基于数字图像处理的森林火灾识别方法研究
|
摘要
森林火灾是对社会、环境及经济发展影响范围最广、破坏性最大的灾害之一。为减少森林火灾的损失,世界各国都非常重视林火监测。传统的火灾监测方法是感烟、感温、感光探测器以及红外对射探测,还有现在比较流行的卫星监测。对于森林这样的大空间火灾监测,本文提出了一种基于图像视觉特征的火灾监测方法。
本文在分析了国内外火灾监测发展现状及先进技术的基础上,尝试将数字图像处理技术与BP神经网络相结合,应用于森林火灾监测。具体内容和研究成果如下:
(1)依据概率论和信号抽样原理,提出对原始图像进行抽样提取的概念,在不损失图像信息的情况下,减少了数据量。
(2)在充分分析各种常用的图像分割技术的基础上,提出了一种结合图像的RGB分量和HIS分量适合于火焰图像分割的方法。
(3)依托数字图像处理技术,对火灾图像依次进行阈值化分割、灰度化、二值化、中值滤波。通过理论分析及对比实验,找到使火焰图像预处理效果最好的算法组合方案;
(4)本文阐述了火灾特征量的提取原则,由火焰燃烧的内外因素及人工经验,选定形状特征,颜色灰度方差及动态特征值三种特征作为火焰模式识别特征;
(5)研究火焰特征提取算法,并在此算法基础上开发了火焰特征提取软件系统,各部分算法按照模块化思想进行设计,保证了源程序的可扩展性;
(6)在采集并分析处理大量森林火灾图片基础上,以VC++6.0为软件开发平台,基于matlab6.5设计建立火灾BP神经网络模式识别系统,训练BP神经网络并进行网络测试。
Forest fire is one of the most devastating disasters for the social,environment and economic development.In order to reduce the loss of forest fire,every country attaches great importance to the monitoring of forest fire around the world.Traditional methods for fire detection are smoke detection, temperature detection,light & infrared detection,and satellite monitoring.For such a large space of forest fire monitoring,a fire detection method based on the visual characters of image is proposed.
Based on the analysis of development of monitor to forest fire home and abroad and advanced technologies,this paper is trying to combine digital image processing technology with BP neural network pattern recognition theory,and applying them to the fire detection process.
(1) Based on probability sampling theory and signal theory,propose a method to sample from original image.It can reduce the amount of data,without losing the information of image.
(2) Based on fully analyzing the common method and theory of image segmentation,propose a suitable method to fire detection,which is based on RGB and HIS.
(3) The fire images were processed by the following steps:threshold of segmentation,median filter and gray level transformation,which rely on the image processing technology.Seek the best method through theoretical analysis and experimental contrast.
(4) This paper described the principle of fire feature extraction and analyzed the characteristic quantity which is common used to be extracted and the factors of fire formation,thus shape feature, color feature and dynamic characterstics were chosen as fire pattern characteristics to be the multidimensional input of the BP neural network.
(5) Based on the algorithms of fire feature extraction,the software of fire feature extraction system is developed.For ensuring the expansibility of the source code,each algorithm was designed by modular programming.
(6) According to numerous specific fire images which were collected and analyzed a fire BP neural network pattern recognition system was established by carrying out Matlab engine calling the functions provided by Matlab neural network toolbox in VC++6.0 development environment.The feasibility of the system could be proved from three areas which included the recognition accuracy and minimum resolution.
