Low illumination images have insufficient local and global light exposure, loss of structural and detail information, and are prone to generating a large amount of noise. The overall image is grayish or even completely dark, and people often cannot recognize the content of the image with the naked eye. Image enhancement technology aims to enhance image brightness, adjust image contrast, restore hidden details in the Fark, and enhance the utilization value of images through corresponding technical means.The traditional low illumination image enhancement methods mainly focus on Histogram equalization and Retinex methods. Based on the drawbacks of the traditional methods, this paper studies the low illumination image enhancement algorithm based on Convolutional neural network, builds a mathematical model, and lays the foundation for subsequent experimental research and application.