In recent years, the growing importance of data in enhancing various aspects of life and production has become increasingly apparent, especially with the rise of the Fourth Industrial Revolution. This is particularly evident in the industrial sector, where analyzing factory operational data plays a crucial role in boosting productivity. However, a major challenge in data collection within factories is the widespread use of traditional meters that lack connectivity and require manual reading. While upgrading to digital meters is an option, the high costs and large-scale implementation difficulties make it less feasible. As a result, a novel approach has emerged leveraging Convolutional Neural Networks (CNNs) to extract key parameters from traditional meters using image recognition. This paper introduces an innovative method for digitizing industrial meter readings through a deep learning model trained on a fully self-constructed dataset. The proposed solution enables automatic meter reading with high accuracy and strong potential for real-world deployment in a Vietnamese building material factory.