Research On The Character Recognition System Of Cockpit Overhead Panel Based On DSP

The development of intelligent civil aviation has enabled more intelligent equipment to be deployed in the field of civil aviation.The artificial intelligence co-pilot can alleviate the shortage of aviation pilots and reduce the cost of training pilots,while also assisting in the execution of certain command operations.The cockpit characters information provide flight status as a basis for instructable operations.The A320 cockpit overhead panels are taken as the research object of character recognition in this dissertation,and through research on software algorithms and hardware platforms,a DSP-based cockpit overhead panels recognition system is designed.In this paper,an embedded platform character recognition system based on TMS320C6748 is designed.The system hardware is composed of DSP development board,camera and640*480 size LCD display.The system software is composed of image pre-processing,character detection,and character recognition.The image collected by the camera is transmitted to the memory by each frame.The core processor’s read and write character recognition operation of the memory data is written back to the frame cache address.Show to the screen.The system software uses two deep convolution algorithms,YOLOv3 and YOLO-Fastest,to detect the position information of the top plate characters respectively.Before character detection,pre-process image characters,including character image collection,noise overlay,character position annotation,and format conversion after annotation.YOLO-Fastest network structure adopts deep separable convolution,which reduces the amount of network parameters while effectively improving the detection time of the network,and the network model is called and partially rewritten by the computer platform,and the results show that the position of the characters in the image can be correctly detected and located,the detection time of the nerwork model for the input 352*352 image characters is about 31.10 ms.The CRNN deep learning algorithm is used for the detection and localization of character recognition.In the character characteristics,predict sequence output using a bidirectional LSTM structure.The traditional CTC loss function algorithm is often used in the field of voice recognition.The algorithm of the entropy loss function and the editing distance of ideas are used as the loss function of the model.The final result is output.The data set used in the character recognition training is used and the model is trained to identify.The results show that the character can be recognized.The recognition time of the network model for character images with an inpur size of 280*32 is about 34.31 ms.Finally,Initialize the camera and LCD.The download program went to the development board to test the camera and LCD.As a result,the video image collected by the camera can be displayed to the LCD,and the character can also be displayed on LCD.