Target Recognition Based on Infrared and Visible Image Fusion and Improved YOLOv8 Algorithm

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Infrared and visible image fusion is crucial for enhanced situational awareness.
• Existing fusion methods are often susceptible to variations in lighting conditions.
• Robust feature extraction is essential for accurate target recognition.

Purpose of the Study:

• To propose an adaptive illumination perception fusion mechanism to mitigate lighting factor impacts.
• To integrate this mechanism into an infrared and visible image fusion network.
• To enhance the performance of target recognition networks trained on fused imagery.

Main Methods:

• Spatial attention mechanisms applied for feature extraction from infrared and visible images.
• Deep convolutional neural networks utilized for advanced feature information extraction.
• An adaptive illumination perception fusion mechanism integrated into image reconstruction.
• A Median Strengthening Channel and Spatial Attention Module (MSCS) incorporated into YOLOv8 backbone.
• A new dataset (ivifdata) created for training the target recognition network.

Main Results:

• The improved YOLOv8 network demonstrated significant enhancements in Recall (2.3%), mAP50 (1.4%), and mAP50-95 (8.2%).
• The fusion network effectively reduced the impact of lighting variations in fused images.
• The enhanced YOLOv8 network showed improved recognition rate and completeness.
• The system achieved a reduction in false negatives and false positives.

Conclusions:

• The proposed adaptive illumination perception fusion mechanism effectively addresses lighting challenges in image fusion.
• Integrating this mechanism into YOLOv8 significantly improves target recognition performance.
• The developed fusion network and dataset contribute to advancements in multi-modal image analysis and object detection.