Rapid Classification of Sugarcane Nodes and Internodes Using Near-Infrared Spectroscopy and Machine Learning...

Area of Science:

• Agricultural Engineering
• Spectroscopy
• Machine Learning

Background:

• Automating sugarcane planting requires precise differentiation between nodes and internodes to protect buds and optimize planting material.
• Current methods may lack the speed and accuracy needed for efficient large-scale automation.

Purpose of the Study:

• To evaluate the efficacy of visible-shortwave near-infrared (Vis-SWNIR) spectroscopy combined with machine learning for classifying sugarcane nodes and internodes.
• To identify the optimal machine learning model and preprocessing technique for this classification task.

Main Methods:

• Spectral data (400-1000 nm) were collected from sugarcane cultivar Khon Kaen 3.
• Various spectral preprocessing techniques were applied to enhance features.
• Linear Discriminant Analysis (LDA), K-Nearest Neighbors (KNN), and Artificial Neural Networks (ANN) were employed for classification.

Main Results:

• All evaluated machine learning models achieved high classification accuracy.
• Artificial Neural Networks (ANN) combined with derivative preprocessing yielded the best performance.
• An F1-score of 0.93 was achieved on calibration and validation sets, and 0.92 on an independent test set.

Conclusions:

• Vis-SWNIR spectroscopy and machine learning offer a feasible solution for rapid and accurate sugarcane node/internode classification.
• This approach supports the automation of sugarcane billet preparation and advancements in precision agriculture.