| Character recognition in specific scenes is important to research in the field of computer vision.It is widely used in real life,such as car license plates automatic identification system,signpost recognition in automatic driving systems,and automatic invoice identification system.With the continuous development of deep learning,it can achieve excellent performance in the implementation of related functions in practical applications.The design of automotive electronic harness diagram character recognition method is a practical application in computer vision research.The character information in the car electronic wire harness diagram is extracted by character recognition technology,and the character information is converted from image form to text form.This has provided key assistance for the design and maintenance of automotive electronic circuits and related systems for automotive maintenance assistance.This paper is based on the automotive circuit maintenance auxiliary system,as a subsystem of the electronics harness auxiliary maintenance system,to achieve the detection and recognition function of characters in the automotive electronics harness diagram.In practical application scenarios,traditional techniques are often used to implement image processing related tasks.Traditional algorithm implements the character detection by manually setting features,extracting the features according to the edges,colors,and textures of the image,and then using the algorithm such as template matching algorithm for character recognition,and finally implement the extraction of the text information in the diagram.However,the versatility of the manual setting features is poor,the traditional algorithm has some limitations,and there are problems such as low accuracy and low efficiency of detection and recognition.Therefore,this thesis innovatively uses the deep learning algorithm to design and implement the automotive electronics harness diagram character recognition and carries out research and design from two aspects of character detection and character recognition.The model trained by the deep learning algorithm has strong robustness and robustness,can solve the shortcomings of traditional algorithms,ensure the accuracy and efficiency of recognition,and better implement the detection and recognition of characters.Considering the traditional algorithm and deep learning algorithm used to realize character recognition,combined with analyzing the specific scenes realized in this thesis,selecting the deep learning algorithm to complete the task of character recognition of automotive electronic harness diagram.Determining the research content,which is using the deep learning algorithm to implement character detection and character recognition tasks for automotive electronic harness diagrams in two steps.The character detection and recognition network model is designed according to the character or text characteristics.The part of character detection is implemented by a neural network based on the multi-dimensional feature combination of the full convolutional neural network and performs end-to-end character detection,and directly positions the characters area without pre-adjusting the direction of characters.The part of characters recognition adopts the structure combined CNN with Bi LSTM,which makes full use of the context information of the characters sequence.Compared with the character recognition algorithm relying on segmentation,the accuracy of character recognition is improved,and the parameter quantity of the model is reduced.This thesis also adjusts the model and model parameters in combination with the specific scene of the characters in the automotive electronic harness diagram.In addition,in order to train the network model suitable for the scene of this paper,the character detection part collects various types of automobile electronic harness diagrams and manually labels the datasets for network training.Since the training of the sequence recognition model requires a large training set,a synthesis of 9 million-images with character is selected as a training set and a verification set in the character recognition section.The character image extracted by the 1000-images with character detection part is selected for manual labeling as a test set of the character recognition model.Finally,based on the Web framework of Flask,the overall implementation of the design of this thesis is demonstrated. |
