基于生成对抗网络的无人机航拍输电线路图像关键信息提取

Extraction of key information fromTransmission Line Images Collected by UAV Based on Generative Adversarial Networks

基于Py Torch开发平台,搭建生成对抗网络模型,通过对无人机航拍图像进行预处理、标签制作、网络训练和算法改进,完成对输电线路特征信息的深度学习,固化生成网络参数,实现从无人机航拍输电线路关键信息的自动提取。在深度卷积神经网络的基础上,提出一种多生成器生成对抗网络模型,在生成网络间引入联同工作机制,以加快模型获取信息并减少参数量,并将Wasserstein距离引入模型的损失函数,避免生成对抗网络在训练过程中出现梯度消失、训练不稳定等问题。通过实验分析,证明该方法对利用高分辨率无人机图像提取输电线路信息具有较好作用。

Based on PyTorch,a Generative Adversarial Networks( GAN) model is build in this paper.Through preprocessing,labeling,network training and algorithm improvement of UAV aerial images,the deep learning of transmission line characteristic information is completed,and network parameters are generated,thereby realizing automatic extraction of key information from transmission line images collected by UAV.Based on deep convolutional neural network,a multi-generator generative adversarial networks model is proposed.A joint working mechanism is introduced between the generative networks to improve the information acquisition rate of the model and reduce the amount of parameters.The Wasserstein distance is introduced into the loss function of the model to avoid problems such as gradient disappearance and unstable training during the training process of the generative adversarial network.The results show that the method can be used as a reference for extracting transmission line information from high-resolution UAV images.