Research And Application Of Multi-angle Visual Recognition Method For The Number Of Coal Transporting Carriage

As an important resource for the development of China,coal is mainly transported by railway.In order to facilitate the management of coal carriage,each coal carriage has its own special number.The traditional way of coal loading station registration is manual inspection and recording.Since the daily coal transportation in the loading station is relatively large,the manual inspection method not only takes a long time but also may wrongly register the vehicle number.This thesis uses unmanned aerial vehicle to collect images of coal carriages at the loading station,using digital image processing technology and lightweight convolutional neural networks.The main work of this thesis is as follows:(1)Positioning of the numbered area of the open coal carriageThe image acquired by the drone contains a complex background.This thesis first uses color information to locate the position of the coal carriage from the original image,and separates the image of the vehicle number area from the coal carriage image based on the relative position information of the vehicle number area and the coal carriage.Extract the maximum stable area and edge of the vehicle number image,enhance the edge image and calculate the stroke image,use the characteristics of the numbered character stroke width to aggregate the numbered characters into connected domains,and remove the connected domains of non-numbered characters,use heuristic rules to make the vehicle number characters form a text line,and filter the text line to complete the positioning of the carriage number area.Experimental results show that this method can accurately locate the numbered area for images acquired by UAVs from multiple angles.(2)Segmentation and recognition of number characters in coal carriagesIn view of the problem of the fracture of the vehicle number characters on the coal carriage,this thesis calculates the number of broken characters based on the results of the vertical projection and the actual number of vehicle numbers,and repairs the projection results according to the width information of the characters,which can correctly divide the broken numbered characters.Aiming at the problem that traditional visual recognition methods require artificial design features and low accuracy,this thesis uses lightweight convolutional neural network technology to build a vehicle number character recognition model.The lightweight convolutional neural network not only has fewer model parameters,but also has a high accuracy rate.This thesis produced a numbered character data set,built a numbered character recognition model based on the MobileNetV3-Small network,and compared the classification performance of the MobileNets lightweight network series in the numbered character data set,including the amount of parameters,accuracy,Multiplication calculation.Experiments show that the method in this thesis can accurately segment and recognize numbered characters.(3)Development of data processing system for patrol inspection along UAV railwayAiming at the problems of backward coal mine railway inspection methods,the large number of single-carriage carriages at loading stations,large workload and low efficiency,and backward data management methods,a data processing system for patrol inspection along the UAV railway was developed.The system consists of two parts: hardware and software.The hardware part includes industrial-grade drones and a ground station used to control the flight.The ground station can choose to manually and automatically control the trajectory of the drone while controlling the movement of the gimbal camera to complete data collection.The software part can complete the functions of collecting the collected original data,marking and querying the abnormal situation of the patrol inspection along the railway,identifying and registering the coal carriages in the loading station,and querying and statistics of the historical data.Experiments show that the system's human-computer interaction not only reduces the labor intensity of field staff,but also meets the real-time requirements of enterprises.This thesis studies the multi-angle recognition method and application of coal carriage number based on airborne vision.Complete the positioning of the numbered area through the maximum stable extreme value area algorithm and the stroke width conversion algorithm;the improved vertical projection method splits the numbered characters with breaks;build and train a lightweight model to identify the numbered characters of the coal carriage;developed a data processing system for patrol inspection along the UAV railway,which greatly improved the registration efficiency of coal carriages in coal loading stations.The thesis includes 66 figures,11 tables and 82 references.