Research On Recognition Algorithm Of Pointer Meter Reading Based On Deep Learning

Aiming at the problems of robustness and poor versatility of existing pointer meter reading recognition algorithms,this paper studies the key steps in the algorithm,and proposes a new set of pointer meter reading recognition algorithms based on deep learning.The main steps include Image correction,image polar coordinate transformation and instrument representation calculation.The main research contents of the full text include:(1)Image correction method based on scale value text.In view of the problems of traditional image correction methods such as adapting only to a specific meter and low robustness,and considering that scale value text is a common feature of meters,this paper proposes a meter image correction method based on scale value text feature matching.It is well used in image correction of different appearance instruments.(2)Polar coordinate transformation center extraction based on scale value text.The current high-precision polar coordinate transformation interpretation methods mostly use the method of extracting the polar coordinate transformation center based on the tick mark feature,and the tick mark contains only a very small amount of image feature information,so the recognition algorithm is susceptible to environmental factors such as strong light and shadows.The scale value text image often has more image feature information.Therefore,this paper proposes a polar coordinate transformation center extraction method based on the scale value text.This method obtains very stable positioning and recognition results by introducing a neural network.The robustness is much higher than the extraction method based on tick mark features.(3)The method of extracting instrument pointer and tick mark based on secondary area search.The pointer and scale line extraction methods in the existing polar coordinate transformation interpretation method are easily affected by other invalid information in the dial,resulting in poor extraction accuracy.In response to this problem,this paper proposes a pointer and tick mark extraction method based on secondary area search.This method effectively eliminates the influence of other invalid interference information in the dial on the pointer and tick mark extraction and improves the algorithm accuracy.This paper also produced a data set containing pointer instrument images in different collection environments,and verified the validity and accuracy of the algorithm proposed in this paper through experimental simulation.Compared with other methods in experiments,this set of recognition algorithms proposed in this paper has higher reliability and effectiveness.