基于LSGAN的水上光电图像生成算法研究

With the advancement of science and technology and economic development, maritime security has attracted much attention. In the field of sea area monitoring, the image feature information of sea area targets is obtained through water-based photoelectricity. Due to factors such as information security, such datasets are slow to be made public. In the real collection process, due to the limitations of the sea environment and other conditions, the number of ship images is small. Aiming at the problem of insufficient training data for target detection and re-identification algorithms in sea areas, this paper proposes a method to upgrade the discriminant network to multi-scale based on LSGAN for the purpose of expanding the water-based photoelectric image data set. By comparing with GAN and LSGAN, and analyzing from two aspects of visual interpretation and FID indicators, the generated image quality is high, which is close to the real picture, which proves the effectiveness of the image generation algorithm.