Joint optimization technique of image enhancement and object detection based on a genetic algorithm for underwater...

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Underwater images suffer from color distortion and haze, degrading object detection.
• Current methods often separate image enhancement and object detection, limiting performance.
• Lack of effective interaction between enhancement and detection hinders accuracy.

Purpose of the Study:

• To propose a novel joint optimization framework for underwater image enhancement and object detection.
• To improve object detection accuracy in degraded underwater imagery.
• To leverage a genetic algorithm (GA) for adaptive parameter tuning.

Main Methods:

• A joint optimization framework integrating an underwater image enhancement module (UIEM) and an object detection module (ODM).
• Utilizing a genetic algorithm (GA) with mean average precision (mAP) as the fitness function to optimize enhancement parameters.
• Employing evolutionary operations (selection, crossover, mutation) for adaptive tuning.

Main Results:

• The proposed framework significantly enhances object detection performance on URPC2019 and UTDAC2020 datasets.
• State-of-the-art detectors (YOLOv5, Faster R-CNN, RetinaNet) show improved accuracy with the joint optimization.
• Demonstrated effectiveness in challenging underwater environments.

Conclusions:

• Joint optimization of image enhancement and object detection is crucial for underwater applications.
• The GA-driven framework effectively bridges the gap between enhancement and detection.
• This approach offers a robust solution for improving underwater object detection accuracy.