基于频域注意力的水下目标检测算法研究

Research on detection algorithm for underwater object based on frequency domain attention

摘要:

针对水下目标检测过程中由于水下成像模糊、目标物与背景对比度低等原因导致的水下图像特征提取与目标理解困难的问题,本文提出了一种基于频域注意力的水下目标检测算法。该方法首先将训练集图像变换到频域,并使用低频特征引导组件(low frequency feature guiding suite,LFGS)计算频率分量,然后该分量将作为参数被应用到低频特征提取模块(low frequency feature extraction model,LFM)来更好地提取图像的低频特征,融合了图像低频信息的特征经过进一步特征提取生成高层特征,最后将高层特征输入到检测头中进行检测。在URPC2021数据集上进行验证,平均精度均值达到了83.35%,验证了本文方法的有效性。

Abstract:

To solve the problem of feature extraction and target understanding in underwater image due to blurry underwater image and low contrast between target and background during underwater target detection,an underwater target detection algorithm based on frequency domain attention is presented.Firstly,the method transforms the training set image into the frequency domain,and uses the low frequency feature guiding suite (LFGS) to calculate the frequency component.Then the component will be applied as a parameter to the low frequency feature extraction model (LFM) to better extract the low frequency features of the image.The features that fuse the low frequency information of the image are further extracted to generate the high-level features.Finally,the high-level features are input into the detection head for detection.The average accuracy is 83.35% on URPC2021 dataset,which verifies the validity of this method.