在光学相机远距离拍摄图像时，由于光线衰减和环境噪声的影响，图像容易变得模糊且难以清晰识别。为应对这一挑战，提出了一种基于权重注意力和密集残差连接的图像超分算法(image super-resolution algorithm based on weighted attention and dense residual connections，WADRNet)。首先，在网络的浅层特征提取阶段，提出一种非对称卷积模块，以替代传统的卷积模块，提高了模型的信息提取能力，尤其是对边缘和纹理等关键特征的提取;其次，采用密集残差结构，在不增加额外计算量的同时实现跨层特征传递和信息的有效利用，增强了模型的上下文特征提取能力，更好地还原图像;最后，在窗口注意力模块融入权重通道注意力模块，有效地利用全局感受野特性。实验结果表明，WADRNet在自制数据集上明显领先于其他模型，尤其在峰值信噪比和结构相似性等方面;同时，该模型在公开数据集上也表现出良好的效果。因此，该方法能够显著提升低分辨图像像素质量，在工程领域具有广泛的应用潜力和价值，尤其适用于需要远距离成像的应用场景。
Image super-resolution algorithm combining weighted attention and dense residual connections
When capturing images from a long distance with optical cameras,the image is prone to becoming blurry and difficult to recognize clearly due to the effect of light attenuation and environmental noise.To address this challenge,an image super-resolution algorithm based on weighted attention and dense residual connections(WADRNet)was proposed.Firstly,in the shallow feature extraction stage of the network,a asymmetric convolution block(ACB)was introduced to replace traditional convolution module,thereby enhancing the model's capability to extract features,especially for extracting key features such as edges and textures.Secondly,dense residual structures were employed to facilitate cross-layer feature transmission and efficient information utilization without increasing additional computational overhead,thereby enhancing the model's ability to extract contextual features and better restore images.Finally,weighted channel attention block(WCAB)was integrated into the window attention module,to effectively leverage global receptive field characteristics.Experimental results demonstrate that,WADRNet significantly outperforms other models on self-made datasets,particularly in terms of peak signal-to-noise ratio and structural similarity.Moreover,the model also exhibits good performance on public datasets.Thus,the approach can significantiy improve the pixel quality of low resolution images,showing application potential and value in various engineering fields,especially in applications requiring long-distance imaging scenarios.
万方数据
维普