本文在分析了国内外火灾监测发展现状及先进技术的基础上,尝试将数字图像处理技术与BP神经网络相结合,应用于森林火灾监测。具体内容和研究成果如下:
(1)依据概率论和信号抽样原理,提出对原始图像进行抽样提取的概念,在不损失图像信息的情况下,减少了数据量。
(2)在充分分析各种常用的图像分割技术的基础上,提出了一种结合图像的RGB分量和HIS分量适合于火焰图像分割的方法。
(3)依托数字图像处理技术,对火灾图像依次进行阈值化分割、灰度化、二值化、中值滤波。通过理论分析及对比实验,找到使火焰图像预处理效果最好的算法组合方案;
(4)本文阐述了火灾特征量的提取原则,由火焰燃烧的内外因素及人工经验,选定形状特征,颜色灰度方差及动态特征值三种特征作为火焰模式识别特征;
(5)研究火焰特征提取算法,并在此算法基础上开发了火焰特征提取软件系统,各部分算法按照模块化思想进行设计,保证了源程序的可扩展性;
(6)在采集并分析处理大量森林火灾图片基础上,以VC++6.0为软件开发平台,基于matlab6.5设计建立火灾BP神经网络模式识别系统,训练BP神经网络并进行网络测试。
Forest fire is one of the most devastating disasters for the social,environment and economic development.In order to reduce the loss of forest fire,every country attaches great importance to the monitoring of forest fire around the world.Traditional methods for fire detection are smoke detection, temperature detection,light & infrared detection,and satellite monitoring.For such a large space of forest fire monitoring,a fire detection method based on the visual characters of image is proposed.
Based on the analysis of development of monitor to forest fire home and abroad and advanced technologies,this paper is trying to combine digital image processing technology with BP neural network pattern recognition theory,and applying them to the fire detection process.
(1) Based on probability sampling theory and signal theory,propose a method to sample from original image.It can reduce the amount of data,without losing the information of image.
(2) Based on fully analyzing the common method and theory of image segmentation,propose a suitable method to fire detection,which is based on RGB and HIS.
(3) The fire images were processed by the following steps:threshold of segmentation,median filter and gray level transformation,which rely on the image processing technology.Seek the best method through theoretical analysis and experimental contrast.
(4) This paper described the principle of fire feature extraction and analyzed the characteristic quantity which is common used to be extracted and the factors of fire formation,thus shape feature, color feature and dynamic characterstics were chosen as fire pattern characteristics to be the multidimensional input of the BP neural network.
(5) Based on the algorithms of fire feature extraction,the software of fire feature extraction system is developed.For ensuring the expansibility of the source code,each algorithm was designed by modular programming.
(6) According to numerous specific fire images which were collected and analyzed a fire BP neural network pattern recognition system was established by carrying out Matlab engine calling the functions provided by Matlab neural network toolbox in VC++6.0 development environment.The feasibility of the system could be proved from three areas which included the recognition accuracy and minimum resolution.
引文
[1].程鑫,王大川,尹东良.图像型火灾火焰探测原理[J].火灾科学,2005,14(4):239-243.
[2].崔屹.图像处理与分析-数学形态学方法及应用[M].北京:科学出版社,2000:67-76.
[3].范华忠,张伯虎,冯艳,等.图像处理技术在火焰目标提取中的应用[J].电光与控制,2006,13(1):99-101.
[4].高凯.立小树干弯曲度自动识别方法的研究[D].2007.
[5].高凯,李文彬,阚江明,等.基于计算机视觉的立木树干曲率自动测量方法[J].北京林业大学学报,2007,18(1):11-14.
[6].冈萨雷斯.数字图像处理[M].北京:电子工业出版社,2002.
[7].霍宏涛.数字图像处理[M].北京:机械工业出版社,2003.
[8].何东健.数字图像处理[M].陕西:西安电子科技大学出版社,2003.
[9].洪子泉.统计模式识别中的特征提取[J].数据采集与处理,1991,6(2):38-43.
[10].贾永红.数字图像处理[M].湖北:武汉大学出版社,2003.
[11].金华彪,夏雨人,张振伟.数字图像处理在火灾探测技术领域的应用[D].微型电脑应用,2006,19(5):25-27.
[12].阚江明,李文彬,孙仁山,等.基于计算机视觉的立木枝干直径自动测量方法[J].北京林业大学学报,2007,29(4):6-7.
[13].李婷.浅谈火灾探测中的图像分析.信息技术.2006,5(7):35-37.
