面向遥感地类变化检测的U型深度学习神经网络改进方法

The improved method of U-type deep learning neural network for remote sensing in land type change detection

遥感影像的变化检测在调查监测等自然资源管理中有着广泛应用。针对样本库建设成本过高、深度学习算法困难等问题,本文提出了多时相变化检测方法,以改进影像变化深度学习检测。该方法将不同时相的数据作为不同波段信息进行融合,将变化发现任务转换为图像分割任务,将土地利用矢量数据作为标签数据用于模型训练,建设深度学习样本库。对原始的U型深度学习神经网络结构进行改进,加速模型训练。试验结果表明:①多时相变化检测方法有利于模型训练过程中学习更多的特征,提升了模型的特征提取能力,可得到更好的预测效果;②模型的查全率和查准率都有一定提升,整体预测效果明显提高。

The change detection of multi-temporal remote sensing image is widely used in natural resource management such as survey and monitoring.According to the high construction cost and deep learning algorithms difficult of the sample library,this paper proposes multi-temporal change detection method to improve image change deep learning detection.This method take multi-temporal images as different band for information fusion,and transform the change detection task into image segmentation task,use land use vector data as label for model training and build deep learning sample library.Improve the structure of the original U-type deep learning neural network,and accelerated model training.Experimental results show that:①Multi-temporal change detection method is conducive to learn more features during model training and improving the feature extraction capability of the model,and finally getting the better prediction effect;②The recall rate and precision rate of the model is improved in a certain degree,and the whole prediction effect is obviously improved.