火焰图像快速检测方法研究及应用

发布时间：2018-07-03 06:11

本文选题：火焰检测 + 二维模态　；参考：《西南科技大学》2017年硕士论文

【摘要】：火的使用标志着人类迈入文明世界,它在推动人类文明进步的同时也对人类的生命财产安全构成严重威胁。因此,如何尽早发现火灾并及时将火灾消灭在初期阶段对消防安全工作有着深远的意义。与传统的采用温度或烟雾传感器相比,视频火焰检测具有探测范围广、反应时间快、能够实时检测火情等优点。本文深入分析和研究了火焰的颜色区域提取、二维经验模态分解提取纹理特征等算法,针对存在的问题,基于支持向量机,结合聚类和闪频等特征,提出了新的快速火焰检测算法。(1)针对使用火焰纹理特征时,提取的有效纹理信息不佳,导致分类效果较低的问题,提出一种采用二维经验模式(Bidimensional Empirical Mode Decomposition,BEMD)和支持向量机(SVM)的火焰检测算法。首先基于累积差分法检测运动目标,根据Ohta颜色空间中火焰的颜色特征,找出图像中疑似火焰区域,其次将疑似火焰区域图像经过BEMD分解,结合局部二值模式(LBP)对所提取到的固有模态函数(IMF)图像进行纹理特征提取。最后将提取的纹理特征结合圆形度、矩形度、重心高度等特征,输入到SVM,进行火焰的判别。(2)针对火焰检测分割火焰区域不完整,导致漏检的问题,提出一种基于模糊C均值(FCM)结合火焰动态特征即闪频特征的火焰检测算法。首先使用累积差分提取出运动目标,然后采用结合爬山法(Hill-climbing algorithm)的FCM图像分割方法(HRFCM),结合火焰颜色特征分割出疑似火焰目标,不仅有效地缩短了时间,还克服了传统聚类算法初始聚类中心和聚类数目随机盲目地选择初始参数的问题。最后提取疑似火焰的闪频特征,使用SVM进行分类判别。实验结果表明,本文算法具有较好的鲁棒性和检测率,在提高检测率的同时,还提高了检测效率。
[Abstract]:The use of fire marks the entry of mankind into the civilized world. It not only promotes the progress of human civilization, but also poses a serious threat to the safety of human life and property. Therefore, how to find the fire as soon as possible and destroy the fire in the early stage is of great significance to the fire safety work. Compared with the traditional temperature or smoke sensor, video flame detection has the advantages of wide detection range, fast reaction time, real-time detection of fire and so on. In this paper, color region extraction of flame and texture feature extraction by two-dimensional empirical mode decomposition (EMD) are deeply analyzed and studied. Based on support vector machine (SVM), clustering and flicker are combined to solve the problems. A new fast flame detection algorithm is proposed. (1) when using flame texture features, the effective texture information extracted is not good, which leads to the low classification effect. A flame detection algorithm based on Bidimensional empirical Mode DecompositionBEMD and support Vector Machine (SVM) is proposed. Firstly, based on the cumulative difference method to detect moving targets and according to the color characteristics of the flame in Ohta color space, the suspected flame region in the image is found, and then the suspected flame region image is decomposed by BEMD. Local binary mode (LBP) is used to extract texture feature of the extracted intrinsic mode function (IMF) image. Finally, the extracted texture features are input into SVM to distinguish the flame by combining the features of roundness, rectangle and height of center of gravity. (2) aiming at the incomplete flame region of flame detection, it leads to the problem of missing detection. A flame detection algorithm based on fuzzy C-means (FCM) and flicker feature is proposed. Firstly, the moving target is extracted by cumulative difference, then the Hill-climbing algorithm (HRFCM) is used to segment the suspected flame target, which not only shortens the time effectively, but also combines the flame color features. It also overcomes the problem of random and blind selection of initial parameters in traditional clustering algorithms. Finally, the flicker feature of suspected flame is extracted, and SVM is used to classify and discriminate. The experimental results show that the proposed algorithm has good robustness and detection rate, and improves the detection efficiency as well as the detection rate.
【学位授予单位】：西南科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】