[14].李丽.基于LVQ神经网络的锯材表面缺陷特征提取方法研究[D].2008.
[15].黎燕,赵望达,甘方成.基于VC6.0的火火试验构件图像处理系统[J].基础自动化,2002,9(2):56-58.
[16].雷艳敏,黄秋元.基于数学形态学的图像边缘检测[J].武汉理工大学学报,2005,27(5):25-27.
[17].刘卫国.MATLAB程序设计教程[M].中国水利水电出版社,2005.
[18].刘晓明,仲元红,欧静兰.基于DSP的火灾图像识别系统设计及应用[J].北京:计算机技术与发展,2006,16(6):95-100.
[19].刘二林.基于神经网络的摄像机标定[J].计算机应用研究,2006.
[20].任晓强.智能楼宇遮阳控制系统的研究[D].北京:北京林业大学控制理论与控制工程,2007.
[21].冉海潮.火灾图像的模糊识别探测方法[J].北京:模糊系统与数学,2001,11(3):102-104.
[22].苏金明.MATLAB实用教程[M].电子工业出版社,2005.
[23].史杰,孙丽华,刘力辉.基于神经网络的火灾探测系统[J].微计算机信息,2000,16(3).
[24].史海山,吕厚余,仲元红.基于遗传神经网络的火灾图像识别及应用[J].计算机科学,2006,1(33):232-235.
[25].孙宏琦,施维颖,巨永锋.利用中值滤波进行图像处理[J].长安大学学报(自然科学版),2003,23(2):104-106.
[26].武林,彭复原,赵坤.基于运动特征分析的森林火灾烟火实时检测方法[J].信号处理,2006,22(2):260-263.
[27].闻新,周露等.MATLAB神经网络仿真与应用[M].北京:科学出版社,2003.
[28].王亚军,徐大芳,陈向成,等.基于夫焰图像动态特征的火灾识别算法.测控技术.2007:7-9
[29].王刚,刘立柱.基于神经网络的模糊模式识别算法研究[J].信息工程大学学报,2006,7(1):54-56.
[30].谢志冰,费敏锐.基于CCD图像的自动定位灭火系统的开发[J].北京:自动化仪表,2007,28(7):67-69.
[31].闫厚,倪金生,等.基于网络摄像机的森林火灾烟火自动识别技术探讨[J].林火监测,2007,93(2):25-26.
[32].余成波.数字图像处理及MATLAB实现[M].重庆:重庆大学出版社,2003.
[33].姚海根.彩色图像转灰度图像技巧[J],印刷杂志,2007,03:52-55.
[34].尹力.一种基于视频多特征融合的火焰识别算法[J].西安:西安交通大学学报,2006,40(7):811-814.
[35].杨淑莹.VC++图像处理程序设计[M].北京:清华大学出版社,2003:138-139.
[36].杨晖,曲秀杰,等.图像分割方法综述[J].电脑开发与应用,2007,18(3):21-22.
[37].杨行峻.人工神经网络与盲信号处理[M].北京:清华大学出版社,2003.
[38].詹福如,徐琼,苏国锋等.基于CCD成像的多角度激光散射探测技术[J].南京:火灾科学,2002,11(1):52-55.
[39].张乃尧,阎平凡.神经网络与模糊控制[M].北京:清华大学出版社,1998.
[40].章毓晋.图像分割[M].北京:科学出版社,2001:2-3.
[41].周小四,杨杰,朱一坦.用于监控智能报警系统的图像识别技术[J].上海交通大学学报,2002,36(4):498-501.
[42].周长发.精通Visual C++图像编程[M].北京:电子工业出版社,2000.
[43].赵建华,方俊,疏学明.基于经网络的火灾烟雾识别方法[J].光学学报,2003,23(9):1087-1089.
[44].Bezdek J C.A Physical interpretation of Fuzzy ISODATA[C].IEEE trans Systems Man Cybern,1976,6(1):23-24.
