基于FHN神经元自适应多通道随机共振图像增强  

Adaptive Multi-Channel Stochastic Resonance of FHN Neurons for Image Enhancement

作　　者：房涛 方琳灵 刘艳 潘树文 FANG Tao;FANG Linglin;LIU Yan;PAN Shuwen(School of Information&Electrical Engineering,Zhejiang University City College,Hangzhou Zhejiang 310015,China;Department of Rich Communication Suite China Mobile Hangzhou R&D Center,China Mobile Hangzhou Information Technology Co.,Ltd.,Hangzhou Zhejiang 311121,China)

机构地区：[1]浙大城市学院信息与电气工程学院,浙江杭州310005 [2]中国移动杭州研发中心融合通信系统部,浙江杭州311121

出　　处：《传感技术学报》2022年第11期1521-1529,共9页Chinese Journal of Sensors and Actuators

基　　金：教育部产学合作协同育人项目(202102019039);浙大城市学院培育基金项目(J-202223)。

摘　　要：为了解决图像受到采集设备以及外界噪声干扰,导致图像质量出现不同程度退化的问题,提出一种基于FHN神经元自适应的多特征并行通道级联随机共振图像增强方法。首先将原始输入图像分解为朝向、亮度以及亮度对比度等特征信息独立通道;然后构建基于遗传算法的自适应随机共振多特征并行通道对输入信息流并行处理机制,通过遗传算法求解每一个通道FHN神经元随机共振系统的最优参数,以实现特定噪声自适应最优FHN神经元参数设置,达到非线性FHN神经元、图像信号以及噪声之间的共振协同作用;最终对并行多通道的输出特征增强信息进行融合,再通过一个级联FHN神经元对融合数据做深度优化,实现FHN神经元非线性系统和带噪声输入图像的最佳匹配。以本实验室基于常规微生物学实验需要而采集的菌落图作为实验对象,采用仿自然光悬浮式暗视野系,以F/1.4大光圈镜头,对培养后皮氏培养皿上所生长的菌落进行了千万像素级的成像。通过与其他图像增强算法实验结果的对比,证明了本方法的有效性和鲁棒性,PSNR值保持在35以上,在保持菌落图像细节的基础上,对原图像具有很好的增强效果。通过利用随机共振机制,设计的多特征独立通道的自适应随机共振信息增强处理方法,为图像增强方法提供了新的思路以及后续的图像处理任务提供了基础。In order to solve the problem that the image quality is degraded by different degrees due to the interference of acquisition devices as well as external noise,an adaptive multi-feature parallel channel cascaded stochastic resonance image enhancement method with FHN neurons is proposed.Firstly,the original input image is decomposed into separate channels of feature information such as orientation,luminance,and luminance contrast.Then,an adaptive stochastic resonant multi-feature parallel channel based on genetic algorithm is constructed to process the input information stream in parallel,and the optimal parameters of each channel FHN neuron stochastic resonant system are solved by using genetic algorithm to achieve noise-specific adaptive optimal FHN neuron parameter settings to fultill resonant synergy between nonlinear FHN neurons,image signals and noise.Finally,the output feature enhancement information of parallel multi-channel is fused,and then the fused data are deeply optimized by a cascaded FHN neuron to achieve the best matching between the FHN neuron nonlinear system and the input with noise signal.The colonies which have grown on the Petri dishes after incubation and collected in our laboratory based on the needs of routine microbiology experiments are imaged at the megapixel level with a large aperture lens of F/1.4 using a natural light-like suspended dark field system.The effectiveness and robustness of this method are proved by comparing the experimental results with those of other image enhancement algorithms,and the PSNR value is kept above 35,which has a good enhancement effect for the original image while maintaining the details of the colony image.By using the mechanism of random resonance,the adaptive stochastic resonance information enhancement method with multiple feature independent channels provides a new idea for image enhancement methods and a basis for subsequent image processing tasks.

关 键 词：FHN神经元 随机共振 遗传算法 并行通道 图像增强 

分 类 号：TP212.3[自动化与计算机技术—检测技术与自动化装置]