Research On Image Segmentation And Object Detection Algorithm Of Water Surface Cleaning Ship

More than two-thirds of the world’s surface is covered by water,of which less than 2.5% is available for human consumption.With the rapid development of society,a large amount of production and domestic waste is produced every day,and a large proportion of the waste enters rivers,lakes and seas.This fact makes it very important to deal with surface waste.At present,the method of cleaning water surface garbage is still mainly manual,or relying on artificially assisted robots.This method makes water surface cleaning expensive and inefficient.Using large machines to clean water surface garbage will damage the surface to a certain extent.existing ecosystem.In recent years,the rapid development of computer vision,the use of visual methods to monitor and clean the water surface is an effective way to solve this problem.According to the above situation and actual needs,this thesis proposes an improved image semantic segmentation and target detection algorithm based on deep learning for the vision system on the water surface cleaning vessel,and integrates the two algorithms.In this thesis,aiming at the problem of water surface segmentation,an image semantic segmentation network based on the encoding-decoding structure is proposed.By introducing the residual structure to alleviate the problem of gradient disappearance during the training process of the network,a new backbone feature extraction structure is adopted.The feature fusion between layers is enhanced,the extraction of context information is strengthened,and a variety of loss functions are added to stabilize the back-propagation process of the network.Experiments show that this method has excellent performance in water surface segmentation.Secondly,for the detection of objects on the water surface,an image object detection network with attention mechanism is proposed.Because the one-stage target detection algorithm solves the position prediction and category prediction of the target through one operation,this type of algorithm is not efficient for small target detection.Based on YOLOv5,this network adds CA attention mechanism,which increases the sensitivity of the network to small targets and improves the success rate of small target detection.Experiments show that the method has excellent performance in the positioning of water targets.Finally,the two networks mentioned above are fused,and the algorithm of this thesis is tested on the embedded platform.The experiments show that the method has better recognition effect and real-time performance.