SMGNet: A Semantic Map-Guided Multitask Neural Network for Remote Sensing Image Semantic Change Detection

Abstract

Semantic change detection (SCD) aims to identify potential Earth surface changes, including their location and class, from multitemporal remote sensing images. However, the underdetection and pseudochange issues in existing SCD methods severally limit their effectiveness in diverse ground scenarios. To address these issues, a semantic map-guided network, namely, SMGNet, is proposed based on a multitask architecture designed to identify potential land-cover changes from bitemporal high-resolution remote sensing images. A robust feature extractor is first developed to extract multiscale contextual information while retaining fine-grained spatial details, thus enhancing the semantic representation of complex objects with irregular shapes and large sizes. To address the issue of underdetection, we integrate historical semantic information derived from pretemporal land-cover maps into the model using a semantic map encoder module. A semantic fusion module based on Bayesian theory is developed to highlight salient changed information, thus reducing pseudochanges caused by the same ground objects with spectra variations. Experimental results obtained in a public SCD dataset demonstrate the effectiveness of the proposed method in identifying various semantic changes. Results indicate that the proposed SMGNet achieved the highest detection accuracy, exceeding nine existing methods by 14.81%–41.28% and 8.45%–40.31% in terms of separated kappa (SeK) and $F1$ -score ( $F1_{\text {scd}}$ ) metrics on the high-resolution SCD (HRSCD) dataset, respectively. The proposed method effectively alleviated pseudochanges induced by spectra and temporal differences, and accurately detecting these changed objects with irregular shapes and large sizes. The detected results exhibited high interclass compactness and well-defined boundaries. Code and data are available at https://github.com/long123524/SMGNet