Language-Guided Semantic Clustering for Remote Sensing Change Detection.

Area of Science:

• Remote Sensing
• Computer Vision
• Machine Learning

Background:

• Existing remote sensing change detection (RSCD) methods often use limited semantic information, leading to inaccurate change masks.
• Semantic-agnostic binary masks hinder the differentiation of various change types in RSCD.

Purpose of the Study:

• To develop a novel Language-guided Semantic Clustering framework for RSCD.
• To leverage the semantic richness of Contrastive Language-Image Pretraining (CLIP) models for improved RSCD accuracy.
• To address the limitations of semantic-agnostic supervision in current RSCD models.

Main Methods:

• A Language-guided Semantic Clustering (LSC-CD) framework is proposed.
• Utilizes CLIP's zero-shot generalization to transfer semantic knowledge.
• Employs a CLIP adapter module (CAM) and semantic clustering module (SCM) for aligning and clustering embeddings.
• A lightweight decoder generates the final change mask prediction.

Main Results:

• The LSC-CD framework effectively transfers semantic information from CLIP for RSCD.
• Achieved state-of-the-art performance on LEVIR-CD, WHU-CD, and SYSU-CD benchmarks.
• Demonstrated accurate change mask prediction with reduced noise.

Conclusions:

• The proposed LSC-CD framework significantly enhances RSCD performance by incorporating language-guided semantic clustering.
• CLIP's semantic transfer capabilities are effectively utilized for robust change detection.
• The method provides accurate and semantically rich change detection masks.