FusionOC: Research on optimal control method for infrared and visible light image fusion

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Single sensor types (infrared/visible light) have limitations.
• Traditional fusion lacks control and feedback, hindering performance.
• Precise perception of fusion task requirements and image quality is needed.

Purpose of the Study:

• To develop an adaptive image fusion model (FusionOC) with optimal control.
• To improve the relationship between fusion tasks, fused image quality, and source image features.
• To enhance target detection performance using fused imagery.

Main Methods:

• Established mathematical models for controlled objects affecting fused image quality.
• Integrated image fusion with a quality evaluation function to determine control factors.
• Designed proportional-integral-derivative (PID) control modes using backpropagation (BP) neural networks.
• Implemented adaptive regulation mode selection and a feedback mechanism for feature perception.

Main Results:

• FusionOC adaptively regulates fusion based on user/task requirements.
• The feedback mechanism improves feature perception and guides the fusion process.
• Demonstrated superior performance over advanced methods on public datasets.
• Showcased benefits of FusionOC in object detection tasks.

Conclusions:

• FusionOC offers improved controllability and feedback for image fusion.
• The model enhances generalization ability and intelligence for diverse fusion tasks.
• FusionOC provides significant advantages for object detection applications.