Development and Application of a Vehicle-Mounted Soil Texture Detector

Chao Meng, Wei Yang, Hong Lan, Xinjian Ren, Minzan Li2020 · DOI

TL;DR

This paper introduces a novel vehicle-mounted soil texture detector that combines electrical conductivity (EC) measurements with soil surface image analysis. This integrated approach allows for rapid, real-time soil texture prediction without the need for traditional time-consuming laboratory methods or soil pretreatment, achieving an impressive 84.86% accuracy.

What Problem Does This Paper Address?

Accurate and timely soil texture information is crucial for effective farmland management, influencing decisions about irrigation, fertilization, and crop selection. Traditional methods for determining soil texture, such as the "straw method" (likely a typo for "standard method" or similar), are highly accurate but labor-intensive, time-consuming, and require complex chemical processes and specialized equipment. Existing automated or remote sensing approaches often suffer from limitations in spatial and temporal resolution, require soil pretreatment, or rely on expensive instrumentation. This leaves a significant gap for a low-cost, real-time, and spatially resolved solution for soil texture assessment directly in the field.

Methods

The core innovation of this work lies in the fusion of two distinct sensing modalities:

  • Electrical Conductivity (EC) Measurement: The detector utilizes a four-terminal method with four disc electrodes to measure soil EC. This method is chosen for its ability to reflect soil texture and porosity. The system employs an improved current-voltage four-terminal method to ensure accuracy by actively monitoring the current.
  • Soil Surface Image Analysis: A vehicle-mounted industrial camera captures images of the soil surface. Texture features are extracted from these images using the Gray Level Co-occurrence Matrix (GLCM) method. Specifically, four GLCM parameters—energy, entropy, moment of inertia, and correlation—were selected based on their strong correlation with soil surface roughness and particle size.
  • Data Fusion and Prediction: The EC data and the extracted GLCM texture features are fed as input into an embedded Support Vector Machine (SVM) model. This SVM model, optimized with an RBF kernel, is trained to predict soil texture (sandy loam, light loam, or medium loam).

The entire system is designed to be vehicle-mounted, allowing for continuous data acquisition as the vehicle traverses the farmland.

Datasets

The study was conducted using soil samples collected from two districts in Beijing: Tongzhou and Haidian. A total of 185 soil samples were collected. For evaluation, these samples were categorized into three soil texture types: sandy loam (63 samples), light loam (97 samples), and medium loam (25 samples). The standard values for soil texture were determined using a laser particle size analyzer, and standard EC values were measured in the laboratory.

Key Insights and Ideas

  • Synergistic Power of Data Fusion: The most significant insight is that combining EC measurements with soil surface image texture features yields substantially higher accuracy in soil texture prediction than using either method alone. This highlights the complementary nature of these two data sources in characterizing soil properties.
  • EC as a Proxy for Soil Texture: The study confirms that soil EC is a valuable indicator of soil texture, showing a decreasing trend as the percentage of finer soil particles (physical clay) increases.
  • GLCM for Surface Roughness and Texture: GLCM features effectively capture the micro-topography and particle arrangement on the soil surface, which are directly related to soil texture. The study found that specific GLCM parameters (energy, entropy, moment of inertia, and correlation) are particularly relevant.
  • SVM for Robust Prediction: The choice of SVM as the embedded model proved effective, demonstrating robustness and good learning performance, especially in scenarios with limited sample sizes.
  • Practicality of Vehicle-Mounted System: The development of a vehicle-mounted system addresses the critical need for rapid, in-situ soil assessment, eliminating the logistical hurdles of traditional laboratory analysis.

Key Results

  • Soil EC Accuracy: The correlation analysis between the detector's EC measurements and laboratory standard values yielded an R² of 0.75, indicating high accuracy for EC measurement.
  • Soil Texture Prediction Accuracy:
    • Using EC alone: 56.21% accuracy.
  • Using GLCM texture features alone: 78.38% accuracy.
  • Using the combined EC and GLCM features: 84.86% accuracy.
  • Performance by Texture Type: The combined method achieved over 87% accuracy for sandy loam and light loam. Medium loam prediction had a lower accuracy of 64%, attributed to the fine distinction between medium loam and light loam in the Kaczynski classification standard, where particle size differences are marginal.

Limitations and Future Work

  • Environmental Sensitivity: The detector's image acquisition can be affected by adverse weather conditions such as rain, fog, snow, and strong light, which can impact camera focus, white balance, and exposure. The experiments were conducted under favorable conditions, and future work aims to enhance the detector's adaptability to diverse environments.
  • Medium Loam Accuracy: The lower accuracy for medium loam prediction suggests that further refinement of the model or feature extraction might be needed for finer distinctions between closely related soil textures.
  • Future Enhancements: The authors plan to expand the detector's capabilities by integrating a deep loose plow to estimate soil bulk density and total porosity, leveraging the predicted soil texture for increased accuracy. They also aim to improve the detector's adaptability to different environments and potentially incorporate more advanced machine learning techniques.

Summary updated

BibTeX
@misc{meng2020development,
  author = {Chao Meng and Wei Yang and Hong Lan and Xinjian Ren and Minzan Li},
  title = {{Development and Application of a Vehicle-Mounted Soil Texture Detector}},
  year = {2020},
  doi = {10.3390/s20247175},
}
Figure