研究目的
To detect changes occurring in a geographical area after a major event by processing a pair of bi-temporal remotely sensed images using a semi-supervised method.
研究成果
The proposed method effectively detects changes in geographical areas after major events by comparing bi-temporal images using a Siamese neural network trained with a novel approach for generating training data from unlabeled images. The method shows robustness to differences in optical and geometrical properties of input images.
研究不足
The method's performance is evaluated subjectively due to the lack of benchmark datasets. The approach may be sensitive to the quality and representativeness of the external images used for generating impostor patch-pairs.
1：Experimental Design and Method Selection:
The proposed method adopts a patch-based approach, comparing successive pairs of patches from input images using a Siamese neural network trained with augmented data.
2：Sample Selection and Data Sources:
Training patch-pairs are generated from transformed maps of the image taken before the event and external images from the Internet resembling the change.
3：List of Experimental Equipment and Materials:
Not explicitly mentioned.
4：Experimental Procedures and Operational Workflow:
Patch-pairs are generated and compared using the Siamese network to detect changes.
5：Data Analysis Methods:
The similarity between patches is measured using Euclidean distance in the feature space.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
