Research On The Algorithm Of Channel Waters And Water Target Recognition Oriented To Ship Intelligent Perception

With the rapid development of modern ship intelligent perception system,it reduces the risk of ship operation and improves the safety of navigation.When ships navigate in densely trafficked narrow sea areas and inland waterways,there are high requirements for the monitoring of the hull itself and the communication between ships.In these application scenarios,ships can collect and transmit data and information to realize functions such as intelligence sharing,intrusion warning,and collision avoidance.With the advancement of video and image processing technology,researchers can easily capture data in related fields from pictures.More and more cutting-edge methods and technologies have emerged in the field of image processing,which has promoted the development of intelligent perception unmanned autonomous ships based on computer vision.The excavation and practical engineering application of the above technologies are inseparable from the accurate identification of the channel waters and the precise capture of the water surface targets.Constructing an image recognition algorithm with high precision and high recognition rate is the basis for video surveillance analysis in the field of shipbuilding and ocean engineering,and is also the key to realizing safe,efficient and reliable navigation of unmanned intelligent ships.This paper takes the development of intelligent sensing channel waters and water target recognition algorithm as the research goal,and the related research work is as follows:(1)Aiming at the problems of human error in the visual inspection method of original channel waters identification and difficulty in combining with unmanned intelligent ships,this paper studies the classic image segmentation algorithm and conducts visualization experiments to complete the identification of channel waters.Aiming at the limitations of the experimental results of traditional image segmentation algorithms,an improved image segmentation algorithm UNet based on deep learning is proposed,which improves the recognition accuracy of channel waters and reduces the parameters of UNet algorithm.The experimental results show that the improved UNet algorithm performs well in pictures and video images,and the loss value is reduced,and the average intersection over union ratio(MIoU)is increased by 15.63%,reaching the 97.18%;average pixel accuracy(MAP)value increased by 8.73%,reaching 98.57%.(2)In order to improve the speed and efficiency of channel waters recognition,the MobileNetV2 lightweight network was used to replace the backbone feature extraction network in the original PSPNet and DeepLabv3+image segmentation network algorithms,realizing the"slimming"of the two image segmentation networks,which greatly The parameter amount and computational complexity of the algorithm used in the channel water area identification process are reduced,and the algorithm training time is shortened.In the range of 1%loss of accuracy,the computing speed of the network is improved.The experimental results show that compared with the original algorithm,the parameters of the improved PSPNet algorithm are reduced from 46.707×10~6 to 2.376×10~6;the parameters of the improved DeepLabv3+algorithm are reduced from 54.714×10~6 to5.818×10~6.(3)In view of the problems of misdetection and ignorance of low-pixel small targets when unmanned smart ships are navigating,this paper collects water targets under various environmental backgrounds such as lighthouses,sailboats,and buoys through research and research.Image processing-related tools build data sets,use the Mixup data enhancement strategy to process the original data sets,optimize the structure of the YOLOv7 target detection algorithm,and nest the"attention mechanism"module to complete the recognition of water surface targets.The experimental results show that compared with the original YOLOv7 and the latest YOLOv8 target detection algorithm,the MAP of the optimized and improved algorithm is increased by 1.9%and 3.2%respectively,and the recognition accuracy of water surface targets is higher,and the detection ability of small targets is stronger.