Research On Coal Train Number Location And Recognition Based On EAST And CRNN

Coal Train is the main tool for coal transportation in China.With the continuous increase of coal transport volume and the arrival of industrial 4.0,the demand for intelligent coal shipping is gradually increasing.The train number is the identity information of each carriage of the coal train,and automatic identification of train number is the key step of intelligent coal shipment.At present,manual copying is still used in the statistics of train numbers in coal mine loading stations,which not only takes a long time for personnel and equipment to work,but also is prone to missed detection and false detection.Therefore,the research on an automatic recognition method of coal train number based on computer vision is great significant to improve the coal turnover speed and realize the intellectualization of coal shipment and management.The character image of the coal train number is used as the research object of this paper,and the main research work is carried out from the following three aspects:(1)Aiming at the missing of data set of train number identification for coal transportation,the data set of train number location and recognition of coal transport train is prepared by manual labeling,which provides data guarantee for the training and testing of depth network.(2)Car number character location is the premise of recognition.Based on EAST(Efficient and Accuracy Scene Text)model,the feature extraction network and classification loss function are redesigned.At the same time,RefineNet multi-scale fusion strategy is introduced to solve the problem of character scale change of coal train number.In order to solve the problem of insufficient real data set,deep migration learning is carried out.(3)On the basis of the accurate location of the train number area,the recognition algorithm of the coal train number is studied.In view of the fact that it is difficult to segment the characters due to the adhesion and ambiguity of the characters of the coal train number,an end-to-end CRNN recognition algorithm is adopted,and the DenseNet with stronger performance is selected as the feature extraction part;Aiming at the problem of the tilt of some train number characters,the spatial transformation network is introduced to correct the train number features,and the expansion of the identification data set is expanded by adding data with typical interference characteristics to the benchmark data set.Combined with the above location and recognition algorithm of coal train number,the train number recognition system of loading station is developed according to the front and rear interactive framework,and the whole system is tested at the same time.The test results show that this method can effectively identify the characters of coal train numbers accurately and quickly.