This page organizes papers related to artificial intelligence published around the world. This page is summarized using Google Gemini and is operated on a non-profit basis. The copyright of the paper belongs to the author and the relevant institution. When sharing, simply cite the source.
This paper addresses key challenges in remote sensing image change detection (CD)—the lack of high-resolution, comprehensive open-source datasets and the difficulty of achieving robust performance across diverse change types—by presenting a large-scale, sub-micron CD dataset, JL1-CD, consisting of 5,000 image pairs. Furthermore, we propose a novel Origin-Partition (OP) strategy and integrate it into the Multi-Teacher Knowledge Distillation (MTKD) framework to improve CD performance. The OP strategy partitions the training set based on the change coverage ratio (CAR) and trains specialized teacher models for each subset. The MTKD framework distills complementary knowledge from these teacher models into a single student model, achieving improved detection results across diverse CAR scenarios without additional inference overhead. The proposed MTKD approach was demonstrated in the 2024 "Jilin-1" Cup competition, winning first place in the preliminary round and second place in the final. Extensive experiments on the JL1-CD and SYSU-CD datasets demonstrate that it consistently improves the performance of CD models with diverse network architectures and parameter sizes, achieving new state-of-the-art results. The code and dataset are available at https://github.com/circleLZY/MTKD-CD .
