基于改进的DenseNet深度网络火灾图像识别算法
|
摘要
现有油田火灾预警系统较多地采用烟感、红外等被动传感器进行烟火检测,其检测范围小,抗干扰能力弱,无法实时准确地进行火灾预警。如何从油田安防设备获取的海量图像数据中检测到烟火信息,提高抢险救灾的预测响应时间,在国内外都是一个具有挑战性的研究课题。提出一种改进的DenseNet深度神经网络架构,解决复杂图像中火灾区域的检测。为了增强特征传播的精度,降低存储数据量,采取结构化稀疏操作。将网络卷积核分为多个组,在训练过程中逐渐减小每个组内不重要的参数连接。针对油田安防领域构建的数据集存在不平衡性,增强火灾检测系统最终分类预测的准确性,引入Focal损失函数对分类层进行火灾识别。大量的定性定量实验表明,该改进网络在检测率与误检率方面均优于现有的其他深度模型。
Most of the existing fire warning systems in the oilfield use passive sensors such as smoke sensors and infrared sensors,to detect smoke and fire.Because of its small detection range and weak anti-jamming ability,it is unable to carry out fire warning in real time and accurately.Therefore,it is a challenging research topic at home and abroad to detect flame information from the massive image data collected by oilfield safety equipment and improve the disaster response time.We proposed an improved DenseNet deep neural network architecture,and used this architecture to solve the detection of fire areas in complex images.To improve the accuracy of feature propagation and reduce the amount of stored data,a structured sparse operation was proposed.Network convolution kernels were divided into several groups,and the unimportant parameter connection in each group was gradually reduced in the training process.In addition,since the dataset constructed in the field of oilfield security was imbalance,we introduced the Focal loss function to identify the fire in the classification layer so as to enhance the accuracy of the final classification prediction of the fire detection system.Qualitative and quantitative experiments show that the improved network is superior to other deep models in terms of detection rate and false positive per image.
Most of the existing fire warning systems in the oilfield use passive sensors such as smoke sensors and infrared sensors,to detect smoke and fire.Because of its small detection range and weak anti-jamming ability,it is unable to carry out fire warning in real time and accurately.Therefore,it is a challenging research topic at home and abroad to detect flame information from the massive image data collected by oilfield safety equipment and improve the disaster response time.We proposed an improved DenseNet deep neural network architecture,and used this architecture to solve the detection of fire areas in complex images.To improve the accuracy of feature propagation and reduce the amount of stored data,a structured sparse operation was proposed.Network convolution kernels were divided into several groups,and the unimportant parameter connection in each group was gradually reduced in the training process.In addition,since the dataset constructed in the field of oilfield security was imbalance,we introduced the Focal loss function to identify the fire in the classification layer so as to enhance the accuracy of the final classification prediction of the fire detection system.Qualitative and quantitative experiments show that the improved network is superior to other deep models in terms of detection rate and false positive per image.
引文
[1]洪伟,李朝锋.基于区域全卷积网络结合残差网络的火焰检测方法[J].激光与光电子学进展,2018,55(4):168-173.
[2]朱晓妤,严云洋,刘以安,等.基于深度森林模型的火焰检测[J].计算机工程,2018,44(7):264-270.
[3]曾思通,吴海彬,沈培辉.基于多特征融合的视频火焰检测方法研究[J].图学学报,2017,38(4):549-557.
[4]严云洋,陈垂雄,刘以安,等.维度加权模式动态纹理特征的火焰检测[J].智能系统学报,2017,12(4):548-555.
[5]Chen T H,Wu P H,Chiou Y C.An early fire-detection method based on image processing[C]//2004 International Conference on Image Processing.IEEE,2004:1707-1710.
[6]Dimitropoulos K,Barmpoutis P,Grammalidis N.Spatiotemporal flame modeling and dynamic texture detection analysis for automatic video-based fire[J].IEEE Transactions on Circuits&Systems for Video Technology,2015,25(2):339-351.
[7]Toreyin B U,Dedeoglu Y,et al.Computer vision based method for real-time fire and flame detection[J].Pattern Recognition Letters,2006,27(1):49-58.
[8]Gubbi J,Marusic S,Palaniswami M.Smoke detection in video using wavelets and support vector machines[J].Fire Safety Journal,2009,44(8):1110-1115.
[9]Listyorini T,Rahim R.A prototype fire detection implemented using the Internet of things and fuzzy logic[J].World Transactions on Engineering&Technology Education,2018,16(1):42-46.
