Scholarlink

This paper aims to evaluate the performance of the implementation of a framework for images classification system (for example, traffic signs of different categories) that can be integrated into an Internet of Things application. A multi-core implementation method based on the NPU - MAX 78000 accelerator from Analog Devices is proposed. Computing time and accuracy are evaluated to determine if such system can be implemented in real time and how such implementation is feasible. A neural network architecture with minimum possible complexity but with maximum accuracy has been established by simulations to have minimum number of weights. The results show that such a system can work with very good performance, high image acquisition rates, with very high accuracy and for a reasonable number of neural networks.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Framework to Efficient Implementation of Neural Network Image Classification for IoT Systems

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Sorin Zoican,
  • Roxana Zoican

摘要

This paper aims to evaluate the performance of the implementation of a framework for images classification system (for example, traffic signs of different categories) that can be integrated into an Internet of Things application. A multi-core implementation method based on the NPU - MAX 78000 accelerator from Analog Devices is proposed. Computing time and accuracy are evaluated to determine if such system can be implemented in real time and how such implementation is feasible. A neural network architecture with minimum possible complexity but with maximum accuracy has been established by simulations to have minimum number of weights. The results show that such a system can work with very good performance, high image acquisition rates, with very high accuracy and for a reasonable number of neural networks.