您的位置：山东大学 -> 科技期刊社 -> 《山东大学学报（工学版）》

山东大学学报 (工学版) ›› 2020, Vol. 50 ›› Issue (2): 100-107.doi: 10.6040/j.issn.1672-3961.0.2019.424

• 机器学习与数据挖掘 • 上一篇    下一篇

基于轻型卷积神经网络的火焰检测方法

严云洋1,2,3(),杜晨锡1,2,刘以安2,高尚兵1   

  1. 1. 淮阴工学院计算机与软件工程学院, 江苏 淮安 223003
    2. 江南大学物联网工程学院, 江苏 无锡 214122
    3. 江苏海洋大学计算机工程学院, 江苏 连云港 222005
  • 收稿日期:2019-07-25 出版日期:2020-04-20 发布日期:2020-04-16
  • 作者简介:严云洋(1967—)，男，江苏淮安人，教授，博士，CCF会员，主要研究方向为数字图像处理,模式识别. E-mail：yunyang@hyit.edu.cn
  • 基金资助:
    国家自然科学基金资助项目(61402192);江苏省“六大人才高峰”项目(2013DZXX-023);江苏省“青蓝工程”;淮安市“533英才工程”

Fire detection based on lightweight convolutional neural network

Yunyang YAN1,2,3(),Chenxi DU1,2,Yian LIU2,Shangbing GAO1   

  1. 1. Faculty of Computer & Software Engineering, Huaiyin Institute of Technology, Huaian 223003, Jiangsu, China
    2. School of Internet of Things Engineering, Jiangnan University, Wuxi 214122, Jiangsu, China
    3. School of Computer Engineering, Jiangsu Ocean University, Lianyungang 222005, Jiangsu, China
  • Received:2019-07-25 Online:2020-04-20 Published:2020-04-16
  • Supported by:
    国家自然科学基金资助项目(61402192);江苏省“六大人才高峰”项目(2013DZXX-023);江苏省“青蓝工程”;淮安市“533英才工程”

RichHTML

10

PDF (PC)

413

摘要：

提出一种基于MobileNet的轻型火焰检测方法,基于深度分离卷积和膨胀卷积的膨胀卷积模块(dilated convolution block, DCB)扩增特征的感受野,加强特征语义信息,提高了视频火焰目标的检测率;优化SSD(Single Shot Multibox Detector)检测框架,提出了一种轻型的检测模型DMSSD(Dilated MobileNet-SSD)。在PASCAL VOC数据集和Bilkent大学VisiFire数据集上进行火焰检测试验,试验结果表明火焰检测的平均精度均值分别提升了1.7%和3.8%,火焰检测速度也可达80帧/s,具有较强的鲁棒性和实用性。

关键词: 火焰检测, MobileNet, 膨胀卷积, 通道重排, DCB

Abstract:

A novel lightweight flame detection method was proposed based on MobileNet. The video flame detection rate was promoted by the feature receptive field of DCB(dilated convolution block)module expand based on depthwise separable convolution and dilated convolution to strengthen the feature semantic information. The SSD(single shot multibox detector) detection framework was also optimized. The lightweight detection model DMSSD(Dilated MobileNet-SSD) was provided. Experiments showed that the mean average precision was increased by 1.7% and 3.8% respectively on the PASCAL VOC dataset and the VisiFire dataset of Bilkent University. Furthermore, the detection speed was up to 80 frames per second. The robustness and real-time performance of DMSSD were strong.

Key words: fire detection, MobileNet, dilated convolution, channel shuffle, DCB

中图分类号: 

  • TP391

图1

MobileNet-SSD模型结构"

图2

常规卷积与膨胀卷积的对比"

表1

各卷积层感受野"

层级 输入尺寸 卷积核尺寸 步长 输出尺寸 感受野
Conv5 38×38 1 1 38×38 43
Conv11 19×19 1 1 19×19 219
Conv13 10×10 1 1 10×10 315
Conv14-2 10×10 3 2 5×5 379
Conv15-2 5×5 3 2 3×3 507
Conv16-2 3×3 3 2 2×2 763
Conv17-2 2×2 3 2 1×1 1 275

图3

网格化示例"

图4

通道重排"

图5

残差连接"

图6

膨胀卷积模块"

图7

DMSSD模型结构"

表2

VOC2007测试集中各组件性能"

方法 移除Conv16 复合膨胀卷积 通道重排 残差连接 mAP/%
MobileNet 72.7
对照组1 √ 73.0
对照组2 √ √ 74.0
对照组3 √ √ √ 74.2
DMSSD √ √ √ √ 74.4

表3

PASCAL VOC 2007数据集检测结果"

方法 MACs/
109		 Parameters/
106		 GPU Inference/
ms		 mAP/
%
Tiny-YOLO[16] 3.49 15.86 6.85 57.1
DSOD_smallest[17] 5.29 5.90 18.74 73.6
Pelee[18] 1.21 5.43 14.17 76.4
MobileNet-SSD[1] 1.16 5.77 5.92 72.7
DMSSD 1.25 5.76 6.50 74.4

