Ablation Study of Multidimensional Attention and Multiscale Convolution on LEVIR-CD Dataset

Dissertation Results of Multidimensional Attention and Multiscale Convolution on the LEVIR-CD Dataset This research investigates the effectiveness of multidimensional attention and multiscale convolution on the LEVIR-CD dataset for change detection. The LEVIR-CD dataset is a benchmark dataset for change detection tasks, consisting of high-resolution aerial images acquired before and after construction projects. The proposed method utilizes multidimensional attention to capture long-range dependencies and contextual information, while multiscale convolution allows for the extraction of features at different scales. The results of an ablation study are presented, showcasing the impact of each component on model performance. The study demonstrates that the combination of multidimensional attention and multiscale convolution significantly improves change detection accuracy compared to traditional methods. The findings highlight the potential of these techniques for enhancing change detection models.