研究目的
To introduce a scheme for performing multi-class change detection in remote sensing hyperspectral datasets by extracting features using Stacked Autoencoders (SAEs), combining multiclass and binary change detection to obtain an accurate multiclass change map.
研究成果
The proposed method for multiclass change detection in hyperspectral images, based on feature extraction using Stacked Autoencoders, effectively combines binary and multiclass change detection to achieve high accuracy. The use of SAEs for feature extraction outperforms other methods like PCA and NWFE, achieving up to 95.52% overall classification accuracy in the experiments.
研究不足
The study is limited to the evaluation on a specific hyperspectral dataset from the Hyperion sensor, and the performance might vary with other datasets. The computational efficiency and scalability of the method with larger datasets or higher dimensional data are not discussed.