表4

PASCAL VOC 2007分类别检测结果"

方法 飞机 自行车 鸟 船 瓶子 大巴 轿车 猫 椅子 奶牛 桌子 狗 马 摩托 人 植物 绵羊 沙发 火车 电视
MobileNet-SSD[1] 73.9 82.4 71.1 61.2 39.1 82.6 80.2 88.2 53.8 67.8 78.4 80.8 87.9 85.6 76.5 43.4 65.0 79.4 86.7 69.6
DMSSD 74.7 83.8 71.3 59.7 40.8 84.6 81.0 88.6 56.8 72.6 77.9 83.2 89.2 86.5 77.3 45.5 71.8 79.7 87.8 75.6

表5

VisiFire数据集中样本数量"

类别 训练集 测试集 合计
图片 5 224 2 582 7 806
目标 10 920 4 130 15 050

表6

VisiFire数据集检测结果"

模型 MACs/
109		 Parameters/
106		 FPS/
(帧·s-1)		 mAP/
%
MobileNet-SSD[1] 1.13 5.52 84 74.3
DMSSD 1.22 5.49 80 78.1

图8

模型性能对比"

图9

视频示例及检测结果"

表7

火焰视频的检测结果"

视频 总帧数 MobileNet-SSD[1] Tiny-YOLO[16] Pelee[18] Tiny-DSOD[19] DMSSD
TP/% FP/% TP/% FP/% TP/% FP/% TP/% FP/% TP/% FP/%
Video1 200 96.0 4.0 94.5 5.5 98.5 1.5 98.0 2.0 97.5 2.5
Video2 216 95.8 4.2 96.8 3.2 98.1 1.9 97.7 2.3 97.2 2.8
Video3 439 90.7 9.3 84.1 15.9 93.2 6.8 91.8 8.2 92.5 7.5
Video4 170 94.7 5.3 97.1 2.9 97.6 2.4 95.3 4.7 96.5 3.5
Video5 595 92.9 7.1 84.5 15.5 96.1 3.9 92.4 7.6 95.0 5.0
Video6 470 92.1 7.9 72.3 27.7 94.7 5.3 91.5 8.8 91.9 8.1
平均 348.3 93.7 6.3 88.2 11.8 96.4 3.6 94.5 5.5 95.1 4.9

表8

非火焰检测视频结果"

