摘要
针对室内复杂环境下火灾识别准确率会降低的问题,提出了一种改进的粒子群算法优化支持向量机参数进行火灾火焰识别的方法;首先在YCrCb颜色空间进行火焰图像分割,对获得的火焰图像进行预处理并提取相关特征量;其次采用PSO算法搜索SVM的最优核参数和惩罚因子,并在PSO算法中加入变异操作和非线性动态调整惯性权值的方法,加快了搜索SVM最优参数的精度和速度;然后将提取的火焰各个特征量作为训练样本输入SVM模型进行训练,并建立参数优化后的SVM分类器模型;最后将待测试样本输入SVM模型进行分类识别;算法的火灾识别准确率达到94.09%,分类效果明显优于其他分类算法;仿真结果表明,改进的PSO优化SVM算法提高了火焰识别的准确率和实时性,算法的自适应性更强,误判率更低。
Due to fire detection is relatively low in the case of complex indoor environment, the proposed support vector machine （SVM） is applied to fire detection in the paper, among which an improved particle swarm optimization （PSO） is used to determine optimal parameters of support vector machine. Firstly, the obtained flame image will be processed ahead of time and extracted related feature quantity after flame image segmentation in YCrCb color space. Secondly, the optimal kernel parameter and penalty factor for support vector machine will be found by PSO algorithms, meanwhile, the ability of searching accuracy and speed of the optimal parameters of SVM are raised by adding mutation and nonlinear dynamic adjustment inertia weight in PSO algorithm; Then, each extracted flame characteristic parameters is re- served as training samples to train the SVM model, meanwhile, the SVM classifier model is established after the optimization of the parame ters. Finally, the test samples will be input the SVM model to classification and recognition. The accuracy rate of algorithm is 94.09%, and the classification effect is better than other algorithms. Simulation results show that the improved SVM algorithm optimized by PSO can enhance the accuracy and real-time performance of flame recognition, as the same time, the algorithm has better adaptability and lower false positive rate.
出处
《计算机测量与控制》
2016年第4期202-205,209,共5页
Computer Measurement &Control
基金
江苏省科技支撑计划项目(社会发展)(BEK2013671)
关键词
火焰检测
支持向量机
粒子群算法
参数优化
flame detection
support vector machine
PSO
parameter optimization
