A novel low light object detection method based on the YOLOv5 fusion feature enhancement

Area of Science:

• Computer Vision
• Artificial Intelligence
• Image Processing

Background:

• Low-light object detection presents significant challenges due to insufficient illumination and noise.
• Existing methods often struggle with accuracy and performance in dim environments.

Purpose of the Study:

• To develop an effective network, NLE-YOLO, for robust object detection in low-light conditions.
• To enhance the feature extraction capabilities of the YOLOv5 architecture for low-light scenarios.

Main Methods:

• Introduced NLE-YOLO, a modified YOLOv5 network incorporating novel modules: C2fLEFEM, AMC2fLEFEM, and AMRFB.
• C2fLEFEM integrates low-frequency enhancement and feature fusion for noise suppression and information preservation.
• AMC2fLEFEM utilizes attention mechanisms for multi-scale feature extraction and adaptability to brightness variations, while AMRFB expands receptive fields using atrous convolution.

Main Results:

• Experiments on the Exdark dataset demonstrated superior detection accuracy and performance compared to existing methods.
• The proposed network effectively suppresses high-frequency noise and enhances crucial target features.
• NLE-YOLO shows improved adaptability to varying brightness and better discrimination between targets and background.

Conclusions:

• NLE-YOLO offers a significant advancement in low-light object detection.
• The novel modules effectively address noise interference and feature extraction limitations in dim environments.
• The method provides a promising solution for real-world applications requiring reliable object detection under poor lighting.