LAUSR

Abstract The pointer meter is widely used in the modern industrial process. This paper proposes a novel pointer meter recognition method based on wireless sensor networks (WSNs) and a lightweight convolutional neural network (CNN), which completes image preprocessing, CNN classification, and reading calculation on the WSN end node, and then only transmits the recognized result in the WSN to reduce its payload transmission data. Meanwhile, a lightweight CNN classifier model with a simple structure and small size is designed for embedding in the resource-constrained WSN node. A set of experiments have been carried out on the fabricated prototype to verify the feasibility and adaptability of the proposed method. Experimental results have shown that the maximum error of the recognized results for real-world applications is around 0.27%, while the payload transmission data of the WSN decrease from 112.5 kB to 5 bytes.