Precision Agriculture: Crop Mapping using Machine Learning and Sentinel-2 Satellite Imagery

TL;DR

This study investigates lavender-field segmentation in Sentinel-2 imagery to support precision agriculture under climate change. It compares deep learning (notably U-Net) and pixel-based methods across spectral-band configurations, showing that RGB information can achieve near-full-band performance, while multispectral inputs can boost accuracy (with All 12 reaching the highest Dice of $0.8685$ for some setups). The tuned U-Net achieves a Dice of $0.8324$ on the test set, and occasional strong performance from logistic regression suggests cost-effective alternatives in data-rich contexts. The findings offer practical guidance for lavender producers seeking scalable, climate-resilient crop monitoring using satellite imagery, balancing accuracy, data requirements, and computation.

Abstract

Food security has grown in significance due to the changing climate and its warming effects. To support the rising demand for agricultural products and to minimize the negative impact of climate change and mass cultivation, precision agriculture has become increasingly important for crop cultivation. This study employs deep learning and pixel-based machine learning methods to accurately segment lavender fields for precision agriculture, utilizing various spectral band combinations extracted from Sentinel-2 satellite imagery. Our fine-tuned final model, a U-Net architecture, can achieve a Dice coefficient of 0.8324. Additionally, our investigation highlights the unexpected efficacy of the pixel-based method and the RGB spectral band combination in this task.

Figures (3)

• Figure 1: Learning curve of tuned U-Net model(a) and default U-Net model(b).
• Figure 2: Patched sample image in RGB and the predicted output for individual patches.
• Figure 3: RGB image(a), predicted output(b), and ground truth mask (c).