基于Faster R-CNN模型的火焰检测  

Flame Detection Based on Faster R-CNN Model

作　　者：严云洋 朱晓妤[1,2] 刘以安 高尚兵[1] Yan Yunyang;Zhu Xiaoyu;Liu Yi’an;Gao Shangbing(Faculty of Computer&Software Engineering,Huaiyin Institute of Technology,Huaian 223003,China;School of Internet of Things Engineering,Jiangnan University,Wuxi 214122,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

出　　处：《南京师大学报（自然科学版）》2018年第3期1-5,共5页Journal of Nanjing Normal University(Natural Science Edition)

基　　金：国家自然科学基金(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"

摘　　要：常规的火焰检测一般是提取火焰的静态或动态特征,然后进行火焰的判别.但是传统特征无法全面描述火焰特性,会导致识别的准确率降低.本文提出一种基于Faster R-CNN模型的火焰检测算法.首先利用候选区域生成网络(Region Proposal Network,RPN)提取火焰候选区域,然后对候选区域进行卷积及池化操作,提取火焰特征,最后利用联合训练的快速区域卷积神经网络(Fast R-CNN)进行火焰识别.实验结果表明该方法能够自动提取火焰特征,有效提高复杂背景下的火焰识别的准确率,具有良好的泛化能力和鲁棒性.Usually the static or dynamic characteristics of the flame is extracted for flame detection.But the traditional characteristics can not fully describe the characteristics of flame,which leads to the reduction of recognition accuracy.To solve this problem,a flame detection based on Faster R-CNN model is proposed in this paper.First,the candidate region of the flame is extracted by RPN.Then the convolution and pool operation of candidate regions are performed to extract the flame characteristics.Finally,Fast R-CNN is used to identify the flame.The experimental results show that the method can automatically extract the flame characteristics,effectively improve the accuracy of flame recognition in the complex background,and have good generalization ability and robustness.

关 键 词：FASTER R-CNN 候选区域生成网络 快速区域卷积神经网络 火焰检测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]