[10]钟玲,张兴坤.基于SVM的视频图像火焰检测[J].软件工程,2017,20(6):1-4.
[11]Ernfjlletal.Fire detection system[J].Optoelectronics Instrumentation&Data Processing,2018,45(2):140-145.
[12]Sermanet P,Chintala S,Lecun Y.Convolutional neural networks applied to house numbers digit classification[C]//2012 21st International Conference on Pattern Recognition(ICPR 2012).IEEE Computer Society,2012.
[13]Springenberg T,Dosovitskiy A,Brox T,et al.Striving for simplicity:The all convolutional net[C]//Proceedings of International Conference on Learning Represenation Workshop Track,2015.
[14]Szegedy C,Vanhoucke V,Ioffe S,et al.Rethinking the inception architecture for computer vision[C]//2016 IEEEConference on Computer Vision and Pattern Recognition(CVPR).IEEE,2016.
[15]Huang U,Liu Z,Weinberger KQ,et al.Densely connected convolutional networks[C]//The IEEE Conference on Computer Vision and Pattern Recognition(CPVR).Las Vegas:IEEE,2017:4700-4708.
[16]Liu W,Anguelov D,Erhan D,et al.SSD:Single Shot MultiBox Detector[C]//European Conference on Computer Vision,2016:21-37.
[17]Xie S,Girshick R B,Dollar P,et al.Aggregated Residual Transformations for Deep Neural Networks[C]//2017 IEEEConference on Computer Vision and Pattern Recognition(CVPR),2017:5987-5995.
[2]朱晓妤,严云洋,刘以安,等.基于深度森林模型的火焰检测[J].计算机工程,2018,44(7):264-270.
[3]曾思通,吴海彬,沈培辉.基于多特征融合的视频火焰检测方法研究[J].图学学报,2017,38(4):549-557.
[4]严云洋,陈垂雄,刘以安,等.维度加权模式动态纹理特征的火焰检测[J].智能系统学报,2017,12(4):548-555.
[5]Chen T H,Wu P H,Chiou Y C.An early fire-detection method based on image processing[C]//2004 International Conference on Image Processing.IEEE,2004:1707-1710.
[6]Dimitropoulos K,Barmpoutis P,Grammalidis N.Spatiotemporal flame modeling and dynamic texture detection analysis for automatic video-based fire[J].IEEE Transactions on Circuits&Systems for Video Technology,2015,25(2):339-351.
[7]Toreyin B U,Dedeoglu Y,et al.Computer vision based method for real-time fire and flame detection[J].Pattern Recognition Letters,2006,27(1):49-58.
[8]Gubbi J,Marusic S,Palaniswami M.Smoke detection in video using wavelets and support vector machines[J].Fire Safety Journal,2009,44(8):1110-1115.
[9]Listyorini T,Rahim R.A prototype fire detection implemented using the Internet of things and fuzzy logic[J].World Transactions on Engineering&Technology Education,2018,16(1):42-46.
[10]钟玲,张兴坤.基于SVM的视频图像火焰检测[J].软件工程,2017,20(6):1-4.
[11]Ernfjlletal.Fire detection system[J].Optoelectronics Instrumentation&Data Processing,2018,45(2):140-145.
[12]Sermanet P,Chintala S,Lecun Y.Convolutional neural networks applied to house numbers digit classification[C]//2012 21st International Conference on Pattern Recognition(ICPR 2012).IEEE Computer Society,2012.
[13]Springenberg T,Dosovitskiy A,Brox T,et al.Striving for simplicity:The all convolutional net[C]//Proceedings of International Conference on Learning Represenation Workshop Track,2015.
[14]Szegedy C,Vanhoucke V,Ioffe S,et al.Rethinking the inception architecture for computer vision[C]//2016 IEEEConference on Computer Vision and Pattern Recognition(CVPR).IEEE,2016.
[15]Huang U,Liu Z,Weinberger KQ,et al.Densely connected convolutional networks[C]//The IEEE Conference on Computer Vision and Pattern Recognition(CPVR).Las Vegas:IEEE,2017:4700-4708.
[16]Liu W,Anguelov D,Erhan D,et al.SSD:Single Shot MultiBox Detector[C]//European Conference on Computer Vision,2016:21-37.
[17]Xie S,Girshick R B,Dollar P,et al.Aggregated Residual Transformations for Deep Neural Networks[C]//2017 IEEEConference on Computer Vision and Pattern Recognition(CVPR),2017:5987-5995.