WUE-Net: a wavelet and EUCB enhanced framework for accurate building change detection in remote sensing imagery

Building change detection (BCD) in very high-resolution remote sensing imagery is crucial for urban planning and disaster management. However, existing methods struggle to detect subtle changes and preserve object boundaries in complex environments, primarily due to data noise, background similarity, model limitations, and insufficient detail handling. These factors collectively compromise the accuracy and robustness of change detection. We propose WUE-Net, a framework that combines Wavelet Convolution (WTConv) and Efficient Upsampling with Convolutional Blocks (EUCB). WTConv decomposes input features into multi-frequency bands, capturing high-frequency details and low-frequency patterns. EUCB enhances spatial fidelity during upsampling, ensuring reliable change detection. Experiments on benchmark datasets—LEVIR-CD, WHU-CD, and GZ-CD—show that WUE-Net achieves state-of-the-art performance, highlighting its robustness and efficiency for practical applications.