Cassava (Manihot esculenta Crantz) is a crucial crop for food security and industrial applications in tropical regions. However, cassava mosaic disease (CMD), driven by begomoviruses such as the Sri Lankan Cassava Mosaic Virus (SLCMV), has become a major threat to cassava production in Thailand and neighboring countries. Traditional field-based CMD detection methods are labor -intensive and lack scalability, prompting the need for remote sensing and machine learning (ML) approaches. This paper proposes a structure-driven framework that combines UAV multispectral imagery, object-based image analysis (OBIA), and machine learning for CMD detection. A Parameter-by-Structure strategy was introduced, tailoring Mean-Shift segmentation parameters—Spatial Radius, Range Radius, and Minimum Segment Size—based on observed cassava canopy morphology types (isolated, partially connected, merged crowns). Segmentation outputs were used to extract statistical and vegetation index-based features, which served as input for four ML classifiers: Random Forest (RF), Support Vector Machine (SVM), multilayer perceptron (MLP), and Logistic Regression (LR). The results demonstrated that structure-adaptive segmentation substantially improved classification accuracy. MLP achieved the highest F1-Scores for isolated and partially connected canopy types (0.972 and 0.970, respectively), whereas RF performed the best for merged canopy structures (F1-Score = 0.957). The findings confirmed that segmentation tuned to biological field characteristics significantly enhanced CMD classification performance. This study highlighted the effectiveness of integrating canopy morphology into segmentation workflows to improve UAV-based disease detection. The proposed Parameter-by-Structure framework offers a practical, interpretable, and scalable solution for CMD monitoring, and this framework enhances UAV-based CMD detection and offers generalizability to other canopy-structured crops.
Promphol et al. (Sun,) studied this question.