近年来, 随着科学技术的高速发展, 深度学习的蓬勃兴起, 实现图像超分辨率重建成为计算机视觉领域一大热门研究课题. 然而网络深度增加容易引起训练困难, 并且网络无法获取准确的高频信息, 导致图像重建效果差. 本文提出基于密集残差注意力网络的图像超分辨率算法来解决这些问题. 该算法主要采用密集残差网络, 在加快模型收敛速度的同时, 减轻了梯度消失问题. 注意力机制的加入, 使网络高频有效信息较大的权重, 减少模型计算成本. 实验证明, 基于密集残差注意力网络的图像超分辨率算法在模型收敛速度上极大地提升, 图像细节恢复效果令人满意.
In recent years, with the rapid development of science and technology and the rise of deep learning, achieving image super-resolution reconstruction has become a hot research topic in the field of computer vision. However, the increase in network depth is easy to cause training difficulties, and the network cannot obtain accurate high-frequency information, resulting in poor image reconstruction. This study proposes an image super-resolution algorithm based on residual dense attention network to solve these problems. The algorithm mainly uses residual dense network, which accelerates the model convergence speed and reduces the gradient vanishing problem. The addition of attention mechanism makes the high-frequency effective information of the network have a larger weight and reduces the model calculation cost. Experiments show that the image super-resolution algorithm based on residual dense attention network greatly improves the model convergence speed, and the image detail recovery effect is satisfactory.