与以往两类单帧图像的超分辨率重建方法相比,卷积神经网络超分辨率(SRCNN)技术以其端对端的映射架构大幅提高了运行效率与复原精准度,然而网络的层数限制以及收敛性能使得部分图像的恢复效果不及基于样例的重建方法。针对网络优化问题,提出了一种将粒子群优化(PSO)算法与SRCNN相结合的方法,利用PSO算法对网络权重进行初始化,同时结合梯度下降(GD)算法对权值进行修正,使得PSO算法的全局搜索能力与GD算法的局部寻优能力相融合。分别对set5、set14数据集和雾霾天气下模糊图片进行对比实验,结果表明,所提算法不仅能以较少参数来获得较高性能的网络,其重建效果优于已有的4种算法,而且对边缘的锐化能力更强。 Compared with the super-resolution reconstruction methods of two types of single-frame images in the past, the convolutional neural network super-resolution (SRCNN) technology greatly improves the operation efficiency and the restoration precision with its end-to-end mapping architecture. However, the network layer limit As well as the convergence performance makes part of the image recovery less effective than the sample-based reconstruction method. Aiming at the problem of network optimization, a method combining particle swarm optimization (PSO) algorithm with SRCNN is proposed. PSO algorithm is used to initialize the network weights, and the weight is modified with the gradient descent (GD) algorithm to make the PSO algorithm The global search ability and the GD algorithm to optimize the ability to find the local fusion. The experiments on set5 and set14 datasets and fuzzy images under foggy and haze weather respectively show that the proposed algorithm not only achieves higher performance networks with fewer parameters but also has better reconstruction effects than the existing four algorithms, And sharpening edge more capable.