Reflectance Estimation from Multispectral Linescan Acquisitions under Varying Illumination-Application to Outdoor...

Area of Science:

• Agricultural Engineering
• Remote Sensing
• Computer Vision

Background:

• Precision spraying requires accurate weed detection, which is challenging due to visual similarities between crops and weeds.
• Color information alone is insufficient for reliable weed identification.
• Multispectral imaging offers high spectral resolution for vegetated surface analysis.

Purpose of the Study:

• To develop a robust spectral reflectance estimation method for weed detection under varying illumination.
• To improve the accuracy and reliability of weed detection systems in agricultural fields.

Main Methods:

• An original image formation model was developed for radiance image acquisition using a linescan camera under changing illumination.
• A novel reflectance estimation method was deduced from the image formation model, considering frame-level illumination variations.

Main Results:

• The proposed reflectance estimation method demonstrated superior robustness against illumination variations compared to existing state-of-the-art techniques.
• Reflectance features derived from the new method proved more discriminant for outdoor weed detection and identification.

Conclusions:

• The developed image formation model and reflectance estimation method significantly enhance the performance of weed detection systems.
• This advancement contributes to more effective precision spraying, reducing herbicide use and safeguarding crop yields.