| The change detection technology is one of the most important research directions in the field of image processing,its goal is to analyze and identify the change information on the ground qualitatively and quantitatively through observing and analyzing the two images obtained in different time periods but in the same area.This technology can be widely used in many important fields,such as disaster assessment,land monitoring and so on.According to whether the image acquisition sources are the same or not,the image change detection problems can be divided into the homologous images change detection and the heterogenous images change detection.In recent years,the homologous images change detection technology has developed rapidly,and many effective algorithms have emerged.But with the continuous development of the image acquisition technology,these homologous images change detection algorithms can't meet the actual needs.The application of deep learning in the image processing field is becoming more and more mature,and it is more and more feasible to apply the deep learning to the problem of image change detection.Therefore,this paper proposes two heterogeneous images change detection algorithms based on deep learning after a full research on the traditional homologous images change detection algorithms and designs an images change detection system to solve practical problems.The main work of this thesis is as follows:(1)A heterogeneous images change detection algorithm based on feature learning and feature classification is proposed.This algorithm turns the change detection problem into the deep feature space through the deep neural network's powerful ability of feature learning.First,we construct the initial change detection result map,and then use stacked autoencoders to learn the deep features of the two images.Finally,we use the convolutional neural networks to classify those features according to the initial result map and get the final change detection result.This algorithm has a novel idea and don't need to produce a difference map.It can get the result map directly from the input images.The memory and time consumption have a certain advantage in case of a similar result compared to other algorithms.(2)A heterogeneous images change detection algorithm based on deep mapping relationship learning is proposed.This algorithm converts the heterologous images to the same feature space by the deep neural network's powerful ability of mapping.We can obtain the difference map through comparing the two mapping relations.Finally,the fuzzy c-means clustering algorithm is applied to get the change detection result.The algorithm does not need the training samples with labels to ensure the performance of the network during the learning process of the mapping relationship.The simulation results prove the effectiveness of this algorithm.(3)An image change detection system based on open source computer vision library is designed.This system integrates the common image change detection algorithms and implements the two heterogeneous image change detection algorithms given in the paper,and we add them as an important module to the system.The design of the system has built a bridge from academic problem study to practical problem application and help the researchers to get the change detection result and compare different algorithms quickly. |
