Automated Recognition of Numeric Display Based on Deep Learning

This paper proposes approach to the automatic recognition of digits from the displays of medical devices such as blood pressure meters, glucometers, analyzers of harmful substances content. A fully connected, convolutional model of deep learning with the use of tensor flow and keras deep learning libraries has been developed. A database was created for training and validation of the proposed network, which contains 4500 images. The evaluation of the proposed recognition system for a large number of images has been carried out. The optimization of the model for practical use with CPU and GPU is carried out and the average performance time is shown.