改进哈里斯鹰优化PCNN参数的图像融合应用

Image Fusion Application of Improved Harris Hawk Optimizing for PCNN Parameters

针对脉冲耦合神经网络(Pulse Coupled Neural Network,PCNN)参数较多且难以优化的问题,提出了一种改进哈里斯鹰算法优化PCNN参数的异源图像融合方法。首先,提出了一种混合种群增量学习的哈里斯鹰优化算法。在初始进化阶段,采用种群增量学习的方法来加强初始化种群在全局开发的能力,扩大了鹰群的搜索范围,使得更好地协调全局开发和局部开发;其次,在开发阶段,将算法原来的逃逸能量公式通过激励函数tanh非线性化,提高局部开采能力;然后将改进的算法用来探索PCNN的三个重要参数的最优值,采用最大化原则融合源图像。选用21个测试函数进行仿真实验,结果表明改进后的算法较原始算法和其他算法拥有更好的寻优性能和更高的精度。通过选用四组图像融合实验,在主观视觉方面相对图像亮度较原算法有一定的提升,在客观评价方面改进后的融合算法较原始的融合算法在多项指标均有提升,四组融合结果表明,在平均梯度、清晰度等四个指标均有提升。融合对比结果证明,该方法在除部分指标外,其余指标优于原始融合算法和其余对比算法。

Aiming at the problem that Pulse Coupled Neural Network(PCNN)has many parameters and is difficult to optimize,a heterogeneous image fusion method is proposed,which improves the Harris hawk algorithm to optimize PCNN parameters.Firstly,a Harris hawk optimization algorithm for incremental learning of mixed populations is proposed.In the initial evolution stage,the method of population incremental learning is used to strengthen the ability of the initial population to develop globally,expand the search range of the eagle colony,and better coordinate global development and local development.Secondly,in the development stage,the original escape energy formula of PCNN is nonlinearized by the excitation function tanh to improve the local mining ability,then the improved algorithm is used to explore the optimal values of the three important parameters of PCNN,and the maximization principle is used to fuse the source images.21 test functions are selected for simulation experiments,and the results show that the improved algorithm has better optimization performance and higher precision than the original algorithm and other algorithms.By selecting four sets of image fusion experiments,the relative image brightness in subjective vision has a certain improvement compared with the original algorithm.In terms of objective evaluation,the improved fusion algorithm has improved in many indicators compared with the original fusion algorithm.The four sets of fusion results show that the average gradient,clarity and other four indicators have improved.The fusion comparison results prove that the proposed method outperforms the original fusion algorithm and other comparison algorithms except for some indicators.