	30×30	$1 × 1,256 3 × 3,256 1 × 1,1024$ ×6	$1 × 1,256 3 × 3,256 1 × 1,1024$ ×23
Conv 5_x	30×30	$1 × 1,512 3 × 3,512, s = 1 1 × 1,2048$ ×3	$1 × 1,512 3 × 3,512, s = 1 1 × 1,2048$ ×3

表1

图1

图2

图3

图4

图5

图6

表2

图7

表3

表4

图8

参考文献 16

[1]	YAN X, WANG L. A Fire Image Segmentation Arithmetic Based on Multi Images Edge Difference[C]// Proceedings of the 2012 International Conference on Modelling,Identification and Control.Piscataway:IEEE, 2012:140-143.
[2]	张丹丹, 章光, 陈西江, 等. 改进YCbCr和区域生长的多特征融合的火焰精准识别算法[J]. 激光与光电子学进展, 2020, 57(6):234-245.
	ZHANG Dandan, ZHANG Guang, CHEN Xijiang, et al. Flame Identification Algorithm Based on Improved Muti-Feature Fusion of YCbCr and Region Growth[J]. Laser & Optoelectronics Progress, 2020, 57(6):234-245.
[3]	TLIG L, BOUCHOUICHA M, SAYADI M, et al. A Fast Segmentation Method for Fire Forest Images Based on Multiscale Transform and PCA[J]. Sensors(Basel), 2020, 20(22):6429.
[4]	BAI X J, LU G, YAN Y. Flame Image Segmentation Using Multiscale Color and Wavelet-Based Texture Features[J]. Computer Engineering and Applications, 2017, 53(9):213-219.
[5]	WANGY B, REN J Y, et al. Application of KNN Algorithm Based on Particle Swarm Optimization in Fire Image Segmentation[J]. Journal of Electrical Engineering & Technology, 2019, 14(4):1707-1715.
[6]	朱莉, 赵俊, 傅应锴, 等. 一种红外热图像目标区域分割的深度学习算法[J]. 西安电子科技大学学报, 2019, 46(4):107-114.
	ZHU Li, ZHAO Jun, FU Yingkai, et al. Deep Learning Algorithm for the Segmentation of the Interested Region of An Infrared Thermal Image[J]. Journal of Xidian University, 2019, 46(4):107-114.
[7]	ZUO Z C, ZHANG W, ZHANG D Y. A Remote Sensing Image Semantic Segmentation Method by Combining Deformable Convolution with Conditional Random Fields[J]. Acta Geodaetica et Cartographica Sinica, 2020, 3(3):39-49.
[8]	RONNEBERGER O, FISCHER P, BROX T. U-Net:Convolutional Networks for Biomedical Image Segmentation[C]// Proceedings of the International Conference on Medical Image Computing and Computer-Assisted Intervention.Heidelberg:Springer, 2015:234-241.
[9]	CHEN L C, ZHU Y, PAPANDREOU G, et al. Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation[C]// Proceedings of the 2018 European Conference on Computer Vision.Heidelberg Springer, 2018:833-851.
[10]	刘志赢, 谢春思, 李进军, 等. 基于改进Deeplabv3+的烟雾区域分割识别算法[J]. 系统工程与电子技术, 2021, 43(2):328-335.
	LIU Zhiying, XIE Chunsi, LI Jinjun, et al. Smoke Region Segmentation Recognition Algorithm Based on Improved Deeplabv3+[J]. Systems Engineering and Electronics, 2021, 43(2):328-335.
[11]	KONG Y Y, LIU Y J, YAN B Y, et al. A Novel Deeplabv3+ Network for SAR Imagery Semantic Segmentation Based on the Potential Energy Loss Function of Gibbs Distribution[J]. Remote Sensing, 2021, 13(3):454. doi: 10.3390/rs13030454
[12]	BAHETI B, INNANI S, GAJRE S, et al. Semantic Scene Segmentation in Unstructured Environment with Modified DeepLabV3+[J]. Pattern Recognition Letters, 2020, 138:223-229. doi: 10.1016/j.patrec.2020.07.029
[13]	汪梓艺, 苏育挺, 刘艳艳, 等. 一种改进DeeplabV3网络的烟雾分割算法[J]. 西安电子科技大学学报, 2019, 46(6):52-59.
	WANG Ziyi, SU Yuting, LIU Yanyan, et al. Algorithm for Segmentation of Smoke Using the Improved DeeplabV3 Network[J]. Journal of Xidian University, 2019, 46(6):52-59.
[14]	LIN T Y, DOLLAR P, GIRSHICK R, et al. Feature Pyramid Networks for Object Detection[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition.Piscataway:IEEE, 2017:936-944.
[15]	CHUNG F L, FU T C, LUK R, et al. Time Series Pattern Matching Based on Perceptually Important Points[C]// Proceedings of the IJCAI Workshop on Learning from Temporal and Spatial Data.San Francisco:Morgan Kaufmann, 2001:1-7.
[16]	周大镯, 李敏强. 基于序列重要点的时间序列分割[J]. 计算机工程, 2008, 34(23):14-16.
	ZHOU Dazhuo, LI Minqiang, Time Series Segmentation Base on Series Importance Point[J]. Computer Engineering, 2008, 34(23):14-16.

网络结构			评价指标
网络结构			P	R	I_OU	S
DeeplabV3+	Resnet 50		0.826	0.797	0.839	0.822
DeeplabV3+	Resnet 101		0.818	0.783	0.818	0.808
文中网络	Resnet 50	add	0.791	0.824	0.832	0.817
	Resnet 50	concat	0.796	0.807	0.860	0.824
	Resnet 101	add	0.799	0.795	0.846	0.817
	Resnet 101	concat	0.807	0.804	0.844	0.822

视频	视频描述
Fire1	共21.2 s:室内场景;在视频第3秒时点燃纸团;火势不断增大,第18秒时达到最大;随后快速下降。
Fire2	共22.0 s:户外场景;帐篷前木棍燃烧;从视频开始一直持续到结束,火势稳定,变化较小。
Fire3	共27.5 s:户外有风场景;在视频第5秒时点燃柴火;火势逐渐增大,在第10秒时被风吹灭;随后复燃,并在第15秒时火势达到最大;之后又因风力影响而下降,在第21秒临近熄灭时,火势有一段小幅增长;最终熄灭于第26秒,该火势整体变化较大。

结果	Fire 1	Fire 2	Fire 3
序列分段数	3	4	6
分段拟合准确率/%	79.6	89.6	61.4
起火时刻	视频第3.6秒		视频第4.8秒
火势最大时刻	视频第18.0秒	视频第8.7秒	视频第15.6秒
熄灭时刻			视频第26.0秒
序列整体及各分段拟合斜率	K₀=219.8 K₁=98.7 K₂=655.7 K₃=-265.6	K₀=0.7 K₁=8.7 K₂=-12.8 K₃=10.8 K₄=-0.5	K₀=-2.7 K₁=100.6 K₂=-109.3 K₃=105.6 K₄=-136.5 K₅=70.0 K₆ =-42.3