Ultra-high-resolution hyperspectral imagery datasets for precision agriculture applications

Area of Science:

• Agricultural Science
• Remote Sensing
• Data Science

Background:

• Technology integration in agriculture is advancing crop identification, yield prediction, and disease detection.
• Hyperspectral remote sensing offers a versatile method for agricultural parameter mapping.
• High-resolution hyperspectral datasets are crucial for developing advanced remote sensing algorithms for crop mapping.

Purpose of the Study:

• To present a high-resolution, ground-based hyperspectral imaging dataset for developing and validating plant-level crop mapping techniques.
• To support research in precision agriculture by providing data on crop type, nitrogen levels, and ground truth.
• To facilitate the development of machine learning models for crop mapping and nutrient management.

Main Methods:

• Acquired ultra-high spatial resolution (3 mm) hyperspectral imagery (400-900 nm, 3 nm spectral resolution) using a ground-based push-broom system.
• Collected data over experimental fields of cabbage and eggplant with three distinct nitrogen levels (high, medium, low).
• Captured imagery in single-crop and mixed-crop configurations, including comprehensive ground truth data.

Main Results:

• A novel, high-resolution hyperspectral dataset for two vegetable crops (cabbage, eggplant) under varying nitrogen conditions was created.
• The dataset includes imagery with single and mixed crop types, offering diverse scenarios for algorithm development.
• Ground truth data correlating with hyperspectral imagery and nitrogen levels is provided.

Conclusions:

• The presented dataset is valuable for advancing machine learning-based crop mapping at the plant level.
• This resource will aid in validating algorithms for precision agriculture, particularly for assessing crop growth responses to nitrogen levels.
• Enables development of more accurate and efficient precision agriculture tools.