深度学习的水下图像增强及目标检测算法研究

Underwater Image Enhancement and Object Detection Algorithm Based on Deep Learning

基于视觉的水下目标检测是自主水下机器人(Autonomous Underwater Vehicle,AUV)视觉感知系统中的关键技术,同时由于水下能见度低,干扰多,也是目标检测领域中的重难点之一。传统水下目标检测方法识别能力有限,识别率较低,速度较慢。本文采用深度学习方法对水下目标检测进行研究,以Aquarium数据集为例,以水底生物产品为检测对象,从图像增强和模型优化改进两个方面展开研究,建立并训练用于水下目标检测的模型,并对其进行轻量化处理,最终在PC平台验证图像处理和模型优化对于增强目标检测效率和识别率的作用。

Underwater object detection based on vision is the key technology in AUV visual perception system.At the same time,due to low underwater visibility and many interferences,it is also one of the key difficulties in the field of object detection.The traditional underwater object detection method has limited recognition ability,low recognition rate and slow speed.Therefore,the deep learning method is used to study underwater object detection.Taking the Aquarium dataset as an example,with underwater biological products as the detection object,the research is carried out from two aspects of image enhancement and model optimization.The model for underwater object detection is established and trained,and the lightweight processing is carried out.Finally,the effect of image processing and model optimization on enhancing the efficiency and recognition rate of object detection is verified on the PC platform.