Zero-Reference Multi-Scale Feature Fusion Network for Monitoring Drone Low-Light Enhancement
Zhugang Chu, Yang Wang, K. Liu, Baisong Wang, Qingtao Chen, Kai Ming Wu · 2025
Monitoring drones are increasingly being used for high-altitude tasks. However, complex lighting conditions, especially low-light and backlight environments, often significantly affect the quality of the images captured by the camera. Image enhancement under such complex lighting conditions is a critical task in the field of drone image processing. Although numerous enhancement methods have been proposed, existing methods often suffer from the degradation of fundamental features during the enhancement process, which impacts image quality and reduces the effectiveness of task execution. To solve these problems, we propose a zero-reference multi-scale feature fusion network (ZMSF-Net) specifically for low-light image enhancement in monitoring drones. By adopting an enhancement matrix generation network and a multi-scale feature fusion strategy, it gradually improves image brightness and ensures that key features are retained at different scales. Extensive experimental results demonstrate that ZMSF-Net achieves significant improvements in both subjective visual quality and objective metrics, particularly in terms of brightness adjustment, color fidelity, and feature preservation, outperforming existing methods.