摘要
针对视频火焰图像识别问题,采取一种结合蚁群算法(Ant Colony Algorithm)优化的BP神经网络火灾火焰检测方法。该方法克服了传统神经网络容易陷入局部最优值和收敛速度慢的问题。使用混合高斯模型建立统计模型分割火焰图像。火焰的判别特征采用面积增加率、圆形度和火焰尖角数,并且各特征量作为神经网络的输入量来得到判别火焰的最终概率。通过对大量实验数据的分析,表明该算法在可接受的时间范围内能有效改善火焰识别的准确度。
For the visible flame detection technology,a BP neural network method optimized by the ant colony algorithm is adopted to detect fire in this paper.This method overcomes the disadvantage of falling into local minimum and slow convergence caused by neural network.The gaussian mixture model was used to build statistical model and to divide the fire image.The growth rate of area,roundness and flame angle numbers were adopted as the feature value of flame recognition.In addition,these values will also be the input quantity for the BP neural network.The analysis of experimental data indecates that the algorithm can effectively improve the flame recognition accuracy within acceptable time.
出处
《常州大学学报（自然科学版）》
CAS
2017年第2期65-70,共6页
Journal of Changzhou University：Natural Science Edition
基金
江苏省科技支撑计划项目(社会发展)(BEK2013671)
关键词
蚁群算法
BP神经网络
混合高斯模型
多特征融合
ant colony algorithm
BP neural network
gaussian mixture model
multi feature fusion
