Research On Underwater Plastic Waste Recognition And Location Technology

In recent years,as human activities have expanded,a large amount of human-made garbage,particularly plastic waste such as plastic bags and bottles,has appeared in inland water environments.Currently,the cleaning of plastic waste on inland water surfaces is mainly accomplished by mechanical equipment such as manual salvage ships,while the cleaning of underwater plastic waste is done through manual manipulation of mechanical arms,which is time-consuming and inefficient.The emergence of Autonomous Underwater Vehicles(AUVs)provides a new solution for underwater plastic waste cleaning.The realization of this task is closely related to whether the sensors carried on the AUV can accurately identify and locate underwater plastic waste.Therefore,this paper takes underwater plastic waste as the research object and based on underwater image enhancement,deep learning technology,and RGB-D cameras,designs and implements a vision-based method for underwater object recognition and positioning,which helps AUVs to achieve autonomous recognition,detection,and collection of plastic waste,and promotes the improvement of the ecological environment of inland water areas.The research work and innovation points of this paper are as follows:(1)In response to the low-quality underwater images captured by the camera,which suffer from degradation,blurriness,color distortion,and contrast imbalance,a weighted fusion-based underwater image enhancement algorithm is proposed.Compared with conventional image enhancement methods,this algorithm not only improves brightness,contrast,and color restoration but also enhances fine detail features,resulting in better overall quality.(2)Based on the YOLOv5 model,we designed a network framework that integrates CNN and Transformer.Mobile Vi T network was used as the backbone.We added a detection layer to improve the detection ability of small objects and introduced self-attention modules from Transformer and hybrid attention modules based on CNN to enhance the recognition ability of important features.CSP structure was introduced in the SPP part to enrich the feature information.Focal EIOU Loss was used to replacing CIOU Loss to accelerate convergence while improving regression accuracy.Experimental results show that the recognition accuracy of our algorithm can reach 0.913,ensuring a recognition speed of 45.56 fps/s,and the overall performance is superior to common object detection algorithms,enabling real-time detection.(3)Based on the open-source underwater garbage dataset Deep Trash,this study constructed the dataset used by adding self-captured and downloaded images of underwater plastic garbage and expanding them.(4)Combining RGB-D cameras with underwater detection and recognition tasks,we achieved the recognition and positioning of underwater targets.We calibrated and aligned the Intel Realsense D415 camera used in the experiment,and based on the camera’s intrinsic and extrinsic parameters obtained from calibration,we converted the pixel coordinates corresponding to the center point coordinates of the underwater target detection box from the pixel coordinate system to the camera coordinate system,obtaining the three-dimensional coordinates of the point based on the camera coordinate system,thus realizing the positioning of underwater targets.We validated the positioning accuracy through multiple experiments.