MSFCR: MultiScale Fusion Method With Color Retention for Reference-Free Low-Light Enhancement in Underground Environments

To facilitate reliable monitoring in challenging underground environments, a novel low-light image enhancement (LLIE) technique, termed multiscale fusion method with color retention (MSFCR), has been developed to accurately and reliably acquire information. Initially, a nonlinear pseudomultiexposure sequence generator (NPMSG) is designed to nonlinearly map images to various luminance levels while preserving the original color and texture distribution of the images. Subsequently, a structural patch-based multiscale fusion with color retention (MFCR) approach is developed to integrate information from different images. The combination of these two components enables the integration of information across multiple luminance levels, thereby facilitating LLIE. Finally, comprehensive subjective and objective assessments of the MSFCR were conducted using three publicly accessible datasets and one underground low-light image dataset, achieving satisfactory results. Furthermore, the application of MSFCR in the underground conveyor belt foreign object detection task significantly improved the performance of all three prevalent detection models based on different architectures.