基于深度学习优化的图像高质量动态降噪算法

Image Quality Dynamic Noise Reduction Algorithm Based on Deep Learning Optimization

电子设备采集图像会受到光照、湿度等因素影响,从而产生噪声污染,影响图像的视觉效果.本研究以残差网络深度学习模型为基础,融合注意力机制构建图像动态降噪方法,采用深度可分离残差网络模型取代传统CNN网络来进行图像去噪的特征提取,同时在残差网络模型中加入了非局部注意力模块和通道注意力模块来优化去噪中的特征提取任务.实验结果表明,构建的模型参数具备更好的图像降噪特征提取效果,其PSNR(Peak Signal-to-Noise Ratio,PSNR)值最大为28.88.同时在不同的数据集性能测试中,AD-ResNet降噪模型在高斯噪声的处理中最高峰值信噪比为27.32,最高结构相似度为0.7826;而在泊松噪声处理中,最高峰值信噪比为26.61,最高结构相似度为0.7374.本文方法实现了高精确率和高效率图像降噪,具备有效的图像降噪处理性能.

Electronic devices may be affected by factors such as light and humidity when collecting images,resulting in noise pollution and damaging the visual effect of the images.This study is based on the residual network deep learning model and integrates attention mechanism to construct a dynamic image denoising method.This study uses a deep separable residual network model to replace traditional CNN networks for feature extraction in image denoising.At the same time,non-local attention modules and channel attention models are added to the residual network model to optimize the feature extraction task in denoising.In the dataset experiment,the model parameters constructed in this study have the best performance in image denoising feature extraction,with a maximum PSNR(Peak signal to noise ratio,PSNR)value of 28.88.In performance tests on different datasets,the highest peak signal-to-noise ratio and structural similarity of the AD ResNet denoising model in Gaussian noise processing were 27.32 and 0.7826,respectively.In Poisson noise processing,the highest peak signal-to-noise ratio is 26.61,and the highest structural similarity is 0.7374.The experiment shows that this research has achieved both high accuracy and efficiency,and has effective image denoising processing performance.