PCINet: A Prior-Guided Correlation Interaction Network for High-Resolution Remote Sensing Image Change Detection
TL;DR
PCINet, a prior-guided correlation interaction network designed to balance reliable change discrimination and computational efficiency, is proposed, demonstrating competitive performance and a favorable accuracy–efficiency trade-off across the three datasets.
Abstract
High-resolution remote sensing image change detection is important for land use monitoring, urban development assessment, and environmental observation. However, illumination differences, seasonal variation, and complex background variation can generate pseudo-change responses, making it difficult to preserve detection accuracy under lightweight computational constraints. This study proposes PCINet, a prior-guided correlation interaction network designed to balance reliable change discrimination and computational efficiency. PCINet is organized around three complementary principles: lightweight representation for efficient multi-scale modeling, correlation-guided temporal interaction based on a correlation-based change prior, and progressive refinement with deep supervision for spatial detail recovery. This problem-oriented design strengthens temporal consistency modeling without relying on a computationally intensive architecture. On the LEVIR-CD, SYSU-CD, and GZ-CD datasets, PCINet achieved F1 score values of 91.30%, 82.61%, and 88.62% and IoU values of 83.99%, 70.37%, and 79.57%, respectively. Under the reported evaluation setting, the model contains 7.93 M trainable parameters and requires 3.56 GFLOPs for a 256×256 bi-temporal image pair. With a batch size of one, automatic mixed precision, and evaluation mode on an NVIDIA RTX 4090 GPU (NVIDIA Corporation, Santa Clara, CA, USA), it processes 120.39 image pairs per second. These results demonstrate competitive performance and a favorable accuracy–efficiency trade-off across the three datasets.