Research On The Method Of Pointer Instrument Reading Recognition Based On Deep Learning

Pointer instrument has the advantages of simple structure,low cost,strong antiinterference and durability,and has been widely used in the fields of power equipment monitoring,industrial manufacturing,military and aerospace.Under normal conditions,data reading and recording of pointer meters always rely on manual observation.However,manual observation can result in heavy workload,poor timeliness,low efficiency and,more importantly,high error rate.With the development of enterprise automation and intelligence,manual reading can no longer meet the requirements of modern enterprises for management and maintenance.Currently,most algorithms for traditional pointer meter reading recognition can only run in a specific environment or fixed position without a high degree of reliability,stability,and long-term availability.With the development of deep learning,digital image processing technology and deep learning related technology are combined in this paper to conduct in-depth research on the methods of reading identification of pointer instrument at each stage.The core content of this paper mainly includes four aspects: the recognition and segmentation of the dashboard area from the image collected by the camera,the preprocessing of the acquired dashboard,the segmentation and refinement of the dial pointer of the instrument,and the segmentation and recognition of the dial number of the instrument.In this paper,based on the relevant knowledge of deep learning,in the identification and segmentation of instrument dial,is proposed and improved Mask RCNN convolution neural network model to the pointer instrument effective information acquisition in the natural environment,and establish the training data set to training and testing of the algorithm,precision fast under the condition of natural scenes,recognition and segmentation of the different size and type of instrument dial,is featured by strong anti-interference advantages.Compared with traditional image processing and positioning dial area,deep learning has better applicability to image extraction dial area in different natural scenes and will not be affected by different background interference.However,traditional image processing requires different correction for different background interference and has poor applicability.In the preprocessing of the acquired dashboard,weighted average method,OSTU algorithm(the most inter-major variance method)binarization and connected domain marker were used for preprocessing to facilitate subsequent operations.In the segmentation and refinement of instrument dial pointer,connected domain marker,OSTU threshold segmentation and least square method are used to extract and fit instrument pointer according to the characteristics of effective information of instrument.Segmentation and recognition in digital instrument dial,first of all to dial the number and scale line segmentation,and then with the improved Lenet-5 convolution Neural Network(Convolutional Neural Network,CNN)model for segmentation of digital identification,and is obtained by statistical dial the number of scale line dividing value,thereby put forward a method of simulation of artificial reading improved the accuracy of the meter said several read.Deep learning technology is used to solve the problem of high working intensity of manual reading.