[45].Dazing L,Pham,Jerry L.Prince.An Adaptive Fuzzy C-Means Algorithm for Image Segmentation in the Presence of Intensity In homogeneities[J].Pattern Recognition Letters,1999,20(1):57-68.
[46].Eric den Breejen,Marcel Breuers,Frank Cremer.autonomous forest fire detection.14~(th)Conference on Fire and Forest Meteorology.1998.
[47].Girosi F,Poggio T.Neural Network and the Best Approximation Property[J].Biologcal Cybemetics,1990,63:169-176.
[48].Ghaleb Faour.Forest fire hazard mitigation in Lebanon using remote sensing and GIS.Space Technology for Sustainable Development.UN/Morocco/ESA.April 2007.
[49].Kohir VV,Desai UB.Face recognition[A].IEEE International Symposium on Circuits and Systems.2000,(05):305-308.
[50].Mayan,Renhong e.The research of the peeling log six points centering theory[J].International Conference on Effective Utilization of Plantation Timber,1999,9:70-74.
[51].Meng.Joo Er.Face Recognition With Radial Function Neural Networks[J].IEEE Transaction on Neural Nerwork,2002,3(3).
[52].Malladi,R.,Sethian.Shape modeling with front propagating:a level set approach[J].IEEE Transaction on Pattern Analysis and Machine Intelligence, 1995,17(2): 158-175.
[53]. N.R.PaI, J.C.Bezdek. Measuring fuzzy uncertainty[C]. IEEE,Fuzzy System, 1994,2(2): 107-118.
[54]. Postaire, R.D.Zhang, Lecoco-Botte, Cluster Analysis by Binary Morphology[C]. IEEE Transaction On Pattern And Machine Intrlligence,1993,15(2): 170-180.
[55]. Russ Berhart, Pat Simpson. Computational Intelligence PC Tools [M]. Academic Press, 1996.
[56]. Sivind Due Trier,Anil K.Jain,Torfinn Taxt.Feature extraction methods for character recognition on a survey[J].Pattern Recognition. 1996,29(4): 641-662.
[57]. Sohoo P.K et al. A Surrey of Thresholding Techniques[J]. Computer Vision Graphics Image Process, 1988,41:233-260.
[58]. Tian Xin, Glen Murphy. Detection of Trimmed and Occluded Branches on Harvested Tree Stems using Texture Analysis [J]. Journal of Forest Engineering, 1997,8(2): 65-78.
[59]. V Caselles, R Kimmel. Geodesic active contours [J]. UCV,1997,22:61-79.
[60]. Venkatesh S, Kitchen L. Edge Evaluation Using Necessary Components [J]. CVCIP. Orlando,FL,USA. 1992.54(1): 23-30.
[61]. Xu C,Prince J L. Snakes, shapes and gradient vector flow[J]. Forest Products journal, 1988, 38 (1):23-26.
[62]. Yao X L, Zhang D S. Adaptive modification of the rolling force prediction[J]. Journal of Beijing Institute of Technology,2002,11(1): 3-36.
[63]. Zobel M. Robust facial feature localization by coupled features, Proceedings[A]. Fourth IEEE International Conference on Automatic Face and Gesture Recognition, 2000, 6(03): 2-7.
[2].崔屹.图像处理与分析-数学形态学方法及应用[M].北京:科学出版社,2000:67-76.
[3].范华忠,张伯虎,冯艳,等.图像处理技术在火焰目标提取中的应用[J].电光与控制,2006,13(1):99-101.
[4].高凯.立小树干弯曲度自动识别方法的研究[D].2007.
[5].高凯,李文彬,阚江明,等.基于计算机视觉的立木树干曲率自动测量方法[J].北京林业大学学报,2007,18(1):11-14.
[6].冈萨雷斯.数字图像处理[M].北京:电子工业出版社,2002.
[7].霍宏涛.数字图像处理[M].北京:机械工业出版社,2003.
