Digital-Display Temperature and Humidity Instrument Recognition Based on YOLOv3 and Character Structure Clustering

In this paper, in order to more efficiently verify digital-display temperature and humidity instruments and better evaluate the quality of digital-display temperature and humidity instruments, we propose a new recognition method of digital-display temperature and humidity instrument based on YOLOv3 and character structure clustering. First, the screen region of digitaldisplay temperature and humidity instrument contains all valid characters, so we define the smallest bounding rectangle region of the screen region as the region of interest. We extract the region of interest through YOLOv3-tiny neural network. Then we use YOLOv3 neural network to detect characters on the region of interest. Finally, according to the intra-class correlation of characters, we use character structure clustering to obtain temperature and humidity values. In addition, in this paper, we verify the effectiveness of this method through experiments.