| The ocean occupies nearly three-quarters of the earth’s surface area,and there are rich biological resources and mineral resources underwater.Underwater images are an important information carrier for understanding the ocean and developing marine resources.However,due to the absorption and scattering of light by water,the underwater image is distorted,and problems such as visual blur,color attenuation,and contrast drop occur,making underwater vision tasks difficult.This thesis studies the problems existing in underwater images,and uses the super-resolution method based on deep learning to reconstruct distorted underwater images.Finally,the underwater image reconstruction model proposed in this thesis is deployed to Android applications.Underwater images are affected by scattering and absorption.Scattering includes forward scattering and backscattering,which respectively cause image blur and contrast reduction.Absorption leads to overall blurring of underwater images.At the same time,absorption is selective,with strong absorption for red light and blue light.The weak absorption of green light causes the original color of the image to attenuate and appear blue-green.Aiming at the problem of image contrast degradation,this thesis uses the salient object detection method to determine the saliency boundary of important objects with the help of channel attention and spatial attention mechanisms,and calculates the average pixel intensity of the foreground relative to the background to help improve the contrast of the reconstructed image.For the problem of color attenuation,this thesis introduces a color matrix to measure the color loss to improve the performance of the model in color restoration.For the blur problem,this thesis introduces a more realistic blur kernel and noise to enhance the details of the reconstructed image.Considering that underwater sensor devices can be controlled and received by Android devices,this thesis deploys the reconstruction model to Android devices to improve the quality of images acquired by underwater devices.Experiments show that compared with the existing algorithms,the underwater image super-resolution reconstruction algorithm proposed in this thesis has better performance in contrast,color,and detail restoration,and the objective indicators have also improved. |
