TRDS-Net: A Resource-Efficient and Lightweight Convolutional Neural Network Model for Digit Recognition
Zhiheng Kong, Wensong Liao, Jiahui Mao, Chong Boon Tan, Hong Liu, Min Zheng · 2024
In remote monitoring systems for meter readings based on image processing technology, deploying digit recognition algorithms effectively on resource-constrained microcontroller units (MCUs) while maintaining high accuracy poses a significant challenge. To address this, this paper introduces the Tiny Residual Depthwise Separable Convolutional Network (TRDS-Net), a lightweight convolutional neural network model tailored for digit recognition. TRDS-Net significantly reduces computational and parameter requirements by adjusting key hyperparameters, without compromising model performance. Tests on the MNIST dataset and the dataset derived from real-world contexts demonstrate that TRDS-Net achieves performance comparable to more complex networks.