Transfer-learning-based text recognition for condition monitoring

In the industrial and medical sectors, the condition information of many devices is often not directly transmittable due to security or commercial considerations, which poses challenges for real-time monitoring and data analysis of text recognition. This paper optimizes the training of the DBNet model for text detection and the Convolutional Recurrent Neural Network (CRNN) for text recognition, and a transfer-learning-based method is proposed to improve the accuracy and processing speed of the text recognition. Experiments show that the proposed method can achieve a higher accuracy and processing speed than traditional method in text recognition.