MLM-LSTM: Multi-layer Memory Learning Framework Based on LSTM for Hyperspectral Change Detection

Hyperspectral change detection plays a critical role in remote sensing by leveraging spectral and spatial information for accurate land cover variation identification. Long short-term memory (LSTM) has demonstrated its effectiveness in capturing dependencies and handling long sequences in hyperspectral data. Building on these strengths, a multilayer memory learning model based on LSTM for hyperspectral change detection is proposed, called MLM-LSTM for hyperspectral change detection is proposed. It incorporates shallow memory learning and deep memory learning. The deep memory learning module performs deep feature extraction of long-term and short-term memory separately. Then fully connected layers will be used to fuse the features followed by binary classification for change detection. Notably, our model has higher detection accuracy compared to other state-of-the-art deep learning-based models. Through comprehensive experiments on publicly available datasets, we have successfully validated the effectiveness and efficiency of the proposed MLM-LSTM approach.