Research And Application Of Fishing Boat’s Detection And Recognition Based On Deep Learning

Waterway transportation is one of the most important ways of transportation for the development of human civilization.In recent years,with the continuous development of my country’s marine economy and the continuous construction of marine digital cities,the number of boats used has also increased correspondingly.Generally,the port is the main gathering point for boats.With the increase in the number of boats,waterway traffic safety management issues such as boat collisions,fishing boats illegally fishing at sea,and boat safety management frequently occur.This makes the research on port boats such as identification and safe transportation of great value.At present,computer vision technologies such as image processing and deep learning have made a lot of progress in fields such as face recognition,OCR character recognition,and autonomous driving.However,the research and application of artificial intelligence technology in the field of waterway transportation is still lacking.The boat is the most important tool for waterway transportation,and the boat’s license plate name is the most important identity certificate of the boat.The character study of the boat’s brand name is conducive to the intelligent management of the boat,and provides convenience for the port of fishing boats and other waterways.This article has carried out a certain exploratory research on the subject of deep learning-based fishing boat brand recognition.The main work is summarized as follows:(1)Collection and processing of fishing boat data.At present,there is no publicly released data set of fishing boat brand.All the data on fishing boat brand are collected by the author himself through various channels,but the amount of data is relatively small.In order to train the model with supervised learning,the data collected are labeled semi-automatically in this thesis.(2)Target detection of fishing boat license plate characters.For the data of the fishing boat license plate,the YoloV3 algorithm and the DB algorithm are used to detect the target of the fishing boat license plate respectively.The output complexity of the DB algorithm is low;while the YoloV3 algorithm has multiple thresholds that are difficult to grasp when outputting the prediction frame,and it is easy to get wrong prediction results.The detection effect based on the YoloV3 algorithm is good,but there are still partial detection errors.And the detection result is a horizontal border,which does not perfectly fit the brand that is not placed horizontally in the image.The detection frame of the DB algorithm can fit well with the brands that are not placed horizontally in the image,and the detection effect is better than that of the YoloV3 algorithm.The accuracy and recall rates of the trained model based on the DB algorithm are 69.01% and 74.35%,respectively.Compared with other algorithms,the accuracy of this model is not satisfactory,and it still needs to be improved in the future.(3)Character recognition of fishing boat license plate.This article mainly uses the CRNN algorithm to recognize the characters of fishing boat brand.Compared with other character recognition algorithms,this algorithm has better performance,satisfies the basic needs of character recognition for boat license plate,and achieves a better recognition effect with an accuracy rate of 99.47%.It can be expected that more data will be collected in the future to train a more effective and universally applicable model for recognition.