[Research on crop-weed discrimination using a field imaging spectrometer].

Area of Science:

• Agricultural Engineering
• Remote Sensing
• Plant Science

Background:

• Effective weed management is essential for optimizing crop yields and reducing herbicide use.
• Precision agriculture relies on accurate identification of crop and weed species for targeted interventions.

Purpose of the Study:

• To investigate the use of hyperspectral imaging for discriminating between crops and weeds.
• To identify optimal spectral bands for weed-crop classification.
• To assess the feasibility of developing low-cost hyperspectral systems for weed detection.

Main Methods:

• Acquisition of hyperspectral images (380-870 nm) using a field imaging spectrometer.
• Data normalization to account for varying illuminance.
• Stepwise forward variable selection and Fisher's Linear Discriminant Analysis (LDA) for band selection and classification.
• Evaluation of band combinations for cost-effective system development.

Main Results:

• Classification accuracy reached 85% (individual species) and over 91% (overall weeds) with selected bands.
• A three-band combination achieved 89% accuracy, comparable to five selected bands.
• The "red edge" spectral region proved informative for crop-weed discrimination.

Conclusions:

• Hyperspectral imaging is effective for crop-weed discrimination.
• Optimal band selection can significantly improve classification accuracy.
• The findings support the development of affordable hyperspectral devices for precision weed management.