基于轻型卷积神经网络的火焰检测方法  

Fire detection based on lightweight convolutional neural network

作　　者：严云洋 杜晨锡[1,2] 刘以安 高尚兵[1] YAN Yunyang;DU Chenxi;LIU Yian;GAO Shangbing(Faculty of Computer&Software Engineering,Huaiyin Institute of Technology,Huaian 223003,Jiangsu,China;School of Internet of Things Engineering,Jiangnan University,Wuxi 214122,Jiangsu,China;School of Computer Engineering,Jiangsu Ocean University,Lianyungang 222005,Jiangsu,China)

机构地区：[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

出　　处：《山东大学学报（工学版）》2020年第2期100-107,共8页Journal of Shandong University（Engineering Science）

基　　金：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”资助。

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。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.

关 键 词：火焰检测 MobileNet 膨胀卷积 通道重排 DCB 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]