Research On Convolutional Neural Network-based Cable Character Recognition

With the development of the economy and the improvement of science and technology,people’s requirements on the machine automation and intelligent are getting higher and higher.Both in the production and common life,the application of automatic character recognition is becoming more and more extensive.Tiny cable’s specification information is recorded by the insignificant character on the insulating layer.Tiny cable is commonly used in the electrical equipment connection of helicopters,cars and other large machinery.When the machine is maintained and refurbished,it is necessary to obtain the character information on the cable,and use it to select and replace cable.Traditional methods of recognizing characters are inefficient and cannot meet the needs of people.Therefore,the desire to achieve automatic recognition of cable character is increasingly urgent.In this paper,a Convolutional Neural Network-based cable character recognition is designed for the characteristics of tiny cable’s character image.Tiny cable has the characteristics of small size,easy bending,low character contrast and non-uniform font structure.These features increase the difficulty of recognition of tiny cable’s characters.In this paper,the following algorithm is designed for these problems.Firstly,for the cable image’s edge bending,uneven lighting conditions and other issues,a combination of binarization method is designed for cable positioning.In this method,the Otsu algorithm is used to separate the cable from the background and obtain the coarse positioning of the cable to obtain the width of the cable image.After that,According to this width,set the template parameters for local binarization processing,which can achieve the precise positioning of the cable,and get the single background,complete character cable area image.The algorithm reduces the number of parameters that need to be set manually in the local binarization algorithm,and has high robustness.Then,the image segmentation algorithm based on the projection method is designed for the cable area image,and the character size is estimated by the voting method.The character size template is used to fine segment the character image after the rough segmentation.This algorithm solves most of the problem of character breakage and character stickiness.Finally,a character image recognition algorithm based on convolution neural network is proposed for the problem of low character contrast and font diversity.The algorithm uses a large amount of training data to extract features automatically and realize the classification and recognition of characters.Through a large number of experiments,the correct rate of the recognition accuracy is 97%.So,the character recognition scheme proposed in this paper has a high correct recognition rate,can realize the automatic recognition of cable characters,and has high practical value.