Hyperspectral Image Change Detection Method Based on the Balanced Metric

Change detection, as a popular research direction for dynamic monitoring of land cover change, usually uses hyperspectral remote-sensing images as data sources. Hyperspectral images have rich spatial–spectral information, but traditional change detection methods have limited ability to express the features of hyperspectral images, and it is difficult to identify the complex detailed features, semantic features, and spatial–temporal correlation features in two-phase hyperspectral images. Effectively using the abundant spatial and spectral information in hyperspectral images to complete change detection is a challenging task. This paper proposes a hyperspectral image change detection method based on the balanced metric, which uses the spatiotemporal attention module to translate bi-temporal hyperspectral images to the same eigenspace, uses the deep Siamese network structure to extract deep semantic features and shallow spatial features, and measures sample features according to the Euclidean distance. In the training phase, the model is optimized by minimizing the loss of distance maps and label maps. In the testing phase, the prediction map is generated by simple thresholding of distance maps. Experiments show that on the four datasets, the proposed method can achieve a good change detection effect.