[8].何东健.数字图像处理[M].陕西:西安电子科技大学出版社,2003.
[9].洪子泉.统计模式识别中的特征提取[J].数据采集与处理,1991,6(2):38-43.
[10].贾永红.数字图像处理[M].湖北:武汉大学出版社,2003.
[11].金华彪,夏雨人,张振伟.数字图像处理在火灾探测技术领域的应用[D].微型电脑应用,2006,19(5):25-27.
[12].阚江明,李文彬,孙仁山,等.基于计算机视觉的立木枝干直径自动测量方法[J].北京林业大学学报,2007,29(4):6-7.
[13].李婷.浅谈火灾探测中的图像分析.信息技术.2006,5(7):35-37.
[14].李丽.基于LVQ神经网络的锯材表面缺陷特征提取方法研究[D].2008.
[15].黎燕,赵望达,甘方成.基于VC6.0的火火试验构件图像处理系统[J].基础自动化,2002,9(2):56-58.
[16].雷艳敏,黄秋元.基于数学形态学的图像边缘检测[J].武汉理工大学学报,2005,27(5):25-27.
[17].刘卫国.MATLAB程序设计教程[M].中国水利水电出版社,2005.
[18].刘晓明,仲元红,欧静兰.基于DSP的火灾图像识别系统设计及应用[J].北京:计算机技术与发展,2006,16(6):95-100.
[19].刘二林.基于神经网络的摄像机标定[J].计算机应用研究,2006.
[20].任晓强.智能楼宇遮阳控制系统的研究[D].北京:北京林业大学控制理论与控制工程,2007.
[21].冉海潮.火灾图像的模糊识别探测方法[J].北京:模糊系统与数学,2001,11(3):102-104.
[22].苏金明.MATLAB实用教程[M].电子工业出版社,2005.
[23].史杰,孙丽华,刘力辉.基于神经网络的火灾探测系统[J].微计算机信息,2000,16(3).
[24].史海山,吕厚余,仲元红.基于遗传神经网络的火灾图像识别及应用[J].计算机科学,2006,1(33):232-235.
[25].孙宏琦,施维颖,巨永锋.利用中值滤波进行图像处理[J].长安大学学报(自然科学版),2003,23(2):104-106.
[26].武林,彭复原,赵坤.基于运动特征分析的森林火灾烟火实时检测方法[J].信号处理,2006,22(2):260-263.
[27].闻新,周露等.MATLAB神经网络仿真与应用[M].北京:科学出版社,2003.
[28].王亚军,徐大芳,陈向成,等.基于夫焰图像动态特征的火灾识别算法.测控技术.2007:7-9
[29].王刚,刘立柱.基于神经网络的模糊模式识别算法研究[J].信息工程大学学报,2006,7(1):54-56.
[30].谢志冰,费敏锐.基于CCD图像的自动定位灭火系统的开发[J].北京:自动化仪表,2007,28(7):67-69.
[31].闫厚,倪金生,等.基于网络摄像机的森林火灾烟火自动识别技术探讨[J].林火监测,2007,93(2):25-26.
[32].余成波.数字图像处理及MATLAB实现[M].重庆:重庆大学出版社,2003.
[33].姚海根.彩色图像转灰度图像技巧[J],印刷杂志,2007,03:52-55.
[34].尹力.一种基于视频多特征融合的火焰识别算法[J].西安:西安交通大学学报,2006,40(7):811-814.
[35].杨淑莹.VC++图像处理程序设计[M].北京:清华大学出版社,2003:138-139.
[36].杨晖,曲秀杰,等.图像分割方法综述[J].电脑开发与应用,2007,18(3):21-22.
[37].杨行峻.人工神经网络与盲信号处理[M].北京:清华大学出版社,2003.
[38].詹福如,徐琼,苏国锋等.基于CCD成像的多角度激光散射探测技术[J].南京:火灾科学,2002,11(1):52-55.
