AI-driven biomass discrimination of palm fronds using low-cost vision sensors for sustainable waste valorization

The development of agricultural machinery for palm trees still faces challenges due to differences in the physical and mechanical properties of fronds and leaves between species. Shredding machines that operate without sensors have difficulty adjusting rotational speed and cutting angle, resulting in decreased performance when faced with variations in raw materials. Therefore, the objective of this study is to classify three types of palm fronds and leaves, including Areca catechu L. (AR), Cocos nucifera (CN), and Elaeis guineensis Jacq. (EG), using a low-cost camera with a resolution of 1920×1080 pixels, combined with ensemble machine learning. Samples were prepared under fresh, incubated, and dried conditions, comprising a total of 810 fronds and 972 leaves. Three feature extraction methods were applied, including RGB, Oriented FAST and Rotated BRIEF (ORB), and Lab color space, which were then analyzed using two machine learning ensemble algorithms, including gradient boosting for classification (GBC) and histogram-based gradient boosting classification tree (HGBC). The combination of ORB with HGBC achieved the highest accuracy for fronds (79.6%), while Lab with HGBC was superior for leaves (84.6%). The Lab is the most consistent feature, while ORB is clear for fronds structure. This classification system has the potential to be used as a smart sensor integrated into palm shredding machines, enabling automated operation control and increasing biomass processing efficiency.