Development of a Portable Low-Cost Multispectral Sensor Integrated with IoT and Machine Learning for Classifying Honey Types

论文详情

#

官方摘要 (Official Abstract)

#

“Accurate honey type authentication is a significant challenge for small-scale producers, as conventional methods are often costly and impractical. This study aims to design and develop a low-cost honey classification prototype by integrating the AS7265X multispectral sensor with Internet of Things (IoT) technology and machine learning. Spectral data from 18 channels of various Indonesian honey types were acquired using the AS7265X sensor and analyzed exploratively using Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA). The data were then normalized and used to train Artificial Neural Network (ANN), Random Forest (RF), and Support Vector Machine (SVM) classification models. An ESP32-based IoT system was developed for real-time monitoring and cloud data storage. The results demonstrate that AS7265X spectral data effectively differentiate honey types, with the ANN model achieving 94.05% accuracy, supported by a responsive IoT system (1–2 seconds) for monitoring and centralized storage. This prototype shows potential as a practical, rapid, accurate, and efficient honey authentication solution for various stakeholders.”

中文摘要翻译：

#

准确的蜂蜜真伪及种类鉴别对于小规模蜂农及生产商而言极其重要，然而传统的化学实验室检测方法往往成本高昂且难以普及。本研究旨在研发一款低成本便携式蜂蜜智能鉴别原型机，深度整合了 AS7265X 18通道多光谱光学传感器、物联网 (IoT) 遥测技术与机器学习算法。使用 AS7265X 传感器采集多种印尼本土蜂蜜在 18 个不同光谱波段下的反射/吸收数据，利用主成分分析 (PCA) 与线性判别分析 (LDA) 进行降维和探索性数据分析。归一化后的数据用于训练人工神经网络 (ANN)、随机森林 (RF) 及支持向量机 (SVM) 等分类模型。基于 ESP32 的物联网系统实现了极速数据上传（响应延迟仅需 1–2 秒）和云端集中存储。实验结果表明，基于 AS7265X 的多光谱数据具备极佳的区分能力，其中 ANN 神经网络模型取得了高达 94.05% 的分类准确率。该原型机具备精准、高效、便携的特点，为蜂业从业者提供了一种极具潜力的现场鉴别解决方案。

关键词 (Keywords): AS7265X Sensor, Honey classification, Internet of Things, Machine Learning, Spectroscopy.

技术指标与核心性能

#

• 光学传感器硬件： 18通道 AS7265X 紫外-可见-近红外多光谱光学传感器 (UV-Vis-NIR)。
• 嵌入式控制器与物联网： 采用 ESP32 微控制器，实现超低延迟 (1–2 秒) 实时遥测及云端集中数据存储。
• 探索性数据分析： 基于主成分分析 (PCA) 和线性判别分析 (LDA) 提取光谱核心特征。
• 机器学习分类对比： 评估人工神经网络 (ANN)、随机森林 (RF) 与支持向量机 (SVM)。
• 最高分类准确率： 人工神经网络 (ANN) 表现最优，准确率达到 94.05%。

推荐引用 (Citation)

#

@article{Muhammad2025Honey,
  author  = {Muhammad, Riki and Isroni and Wisesa, T. P. and Syahputra, T. S.},
  title   = {Development of a Portable Low-Cost Multispectral Sensor Integrated with IoT and Machine Learning for Classifying Honey Types},
  journal = {Journal of Energy, Material, and Instrumentation Technology},
  volume  = {6},
  number  = {3},
  pages   = {143--153},
  year    = {2025},
  doi     = {10.23960/jemit.343},
  url     = {https://doi.org/10.23960/jemit.343}
}