In order to solve the problems of low accuracy and poor performance in the field of image recognition in massive image recognition, this paper proposes an image denoising algorithm based on improved CGAN after optimizing and improving the traditional conditional generative adversarial network (CGAN) algorithm on the basis of the neural network model by using the least-squares loss function and adopting techniques such as attention pyramid network. After that, the loss entropy of convolutional neural network is optimized, and the model network and shared feature parameters are constructed by inductive migration algorithm in the model training stage to achieve accurate recognition of denoised images. The results show that the denoised image repaired by this paper’s algorithm has the best quality and structural similarity, and its PSNR and SSIM values are improved by 2.28dB and 0.124 respectively compared with those of the traditional CGAN model, which can significantly reduce the checkerboard effect phenomenon and improve the quality of the image. The improved convolutional neural network model achieves a correct rate of 99.47% for image recognition, and the recognition accuracy of 10 different types of images in the MNIST dataset is improved by 2.87% to 17.85% compared with the traditional convolutional neural network model. Comprehensive analysis shows that the model proposed in this paper has good application in both image denoising and recognition accuracy.
