Scholarlink

<p>Traditional colorimetric pH strips rely on subjective visual interpretation, while camera specifications and lighting variability often influence smartphone-based systems. We developed a compact, machine learning (ML)-enhanced pH sensing platform combining a custom ESP32-S3 device with controlled LED illumination and a mobile application. The system utilizes a 787-sample dataset spanning pH 0-14, extracting 45 statistical features from the red-green-blue (RGB) channels, hue-saturation-value (HSV) channels, and CIELAB color spaces. An AutoML pipeline was used to select an ExtraTreesRegressor model, achieving an <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(R^2\)</EquationSource> </InlineEquation> value of 0.967 and outperforming traditional RGB-based methods (<InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(R^2\)</EquationSource> </InlineEquation> = 0.618). The integrated Android application enables real-time, offline pH prediction via Wi-Fi communication. This fully embedded, end-to-end platform eliminates dependency on smartphone cameras and illumination while maintaining high accuracy across the full pH range, making it well-suited for portable pH monitoring in resource-limited settings.</p>

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

High-accuracy machine learning–based colorimetric pH quantification using a custom-built portable strip-imaging device and smartphone application

  • 期刊论文

  • 发表日期：

    2025年11月25日

查看全文
  • Ece Minel Bursalı,
  • Mehmet Akif Özdemir,
  • Mustafa Şen

摘要

Traditional colorimetric pH strips rely on subjective visual interpretation, while camera specifications and lighting variability often influence smartphone-based systems. We developed a compact, machine learning (ML)-enhanced pH sensing platform combining a custom ESP32-S3 device with controlled LED illumination and a mobile application. The system utilizes a 787-sample dataset spanning pH 0-14, extracting 45 statistical features from the red-green-blue (RGB) channels, hue-saturation-value (HSV) channels, and CIELAB color spaces. An AutoML pipeline was used to select an ExtraTreesRegressor model, achieving an \(R^2\) value of 0.967 and outperforming traditional RGB-based methods ( \(R^2\) = 0.618). The integrated Android application enables real-time, offline pH prediction via Wi-Fi communication. This fully embedded, end-to-end platform eliminates dependency on smartphone cameras and illumination while maintaining high accuracy across the full pH range, making it well-suited for portable pH monitoring in resource-limited settings.