计算机科学
目标检测
水下
人工智能
渡线
计算机视觉
失真(音乐)
遗传算法
图像质量
对象类检测
适应度函数
过程(计算)
图像(数学)
接头(建筑物)
模式识别(心理学)
人脸检测
机器学习
操作系统
海洋学
地质学
工程类
面部识别系统
建筑工程
计算机网络
放大器
带宽(计算)
作者
Xiuwen Bi,Wei Guo,Fusheng Zha,Pengfei Wang,Yapeng Shi
出处
期刊:Applied Optics
[Optica Publishing Group]
日期:2025-07-07
卷期号:64 (22): 6479-6479
摘要
Underwater images often suffer from severe degradation issues such as color distortion and haze, significantly impacting object detection performance. Most existing approaches treat image enhancement and object detection as two separate tasks, first enhancing the image, then performing object detection. However, the enhancement process does not necessarily improve the detection accuracy due to the lack of effective interaction between these two tasks. To address this limitation, we propose, to our knowledge, a novel joint optimization framework for underwater image enhancement and object detection driven by a genetic algorithm (GA). The proposed framework consists of three key components: an underwater image enhancement module (UIEM), an object detection module (ODM), and a genetic optimization module. The UIEM improves image quality by correcting color distortion and removing haze, while the ODM performs object detection on the enhanced images and evaluates the detection performance using mean average precision (mAP). To optimize the synergy between enhancement and detection, we employ GA as the core optimization strategy, using mAP as the fitness function. Through evolutionary operations such as selection, crossover, and mutation, GA adaptively tunes the enhancement parameters to improve detection accuracy. Extensive experiments conducted on two public underwater datasets, URPC2019 and UTDAC2020, demonstrate that the proposed method significantly increases the detection performance of state-of-the-art detectors, including YOLOv5, Faster R-CNN, and RetinaNet, in challenging underwater environments.
科研通智能强力驱动
Strongly Powered by AbleSci AI