| Intelligent underwater robots can complete tasks such as seafood farming,seabed exploration,and pipeline cleaning on the seabed,keeping the staff away from the complex seabed environment and ensuring the safety of the staff.However,the seabed environment is very complex and changeable,and intelligent underwater robots need to carry out autonomous operations underwater.At this time,visual information becomes the key to underwater autonomous operations of underwater robots.Due to the complexity and danger of the underwater environment,the requirements for underwater image acquisition equipment are very high,and both manual and machine acquisition need to face higher challenges.These reasons lead to high acquisition cost of underwater datasets,and there are relatively few existing underwater image datasets.Moreover,the scattering and attenuation effects of light in underwater imaging will lead to color distortion and blurring of underwater images.This poor image quality not only affects the subsequent image processing tasks,but also brings serious problems to the operation of underwater robots.up the challenge.At the same time,the development of related technologies of intelligent underwater robots requires the joint efforts of researchers.At present,many domestic universities have carried out related studies and research on underwater robots,but the learning cost of underwater robots is very high,and water flow,equipment,weather,etc.need to be considered Affected by many factors,these factors lead to great difficulties and challenges in the learning and research of underwater robots.This paper focuses on underwater image segmentation.According to the characteristics of underwater environment,an image enhancement and segmentation algorithm is proposed,and the corresponding underwater robot simulation system software is developed.Specifically,this paper proposes an end-to-end underwater image enhancement algorithm based on segmentation drive for the problems of color distortion and blurring of underwater images.The back-end task of image enhancement,namely image segmentation,is used as the algorithm driver to assist the front-end image enhancement network.Training for underwater images of higher quality and better for image segmentation.Aiming at the scarcity of underwater image datasets,a semisupervised underwater image segmentation algorithm based on pseudo-labels and transformation consistency is proposed.This algorithm uses an image enhancement module based on a fusion strategy to enhance underwater images,and then utilizes The semi-supervised network is trained with pseudo-labels and consistency principles,and the performance of underwater image segmentation is improved by introducing more layers of perturbation and image enhancement algorithms.Aiming at the problems that the learning and research of intelligent underwater robots in colleges and universities are restricted by the experimental environment,weather and other factors,this paper designs and develops a simulation system for intelligent underwater robots based on ROS.The simulation system adopts the architecture of client and server.Finally,the core module is described in detail,and the completed application results of the simulation system are displayed. |
