Scholarlink

Purpose <p>Smartphone color imaging offers a promising, field-ready alternative to laboratory or hyperspectral techniques for free iron oxide (Fe<sub>d</sub>), a key control on soil carbon sequestration, nutrient buffering and site-specific agronomic decisions.</p> Methods <p>We imaged 150 B-horizon samples from 70 soil profiles across subtropical Jiangxi, China, with five flagship smartphones (Samsung Galaxy Note 10+ 5G (SP1), Huawei Mate 30 Pro (SP2), Apple iPhone 11 (SP3), OPPO Reno3 Pro 5G (SP4), and Xiao Mi 10 Pro (SP5)) under controlled illumination, alongside a laboratory SOC710VP hyperspectral imager. The RGB data were converted to five color-space models, and hyperspectral curves were smoothed with a Savitzky-Golay filter. Support vector machine (SVM) and partial least squares regression (PLSR) models were calibrated on a Kennard-Stone algorithm (70% training, 30% validation).</p> Results <p>Iron oxides darkened soils and shifted hue towards the red-yellow domain, yielding absolute Pearson correlations up to 0.74 with saturation and hue indices. Across the independent validation set, SVM consistently surpassed PLSR; the iPhone 11-SVM combination produced the best smartphone performance (R<sup>2</sup> = 0.64, RMSE = 5.92 g kg<sup>-1</sup>, RPD = 1.66), within 2% of the hyperspectral benchmark (R<sup>2</sup> = 0.65). These results place the iPhone 11-SVM workflow in the “approximate quantitative” accuracy class (0.60 ≤ R<sup>2</sup> &lt; 0.80), suitable for variable-rate liming, micronutrient placement and carbon-budget modelling.</p> Conclusion <p>Future research should integrate robust field-calibration targets and ensemble optimisation to generalise this low-cost approach across diverse soils and illumination conditions, accelerating the adoption of smartphone-based soil diagnostics in precision agriculture.</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

Smartphone-based soil color analysis and machine learning for free iron oxide prediction in subtropical B-horizon soils

  • 期刊论文

  • 发表日期：

    2025年8月9日

查看全文
  • Feilong Shen,
  • Jiawei Yang,
  • Tianwei Wang,
  • Nian Li,
  • Shuxin Que,
  • Mengyang Xu,
  • Jifeng Li,
  • Zhongbing Chen

摘要

Purpose

Smartphone color imaging offers a promising, field-ready alternative to laboratory or hyperspectral techniques for free iron oxide (Fed), a key control on soil carbon sequestration, nutrient buffering and site-specific agronomic decisions.

Methods

We imaged 150 B-horizon samples from 70 soil profiles across subtropical Jiangxi, China, with five flagship smartphones (Samsung Galaxy Note 10+ 5G (SP1), Huawei Mate 30 Pro (SP2), Apple iPhone 11 (SP3), OPPO Reno3 Pro 5G (SP4), and Xiao Mi 10 Pro (SP5)) under controlled illumination, alongside a laboratory SOC710VP hyperspectral imager. The RGB data were converted to five color-space models, and hyperspectral curves were smoothed with a Savitzky-Golay filter. Support vector machine (SVM) and partial least squares regression (PLSR) models were calibrated on a Kennard-Stone algorithm (70% training, 30% validation).

Results

Iron oxides darkened soils and shifted hue towards the red-yellow domain, yielding absolute Pearson correlations up to 0.74 with saturation and hue indices. Across the independent validation set, SVM consistently surpassed PLSR; the iPhone 11-SVM combination produced the best smartphone performance (R2 = 0.64, RMSE = 5.92 g kg-1, RPD = 1.66), within 2% of the hyperspectral benchmark (R2 = 0.65). These results place the iPhone 11-SVM workflow in the “approximate quantitative” accuracy class (0.60 ≤ R2 < 0.80), suitable for variable-rate liming, micronutrient placement and carbon-budget modelling.

Conclusion

Future research should integrate robust field-calibration targets and ensemble optimisation to generalise this low-cost approach across diverse soils and illumination conditions, accelerating the adoption of smartphone-based soil diagnostics in precision agriculture.