ELGC-Net: Efficient Local-Global Context Aggregation for Remote Sensing Change Detection
This work addresses the challenge of efficient and accurate change detection in remote sensing for applications like environmental monitoring, offering a new state-of-the-art method with a lighter variant for resource-constrained settings.
The paper tackles the problem of accurately segmenting semantic change regions in remote sensing change detection by proposing ELGC-Net, which introduces an Efficient Local-Global Context Aggregator module to capture enhanced context with reduced computational cost, achieving a 1.4% gain in intersection over union on the LEVIR-CD dataset compared to ChangeFormer while reducing trainable parameters.
Deep learning has shown remarkable success in remote sensing change detection (CD), aiming to identify semantic change regions between co-registered satellite image pairs acquired at distinct time stamps. However, existing convolutional neural network and transformer-based frameworks often struggle to accurately segment semantic change regions. Moreover, transformers-based methods with standard self-attention suffer from quadratic computational complexity with respect to the image resolution, making them less practical for CD tasks with limited training data. To address these issues, we propose an efficient change detection framework, ELGC-Net, which leverages rich contextual information to precisely estimate change regions while reducing the model size. Our ELGC-Net comprises a Siamese encoder, fusion modules, and a decoder. The focus of our design is the introduction of an Efficient Local-Global Context Aggregator module within the encoder, capturing enhanced global context and local spatial information through a novel pooled-transpose (PT) attention and depthwise convolution, respectively. The PT attention employs pooling operations for robust feature extraction and minimizes computational cost with transposed attention. Extensive experiments on three challenging CD datasets demonstrate that ELGC-Net outperforms existing methods. Compared to the recent transformer-based CD approach (ChangeFormer), ELGC-Net achieves a 1.4% gain in intersection over union metric on the LEVIR-CD dataset, while significantly reducing trainable parameters. Our proposed ELGC-Net sets a new state-of-the-art performance in remote sensing change detection benchmarks. Finally, we also introduce ELGC-Net-LW, a lighter variant with significantly reduced computational complexity, suitable for resource-constrained settings, while achieving comparable performance. Project url https://github.com/techmn/elgcnet.