[Research on distinguishing weed from crop using spectrum analysis technology]

Area of Science:

• Agricultural Science
• Remote Sensing
• Plant Biology

Background:

• Site-specific weed management requires accurate automatic detection methods.
• Leaf spectral reflectance is influenced by pigments, structure, and water content, offering potential for plant discrimination.

Purpose of the Study:

• To identify characteristic wavelengths for distinguishing weeds from cotton and rice.
• To develop discriminant models for accurate weed classification using spectral data.

Main Methods:

• Spectral reflectance measurements (350-2500 nm) of cotton, rice, and weeds in a laboratory setting.
• Statistical analysis using SAS, including STEPDISC for wavelength selection and DISCRIM for model development.
• Validation of classification accuracy for weed discrimination models.

Main Results:

• 100% classification accuracy achieved for distinguishing spine-greens from cotton using wavelengths 385, 415, and 435 nm.
• 100% classification accuracy achieved for distinguishing barnyard-grass from rice using wavelengths 375, 465, 585, 705, and 1035 nm.
• Wavelengths 415 and 435 nm were most critical for spine-green discrimination; 585 nm and 705 nm were key for barnyard-grass discrimination.

Conclusions:

• Spectral reflectance analysis is a highly effective method for automated weed detection in agricultural settings.
• Specific wavelengths in the visible and near-infrared spectrum are crucial for differentiating crops from weeds.
• This approach supports precision agriculture by enabling targeted weed control strategies.