Soil profile analysis using interactive visualizations, machine learning, and deep learning

Date

2021

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

Soil is an essential element of life, and soil properties are crucial in analyzing soil health. Recent developments of proximal sensor technologies, such as portable X-ray fluorescence (pXRF) spectroscopy or visible and near-infrared (Vis–NIR) spectroscopy, offer rapid and non-destructive alternatives for quantifying data from soil profiles. While the data collection time using these technologies decreases significantly, the subsequent analysis remains time-consuming, and current analysis solutions only provide basic visualizations. Furthermore, the use of collected data from proximal sensors to predict high-level soil properties has garnered worldwide attention in the past decade, owing to its convenience. Therefore, this paper discusses the objectives for software solutions in this area, consolidated from interviewing 102 stakeholders. Following these requirements, data visualizers work closely with soil scientists to propose a set of interactive visualizations for analyzing soil profiles using pXRF data. These interactive visualizations receive positive feedback from the domain experts. This project also explores various machine learning and deep learning approaches to predict soil properties from spectral data. This work then proposes a deep learning model called RDNet that achieves state-of-the-art results in predicting pHH2O and pHKCl from Vis–NIR spectra acquired from a set of globally distributed soil samples.

Description

© 2021 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

Rights

Rights Availability

Keywords

Chemical Measurement Data Analysis, Intelligent Visual Analytics, pXRF Data Visualization, Soil Property Predictions, Vis-NIR Spectra, Machine Learning and Deep Learning

Citation

Pham, V., Weindorf, D. C., & Dang, T. (2021). Soil profile analysis using interactive visualizations, machine learning, and Deep Learning. Computers and Electronics in Agriculture, 191, 106539. https://doi.org/10.1016/j.compag.2021.106539

Collections