摘要
超分辨率重建技术可以提高图像质量,使原图像具有更丰富的细节信息。针对现有的超分辨率重建算法存在提取特征单一、不利于对图像信息进一步提取的问题,提出了一种基于多尺度特征融合的超分辨率重建算法。采用多特征提取模块获取更多浅层信息,并在网络中添加密集连接结构,增强特征的传播,减少相关参数计算,减轻梯度消失问题。在Set5和Set14基准数据集上进行了测试,并在电力巡检数据集上进一步验证了算法的有效性。与主流的超分辨率重建方法进行了对比,实验结果表明,该方法生成的图像有更加丰富的细节信息,能够有效地改善图像质量,峰值信噪比与结构相似度值较其他主流算法均有一定的提高。
The super-resolution reconstruction technology can improve the image quality and make the original image have richer details.Aiming at the problem that the existing super-resolution reconstruction algorithm have a single extraction feature,which is not conducive to the further extraction of image information,a super-resolution reconstruction algorithm based on multi-scale feature fusion is proposed.The multi-feature extraction module was used to obtain more shallow information,and densely connected structures were added to the network to enhance the propagation of features,reduced the calculation of related parameters,and alleviated the problem of gradient disappearance.Experiments were carried out on the Set5 and Set14 benchmark datasets,and further verified the effectiveness of the algorithm on the power inspection datasets.Compared with mainstream super-resolution reconstruction methods,the experimental results show that the image generated by this method has richer detailed information,which can effectively improve the image quality.The peak signal-to-noise ratio and structural similarity values have a certain improvement compared with other mainstream algorithms.
作者
仝卫国
蔡猛
庞雪纯
翟永杰
TONG Wei-guo;CAI Meng;PANG Xue-chun;ZHAI Yong-jie(College of Control and Compute Engineering,North China Electric Power University,Baoding 071003,China)
出处
《科学技术与工程》
北大核心
2022年第26期11507-11514,共8页
Science Technology and Engineering
基金
国家自然科学基金(61773160)。
关键词
超分辨率重建
卷积神经网络
多尺度特征
密集连接
super-resolution reconstruction
convolutional neural network
multi-scale features
dense connection