PSRNet: a phase-guided frequency-domain structure reconstruction network for RGB-T salient object detection.

INTRODUCTION

RGB-T salient object detection remains challenging because visible and thermal features often show structural shifts, weak thermal boundaries, and background interference.

METHODS

This study proposes PSRNet, a phase-guided frequency-domain structure reconstruction network. RGB and thermal features are decomposed into amplitude and latent-phase components in the Fourier domain. Reliable cross-modal structural cues are aligned through amplitude-weighted phase consistency, and boundary-related high-frequency responses are reconstructed with a bounded Gaussian high-pass gate before adaptive phase-modulated fusion and multi-scale supervision.

RESULTS

On VT1000, PSRNet achieved an S-measure of 0.903, an MAE of 0.028, and an F-measure of 0.794 at threshold 0.80, with a boundary IoU of 0.862.

DISCUSSION: The results indicate improved structural preservation and boundary recovery under challenging RGB-T conditions.