基于深度置信网络的无人机航拍图像变化检测

Change Detection of Unmanned Aerial Vehicle Images Based on Deep Belief Networks

无人机(UAV)图像的变化检测对城市规划发展中的城市空间布局、土地利用信息具有重要的意义.针对航拍图像地面背景复杂,传统的变化检测方法鲁棒性低等问题,研究基于深度置信网络(DBN)的无人机航拍图像变化检测方法.利用DBN能够自动学习、提取特征的特点,将DBN应用到无人机图像的变化检测中.首先通过模糊C聚类联合算法对采集的图像进行预分类,并在此基础上进行DBN的逐层预训练;然后利用反向传播算法微调整个网络,从而使网络权值达到最优;最后,将该DBN模型应用于无人机图像的变化检测.实验结果表明,该方法能够有效地提取出图像的变化区域,检测准确率达到95%以上,该方法提高了图像变化检测的精度,实现了图像变化检测的智能化.

Change detection of unmanned aerial vehicle(UAV)image is of great significance to urban spatial layout and land use information in urban planning and development.A change detection method for aerial image of UAV based on deep belief network(DBN)is studied.to solve the proble of the complex ground background of aerial image and the low robustness of traditional change detection methods.DBN is applied to change detection of UAV image,which can automatically learn and extract features.Firstly,the collected images are pre-classified by the joint algorithm of fuzzy C-means clustering,and on this basis,the layer-by-layer pre-training of DBN is carried out.Then,the back-propagation algorithm is used to fine-tune the network so as to optimize the weight of the network.Finally,the DBN model is applied to the change detection of UAV images.The experimental results show that this method can effectively extract the change area of the image,and the detection accuracy reaches more than 95%.This method improves the accuracy of image change detection and realizes the intellectualization of image change detection.