Abstract: In order to solve the high false probability caused by FCS, a new color space model IFCS based on chaos theory and k-medoids PSO algorithm is proposed in this paper. The flame recognition with IFCS can compute simply and quickly, and highlight the attribute differences between the flame pixels and non-flame pixels more than FCS. Particles are initialized by chaos sequence. Inertia weight is adjusted adaptively. Learning factors are nonlinear adjusted dynamically. Locally optimal solutions are avoided with chaotic searching. Then IFCS is obtained. Further, IOFR algorithm based on IFCS and Otsu is established with the classic Otsu method in IFCS fire color space. Results show that IOFR algorithm can reduce the false probability effectively.

Key words: flame recognition, fire color space, Particle Swarm Optimization（PSO）, chaos, k-medoids

摘要： 针对火焰识别颜色空间模型FCS带来的错判率高问题，基于混沌理论和[k]-medoids的粒子群算法，提出了一种改进的火焰识别颜色空间——IFCS。利用IFCS颜色空间进行火焰识别，可以在保证计算的简单快捷的同时，相对FCS更加突出火焰/非火焰像素颜色属性差异特性。采用混沌序列初始化粒子、自适应调整惯性权重、动态非线性调整学习因子、混沌搜索跳出局部最优等方法，得到了IFCS火焰识别颜色空间；进一步，在IFCS火焰颜色空间中通过经典Otsu阈值方法得到二值图像，建立了基于IFCS和Otsu的火焰识别算法——IOFR算法。实验结果表明：IOFR算法有效降低了当前基于FCS颜色空间火焰识别算法的火焰错判率。

关键词: 火焰识别, 火焰颜色空间, 粒子群算法, 混沌, k-medoids