Weather Recognition Method Of Inland Waterways Based On Visual Features

Knowing the real-time weather conditions on the planned route is essential for the safe navigation of the ship.For a long time,the main means for the crew to obtain the weather of the waterway is to collect the weather forecast data released by each weather station,but due to the low frequency of forecast release and the insufficient number of specialized weather observation stations arranged along the waterway with large spacing between stations,it is impossible to detect the bad weather in local waters early,which may lead to accidents such as ship capsizing,collision and grounding.Based on the existing inland river vessel video monitoring system,this paper researches the channel weather identification method based on visual features,designs and develops a channel weather identification and sharing system,identifies and shares the weather of all vessel navigation waters,upgrades all vessels into a mobile weather observation station,enables vessels to obtain real-time weather of the channel,and at the same time provides basic data for channel weather forecasting.Specifically,the following three main areas of work were accomplished.(1)A weather dataset for inland waterways was constructed.Through inland vessel monitoring video image acquisition and web crawler,a large number of inland waterway images were obtained,and then the images with high similarity were eliminated,and the images were divided into four categories of weather: sunny,rainy,foggy and cloudy by manual annotation,and a total of 8406 annotated inland waterway weather images were obtained.Then,the inland waterway weather dataset was enhanced using rotation,offset,flip,zoom,and angle transformation.(2)A two-channel Res2Net50 model-based weather identification method for inland waterways is proposed.Analysis of the inland waterway weather image features reveals that the image background contains most weather features and the target area contains fewer weather features.Therefore,in this paper,a two-channel Res2Net50 model for inland waterway weather recognition is constructed by semantic segmentation of waterway images through Mask R-CNN to obtain the image background,and then a twochannel Res2Net50 model for inland waterway weather recognition is constructed using the image background and the original image as input,and the model is trained and experimentally validated using the dataset.The experimental results show that the recognition accuracy of the method is as high as 93.65%.Compared with other methods,the accuracy rate is significantly improved.(3)A B/S architecture-based inland waterway weather identification and sharing system was designed and developed.The system implements functions including user login,news headlines,weather recognition,weather sharing,etc.,which can help vessel operators to obtain and share real-time weather data of the waterway.In this paper,the existing video monitoring system for inland vessels is used as the basis for application development,which realizes the function of automatic weather identification and weather sharing for vessels.Compared with the construction of weather service stations,this method has less investment and high efficiency,and is conducive to improving the navigation safety of inland vessels.