Research On Intelligent Recognition Method Of Embossed Characters Of Thermal Protector

Thermal protector is an important overheat and overcurrent protection device in the operation of electrical equipment.The code on its surface records important information such as the product model and production date.The identification and detection of the code is an indispensable link in product production.At present,the detection of imprinted characters of thermal protectors based on machine vision has the problems of low detection accuracy and low detection efficiency.This subject researches the problems existing in the existing detection system of the cooperative enterprise.Aiming at the characteristics of the thermal protector,which is made of metal material,small in size,contains multiple ineffective information circular holes,and has many interference factors,a set of machine vision is designed.Technical high-precision recognition system for characters on the surface of the thermal protector.The system focuses on solving three problems: one is that the image collected on the production line transmission device contains multiple thermal protectors,and the image characters are inclined,which is not conducive to subsequent processing;the second is that the surface of the thermal protector is complex,the metal surface is reflective and there are interference areas.The impact is calculated as character segmentation;the third is that the characteristics of imprinted characters are not obvious,and the recognition accuracy of traditional methods is not high.The specific work content of this paper is as follows:(1)The construction of the character recognition system for embossing on the surface of the thermal protector.First,the characteristics of the thermal protector are analyzed,and then the advantages and disadvantages of various cameras,lenses,and light sources are compared according to their characteristics,and the hardware suitable for the system is selected.Then the lighting scheme is designed,using LED as the lighting source,and using structured light lighting.Provide uniform illumination for the thermal protector,and install a polarizer,which effectively suppresses light reflection and highlights character features.(2)Aiming at the problem that the image contains multiple thermal protectors and the image is tilted,an image segmentation and correction algorithm is proposed.First,the minimum enclosing moment method is used to segment the image.This method searches the connected domains in the binarized image,records its horizontal and vertical boundaries,locates all connected domains with the minimum enclosing moment method,eliminates noise interference according to the set threshold,and separates a single thermal protector.Then for the segmented oblique image,a correction algorithm based on Radon transform is proposed.This method performs multiple downsampling processing on high-precision images,and limits the search range of Radon transform projection angles by column-by-column scanning,reduces the amount of calculation,and improves the recognition efficiency of the algorithm.And considering that the Radon transform cannot distinguish between inverted devices,the concept of duty cycle is introduced to solve the correction problem of inverted images.Experiments show that,in terms of correction speed,the improved Radon transform reduces the average running time by 0.95 seconds compared with the traditional Radon transform.(3)Aiming at the problem that the surface of the thermal protector is complex,which is not conducive to character segmentation and recognition,the method of suppressing the reflection and the method of eliminating the interference area are studied.First,according to the surface reflection characteristics of the thermal protector,the surface reflection suppression algorithm based on the gray-scale median is used to suppress the surface luminescence.Then,for the problem of the reduction of contrast caused by the suppression of reflection,the gray-scale histogram equalization method is adopted to improve The contrast between the character and the background area is determined,and then the interference area is eliminated through the improved Hough circle detection algorithm,the coding area is segmented,and the character segmentation is completed.Experiments show that the improved Hough circle detection algorithm saves 2/3 of the running time and significantly improves the recognition speed of the system.(4)Aiming at the problems of low recognition accuracy and slow speed in the field of imprinted character recognition in traditional image recognition algorithms,an imprinted character recognition method based on improved Le Net-5 is proposed.Different from the traditional Le Net-5 network,each convolutional layer of this network adopts a small-size convolution kernel to extract more features and speed up the training speed of the model;use the Inception V2 convolution module to replace the C5 fully connected layer,which can deepen the network Width,thereby improving the recognition accuracy of the network;abandon the fully connected layer F6,switch to the global average pooling layer,and select the Relu function with superior performance as the activation function,in order to reduce the training parameters and improve the training speed of the network.Experiments show that the recognition accuracy of the improved Le Net-5 convolutional neural network reaches 98.57%,which is about 4% higher than the traditional network.