[39].张乃尧,阎平凡.神经网络与模糊控制[M].北京:清华大学出版社,1998.
[40].章毓晋.图像分割[M].北京:科学出版社,2001:2-3.
[41].周小四,杨杰,朱一坦.用于监控智能报警系统的图像识别技术[J].上海交通大学学报,2002,36(4):498-501.
[42].周长发.精通Visual C++图像编程[M].北京:电子工业出版社,2000.
[43].赵建华,方俊,疏学明.基于经网络的火灾烟雾识别方法[J].光学学报,2003,23(9):1087-1089.
[44].Bezdek J C.A Physical interpretation of Fuzzy ISODATA[C].IEEE trans Systems Man Cybern,1976,6(1):23-24.
[45].Dazing L,Pham,Jerry L.Prince.An Adaptive Fuzzy C-Means Algorithm for Image Segmentation in the Presence of Intensity In homogeneities[J].Pattern Recognition Letters,1999,20(1):57-68.
[46].Eric den Breejen,Marcel Breuers,Frank Cremer.autonomous forest fire detection.14~(th)Conference on Fire and Forest Meteorology.1998.
[47].Girosi F,Poggio T.Neural Network and the Best Approximation Property[J].Biologcal Cybemetics,1990,63:169-176.
[48].Ghaleb Faour.Forest fire hazard mitigation in Lebanon using remote sensing and GIS.Space Technology for Sustainable Development.UN/Morocco/ESA.April 2007.
[49].Kohir VV,Desai UB.Face recognition[A].IEEE International Symposium on Circuits and Systems.2000,(05):305-308.
[50].Mayan,Renhong e.The research of the peeling log six points centering theory[J].International Conference on Effective Utilization of Plantation Timber,1999,9:70-74.
[51].Meng.Joo Er.Face Recognition With Radial Function Neural Networks[J].IEEE Transaction on Neural Nerwork,2002,3(3).
[52].Malladi,R.,Sethian.Shape modeling with front propagating:a level set approach[J].IEEE Transaction on Pattern Analysis and Machine Intelligence, 1995,17(2): 158-175.
[53]. N.R.PaI, J.C.Bezdek. Measuring fuzzy uncertainty[C]. IEEE,Fuzzy System, 1994,2(2): 107-118.
[54]. Postaire, R.D.Zhang, Lecoco-Botte, Cluster Analysis by Binary Morphology[C]. IEEE Transaction On Pattern And Machine Intrlligence,1993,15(2): 170-180.
[55]. Russ Berhart, Pat Simpson. Computational Intelligence PC Tools [M]. Academic Press, 1996.
[56]. Sivind Due Trier,Anil K.Jain,Torfinn Taxt.Feature extraction methods for character recognition on a survey[J].Pattern Recognition. 1996,29(4): 641-662.
[57]. Sohoo P.K et al. A Surrey of Thresholding Techniques[J]. Computer Vision Graphics Image Process, 1988,41:233-260.
[58]. Tian Xin, Glen Murphy. Detection of Trimmed and Occluded Branches on Harvested Tree Stems using Texture Analysis [J]. Journal of Forest Engineering, 1997,8(2): 65-78.
[59]. V Caselles, R Kimmel. Geodesic active contours [J]. UCV,1997,22:61-79.
[60]. Venkatesh S, Kitchen L. Edge Evaluation Using Necessary Components [J]. CVCIP. Orlando,FL,USA. 1992.54(1): 23-30.
[61]. Xu C,Prince J L. Snakes, shapes and gradient vector flow[J]. Forest Products journal, 1988, 38 (1):23-26.
[62]. Yao X L, Zhang D S. Adaptive modification of the rolling force prediction[J]. Journal of Beijing Institute of Technology,2002,11(1): 3-36.
[63]. Zobel M. Robust facial feature localization by coupled features, Proceedings[A]. Fourth IEEE International Conference on Automatic Face and Gesture Recognition, 2000, 6(03): 2-7.