基于小波神经网络的激光主动成像视觉图像去噪方法  

Laser active imaging vision image denoising method based on wavelet neural network

作　　者：杨惠烽[1] 曹建芳[1] YANG Huifeng;CAO Jianfang(Xinzhou Normal University,Xinzhou Shanxi 034000,China)

机构地区：[1]忻州师范学院,山西忻州034000

出　　处：《激光杂志》2024年第11期123-127,共5页Laser Journal

基　　金：山西省自然科学基金项目(No.202203021221222)。

摘　　要：激光主动成像技术在许多领域中发挥着重要作用,然而,由于成像过程中会受到各种因素的干扰,导致图像产生噪声,影响后续的信息提取和处理,设计基于小波神经网络的激光主动成像视觉图像去噪方法。设计基于图像配准算法的超分辨率重构方法,将每个子区域的多帧光斑图像集中起来实施矫正图像间差异的处理。鉴于激光成像所捕获的彩色图像富含基原色,这导致了庞大的数据量和处理上的效率瓶颈。为了优化后续的预处理和识别流程,使用平均值法实施图像的灰度化处理。设计单隐层结构的小波神经网络结构,在输入层中仅设置一个节点,用以接收输入信息,在输出层中仅设置一个节点,负责输出处理后的结果,根据设计方法确定隐层节点数,样本数量取用于学习的图像像素数量值,实现激光主动成像视觉图像的去噪处理。实验测试结果表明,设计方法的去噪图像比较清晰同时保留了图像细节,range指标差异较小,去噪后图像的像素分布比较均匀。Laser active imaging technology plays an important role in many fields.However,due to the interference of various factors during the imaging process,images may generate noise,which affects subsequent information extraction and processing.Therefore,a laser active imaging visual image denoising method based on wavelet neural network is designed.Design a super-resolution reconstruction method based on image registration algorithm,which concentrates multiple frames of spot images in each subregion to correct the differences between images.Given that the color images captured by laser imaging are rich in primary colors,this leads to a huge amount of data and efficiency bottlenecks in processing.In order to optimize the subsequent preprocessing and recognition process,the average method is used to implement grayscale processing of the image.Design a wavelet neural network structure with a single hidden layer structure,with only one node set in the input layer to receive input information,and only one node set in the output layer to output processed results.Determine the number of hidden layer nodes according to the design method,and take the number of samples as the value of the number of image pixels used for learning,to achieve denoising processing of laser active imaging visual images.The experimental test results show that the denoised image of the design method is relatively clear while retaining image details.The difference in range index is small,and the pixel distribution of the denoised image is relatively uniform.

关 键 词：小波神经网络 激光主动成像视觉图像 超分辨率重构 图像配准 隐层 图像去噪 

分 类 号：TN249[电子电信—物理电子学]