| Kitchen appliance control panel button function durability test is an essential test to verify the life of the product during the production of kitchen appliances.According to the requirements of the project partner for the function durability test of the control panel of kitchen appliances,the button function durability test task requires identifying the function and position of each button in the panel,and conducting repeated button tests in a certain order.Until the required number of tests is reached and the test is completed.However,at present,the appliance control panel button function durability test is still performed manually,which is time-consuming and inaccurate.In addition,it takes a lot of time to test the function durability of buttons,and human beings cannot work continuously for a long time like machines,which inevitably causes visual fatigue,resulting in leakage and mis-checking and which makes it difficult to ensure the authenticity of the test data.Through extensive observations,it is found that the buttons in the kitchen appliance control panel are usually composed of two elements: pattern and text annotation.The method of directly detecting the buttons by recognizing the button pattern is difficult and less adaptable,and belongs to the text recognition of complex scenes.Therefore,this paper proposes a character recognition algorithm for kitchen appliance control panel scenes,which belongs to a complex scene of text recognition.This algorithm can assist in completing the function durability test of the control panel buttons of kitchen appliances by recognizing the corresponding text and position of each button on the control panel,and combining it with the manipulator.The main work of this paper is as follows:(1)The kitchen appliance control panel dataset and the mixed Chinese character symbol dataset required for model training are established.(2)This paper proposes an improved Faster R-CNN-based algorithm for scene text target detection,which aims to exclude complex background interference and quickly identify text regions scattered in different areas of the control panel.In order to improve the capability of RPN feature characterization in Faster R-CNN,a deep region suggestion network(Deep Region Proposal Network,D-RPN)is proposed in this paper.In D-RPN,a feature reinforcement block is proposed with multi-scale convolution and feature fusion,which solves the problem of single scale of RPN convolution kernel and enables the network to obtain a richer set of feature information.In addition,D-RPN adopts a U-shaped network structure with several feature reinforcement blocks as nodes,which are connected by maximum pooling and upsampling operations to improve the learning ability of the network for image features.The experimental results show that the improved algorithm proposed in this paper has better recognition capability compared with the currently used target detection networks.(3)In this paper,a projection method with higher adaptability is used to segment the detected text regions,which performs better than the maximally stable extremal regions methods and connected region analysis methods.In addition,since there are many types of characters related to home appliances,a 9-layer convolutional neural network is built in this paper in order to accomplish the task of multiple classifications,which in turn enables the character recognition task.(4)In this paper,an automatic test system platform for the function durability of kitchen appliance control panel buttons is built.Firstly,the image is captured by the image acquisition unit.Then the function and position of the buttons are detected and recognized by the vision algorithm,and the pixel coordinates of the buttons are converted into the coordinates of the manipulator and finally transmitted to the manipulator through the RS485 serial port to complete the test. |
PDF Full Text Request