基于生成对抗网络的遥感样本生成方法

A remote sensing samples generation method based on generative adversarial networks

针对扩充建筑垃圾遥感图像样本集问题,本文基于Wasserstein对抗生成网络,在网络整体流程中加入结合自适应直方图均衡化与图像融合的对比度调整方法,并对网络内部模型生成器与判别器模型进行调整。经原始网络与改进网络生成图像试验后,使用CS-LBP算子与颜色直方图进行评估分析。试验证明,经过改进的WGAN网络生成图像有显著的纹理特征与颜色特征,均方根误差降低22.6%,更符合原始图像的特征分布,边界明显色块鲜明,符合现实世界遥感图像特点。可生成大量样本进行后续的基于大样本的建筑垃圾识别研究。

To expand the sample set of remote sensing image of construction waste,this paper proposes a method of sample expansion of remote sensing image of construction waste based on Wasserstein Generative Adversarial Networks.In this paper,a contrast adjustment method combining adaptive histogram equalization and graph fusion is added to the whole network process,and the internal model generator and discriminator model of the network are adjusted.After the experiments of original and improved networks,CS-LBP operator and Color Histogram are used to evaluate and analyze the generated images.Experiments show that the improved WGAN network generates pictures with significant texture and color features,and the root mean square error is reduced by 22.6%.Moreover,it is more in line with the feature distribution of the original image,the boundary is obvious,and the color block is distinct in line with the characteristics of remote sensing image in the real world.This method can generate a large number of samples for subsequent research on building waste identification based on large samples.