Open AccessChange Detection Method based on Block Similarity Measure
Author(s) -
Lie Yu,
Tao Sun
Publication year - 2019
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1237/2/022047
Subject(s) - artificial intelligence , computer science , speckle pattern , change detection , block (permutation group theory) , image (mathematics) , convolutional neural network , pixel , pattern recognition (psychology) , similarity (geometry) , speckle noise , computer vision , measure (data warehouse) , contrast (vision) , mathematics , data mining , geometry
Due to the random distribution of speckle noises in SAR images, the method based on direct pixel contrast cannot correctly judge the change of the pixels. In this paper, Convolutional Neural Network (CNN) is used to describe the multitemporal image blocks. Then the learned image block features are input into the decision network for further learning. The change detection of the whole image is completed on the basis of comparing the image blocks. The innovation of this method is that the input of the network is not the difference image produced by traditional methods, but the corresponding multitemporal image blocks. The output of the network is the judgment of the change between blocks. In addition, CNN is adopted to describe the features, which can extract the main features of the image blocks and is more robust to coherent speckle noises. This method has excellent performance in the accuracy of change detection.