Remote Sensing Change Detection via a Spatial–Temporal Attention-Driven CNN–Transformer Architecture
Abstract
In this study, we examine the efficacy of a model in identifying variations in satellite imagery before and after event occurrences. Our evaluation of the model is based on its ability to detect and predict change areas by comparing pre-event and post-event imagery along with other performance measures, including precision, recall, F1-score, accuracy, and Intersection over Union (IoU). Confusion matrices and evaluation outcomes are used to determine the weaknesses and strengths of the model, especially with regard to false positives and false negatives. The findings demonstrate a high level of accuracy but also reveal room for improvement in recall and precision to achieve better results in identifying minor changes. The results provide insights into future improvements of change detection models in remote sensing applications.