Underwater Motion Deblurring Based on Cascaded Attention Mechanism

The images captured in the underwater scene frequently suffer from blur effects due to insufficient light and relative motion between scenes imaging system, which severely hinders visual-based exploration investigation of ocean. In this article, we propose a feature pyramid attention network (FPAN) remove restore blurry images. FPAN incorporates cascaded modules into network, enabling it learn more discriminative information. To facilitate training FPAN, construct weighted loss function, consists content loss, an adversarial perceptual loss. module function enable our proposed generate realistic high-quality addition, deal with lack publicly available datasets image deblurring, built two specific deblurring datasets, namely Underwater Convolutional Deblurring Dataset Multiframe Averaging Dataset, train examine different deep learning-based networks. Finally, conduct sea trial experiments on autonomous vehicle. Experimental results demonstrate that method achieves satisfactory results, validates potential practical values real-world applications.