基于轻量级卷积神经网络的水表读数自动识别系统

Water Meter Reading Automatic Recognition System Based on Lightweight Convolutional Neural Network

本文基于ESP32微控制器设计了一种轻量化的卷积神经网络用于自动识别水表的数字读数,该神经网络通过Tensorflow Lite深度学习开源框架部署到微控制器上,通过OV2640摄像头采集图像并传输给ESP32微控制器调用神经网络模型执行数字分类推理,实现读数识别。实验结果表明,该网络模型可以部署在硬件资源有限的ESP32微控制器上运行,对于清晰数字样本的预测准确率可达96%以上。

In the paper,a lightweight convolutional neural network is designed,which is based on ESP32 microcontroller to automatically recognize the digital reading of water meter.The neural network is deployed to the microcontroller through Tensorflow Lite deep learning open source framework,and the image is collected by OV2640 camera and transmitted to ESP32 microcontroller.The neural network model is executed to perform digital classification inference to realize reading recognition.The experimental results show that the network model can be deployed on the ESP32 microcontroller with limited hardware resources,and the prediction accuracy of clear digital samples can reach more than 96%.