视频 总帧数 MobileNet-SSD[1] Tiny-YOLO[16] Pelee[18] Tiny-DSOD[19] DMSSD
TP/% FP/% TP/% FP/% TP/% FP/% TP/% FP/% TP/% FP/%
Video7 144 93.1 6.9 85.4 14.6 95.1 4.9 92.4 7.6 93.8 6.2
Video8 120 93.3 6.7 91.6 8.4 95.0 5.0 94.2 5.8 95.0 5.0
Video9 228 96.9 3.1 94.7 5.3 97.8 2.2 96.1 3.9 97.4 2.6
平均 164 94.4 5.6 90.6 9.4 96.0 4.0 94.2 5.8 95.4 4.6
1 HOWARD A G, ZHU Menglong, CHEN Bo, et al. Mobilenets: efficient convolutional neural networks for mobile vision applications[J]. arXiv preprint arXiv: 1704.04861, 2017. https://arxiv.org/abs/1704.04861.
2 CHENG Yu, WANG Duo, ZHOU Pan, et al. A survey of model compression and acceleration for deep neural networks[J]. arXiv preprint arXiv: 1710.09282, 2017. https://arxiv.org/abs/1710.09282.
3 HAN Song, MAO Huizi, DALLY W J. Deep compression: compressing deep neural networks with pruning, trained quantization and huffman coding[J]. arXiv preprint arXiv: 1510.00149, 2015. https://arxiv.org/abs/1510.00149.
4 LUO Ping, ZHU Zhenyao, LIU Ziwei, et al. Face model compression by distilling knowledge from neurons[C]//Proceedings of 30th AAAI Conference on Artificial Intelligence. Menlo Park, USA: Association for the Advancement of Artificial Intelligence, 2016: 3560-3566.
5 JADERBERG M, VEDALDI A, ZISSERMAN A. Speeding up convolutional neural networks with low rank expansions[J]. arXiv preprint arXiv: 1405.3866, 2014. https://arxiv.org/abs/1405.3866.
6 SZEGEDY C, VANHOUCKE V, IOFFE S, et al. Rethinking the inception architecture for computer vision[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, USA: IEEE Computer Society, 2016: 2818-2826.
7 CHOLLET F. Xception: deep learning with depthwise separable convolutions[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu, USA: IEEE Computer Society, 2017: 1800-1807.
8 ZHANG Xiangyu, ZHOU Xinyu, LIN Mengxiao, et al. Shufflenet: an extremely efficient convolutional neural network for mobile devices[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA: IEEE Computer Society, 2018: 6848-6856.
9 CHEN Liangchieh, PAPANDREOU G, SCHROFF F, et al. Rethinking atrous convolution for semantic image segmentation[J]. arXiv preprint arXiv: 1706.05587, 2017. https://arxiv.org/abs/1706.05587.
10 DUMOULIN V, VISIN F. A guide to convolution arithmetic for deep learning[J]. arXiv preprint arXiv: 1603.07285, 2016. https://arxiv.org/abs/1603.07285.
11 WEI S E, RAMAKRISHNA V, KANADE T, et al. Convolutional pose machines[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, USA: IEEE Computer Society, 2016: 4724-4732.
12 CAO Xudong. A practical theory for designing very deep convolutional neural network[R/OL].[2018-08-01]. https://www.kaggle.com/c/datasciencebowl/discussion/13166.
13 YU Fisher, KOLTUN V. Multi-scale context aggregation by dilated convolutions[J]. arXiv preprint arXiv: 1511.07122, 2015. https://arxiv.org/abs/1511.07122.
14 WANG Panqu, CHEN Pengfei, YUAN Ye, et al. Understanding convolution for semantic segmentation[C]//2018 IEEE Winter Conference on Applications of Computer Vision (WACV). Nevada, USA: IEEE Computer Society, 2018: 1451-1460.
15 CHEN Liangchieh, ZHU Yukun, PAPANDREOU G, et al. Encoder-decoder with atrous separable convolution for semantic image segmentation[J]. arXiv preprint arXiv: 1802.02611, 2018. https://arxiv.org/abs/1802.02611
16 REDMON J, FARHADI A. Yolo9000: better, faster, stronger[C]//Computer Vision and Pattern Recognition. Washington, USA: IEEE Computer Society, 2017: 6517-6525.
17 SHEN Zhiqiang, LIU Zhuang, LI Jianguo, et al. Dsod: learning deeply supervised object detectors from scratch[C]//Proceedings of the IEEE International Conference on Computer Vision. Venice, Italia: IEEE Computer Society, 2017: 1919-1927.
18 WANG R J, LI Xiang, LING C X. Pelee: a real-time object detection system on mobile devices[C]//Advances in Neural Information Processing Systems. Montreal, Canada: Neural Information Processing Sys-tems Foundation, Inc., 2018: 1963-1972.
19 LI Yuxi, LI Jiuwei, LIN Weiyao, et al. Tiny-dsod: lightweight object detection for resource-restricted usages[J]. arXiv preprint arXiv: 1807.11013, 2018. https://arxiv.org/abs/1807.11013.
20 ZOPH B, VASUDEVAN V, SHLENS J, et al. Learning transferable architectures for scalable image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA: IEEE Computer Society, 2018: 8697-8710.
[1] 严云洋,张慧珍,刘以安,高尚兵. 基于GMM与三维LBP纹理的视频火焰检测[J]. 山东大学学报 (工学版), 2019, 49(1): 1-9.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
[1] 王素玉,艾兴,赵军,李作丽,刘增文 . 高速立铣3Cr2Mo模具钢切削力建模及预测[J]. 山东大学学报(工学版), 2006, 36(1): 1 -5 .
[2] 于海波,李宇,余恬,雷虹 . W波段折叠波导慢波系统的尺寸对其冷特性的影响[J]. 山东大学学报(工学版), 2008, 38(3): 90 -94 .
[3] 庞志俭 张长桥. 甲基丙烯酸十二酯基二元共聚制备缔合减阻剂的合成与性能研究[J]. 山东大学学报(工学版), 2009, 39(5): 128 -132 .
[4] 李新平 代翼飞 胡静. 某岩溶隧道围岩稳定性及涌水量预测的流固耦合分析[J]. 山东大学学报(工学版), 2009, 39(4): 1 -6 .
[5] 高阳 张庆松 原小帅 许振浩 刘斌. 地质雷达在岩溶隧道超前预报中的应用[J]. 山东大学学报(工学版), 2009, 39(4): 82 -86 .
[6] 翟新献 陈东海 郭念波 勾攀峰. 济三煤矿沿空巷道矿压显现规律研究[J]. 山东大学学报(工学版), 2009, 39(4): 92 -96 .
[7] 乔小燕. 赤潮藻显微图像自动识别方法[J]. 山东大学学报(工学版), 2016, 46(3): 1 -6 .
[8] 吕国仁,闫书明,白书锋,贾 宁,马 亮 . 高速公路新型波形梁护栏端头实车碰撞性能研究[J]. 山东大学学报(工学版), 2008, 38(4): 47 -52 .
[9] 李杰 刘弘. 基于遗传算法的分形艺术图案生成方法[J]. 山东大学学报(工学版), 2008, 38(6): 33 -36 .
[10] 李春晓 岳钦艳 卢磊 高宝玉 杨忠莲 司晓慧 倪寿清 王元芳. 疏水缔合阳离子聚丙烯酰胺的合成与应用[J]. 山东大学学报(工学版), 2008, 38(6): 99 -104 .
版权所有 © 2018 《山东大学学报(工学版)》编辑部　
地址: 济南市山大南路27号（250100）　 电话: 0531-88366735　 E-mail: xbgxb@sdu.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发