Research On Method Of Reading Time Of Pointer Watch Based On Convolutional Neural Network

The watch can be used not only for display time,but also as a daily decoration,beautiful and portable.When people buy watches,they mainly choose between appearance and performance.Among them,performance is mainly measured by the accuracy of travel time.At present,the domestic watchmaker mainly relies on manual measurement of watches’ travel time accuracy.Watches are inspected for a long period and are easily influenced by the mood of the inspectors,which results in unstable results,which are not conducive to the mass production of enterprises.Therefore,it is urgent to design an automatic watches travel time accuracy detection instead of manual detection,which is conducive to improving the detection accuracy and is suitable for mass production of the company.The application of image processing technology in the field of automatic reading time of pointer instrument can replace manual reading time,save manpower cost and improve work efficiency.But there are also some problems such as inaccurate dial positioning,low pointer recognition rate and weak anti-interference ability.With the development of hardware and people’s exploration of knowledge,in-depth learning has become a hot topic and has been achieved breakthroughs in various fields.In the field of target detection,the accuracy of target detection based on convolution neural network is much higher than that of the algorithm mentioned above.Therefore,this paper studies the detection method of watch reading time based on convolution neural network.The main work includes the following aspects:1.System design.Hardware design is very important for obtaining clear and easily recognizable watch images and subsequent detection.The required pixels,lens focal length and light source requirements can be calculated according to the watch size,chassis characteristics,detection accuracy and other characteristic parameters.The software design includes the research of pointer reading time method,the flow chart of automatic reading time of pointer watch based on traditional image technology,and the concrete realization of reading time based on convolution neural network.2.Automatic Reading of Pointer Watch Based on Traditional Image Processing Technology.The automatic reading time is realized by several steps,such as pre-processing,target location determination,target segmentation and anglecalculation.The analysis of the experimental results shows that the traditional image processing technology can achieve automatic reading time detection,but there are also some problems,such as inaccurate positioning,low recognition rate,and the inability to achieve the required accuracy of enterprises.3.Aiming at the problem of reading time detection of pointer watch based on traditional image processing technology.Convolutional neural network algorithm for automatic reading of pointer watch is proposed.Firstly,it analyzes the structure of convolutional neural network,Then it briefly introduces the development process of a series of target detection algorithms from RCNN(Region Convolutional Neural Network)to Mask RCNN(Mask Region Convolutional Neural Network).Mask RCNN algorithm achieves instance segmentation and has positive significance for pointer recognition.Mask RCNN algorithm builds and trains convolutional neural network by annotating four classes of image: time-hand,minute-hand,second-hand and Logo.The test set is input into the convolution network to get the recognition results of the pointer and logo,and the coordinates to be calculated are output.Find logo center point and dial center circle point,connect these two points to get the central axis,and finally calculate the angle between the pointer and the central axis to get the reading time.Experiments show that the proposed method based on convolution neural network can solve the problems existing in traditional image processing technology,and has reference significance for the research of automatic reading of pointer watch.