A Multi-Scale Enhancement Network for Object Detection in Low-Light Conditions

Object detection in low-light environments suffers significantly from poor image visibility, which impairs both accuracy and stability. To address this issue, we propose the MultiScale Feature Enhancement Network (MENet), designed to enhance detail and contrast in low-light images. MENet integrates advanced object detection algorithms, create a robust framework for low-light object detection. Our method utilizes a Laplacian pyramid for multi-scale image decomposition, incorporating a Detail Enhancement Module (DEM) and dual attention mechanisms to capture and enhance crucial features across various scales, thereby improving detection accuracy. By optimizing feature extraction and enhancement at multiple scales, MENet ensures comprehensive utilization and refinement of features across different resolutions through the DEM and dual attention mechanisms. Experimental results on the ExDark dataset reveal that our approach significantly surpasses existing low-light enhancement and object detection techniques, demonstrating superior accuracy and stability under challenging lighting conditions. This study introduces a novel solution for lowlight object detection, highlighting its potential for real-world applications and providing valuable insights for future research.