Machine Learning Approach to Assess Rubber Plant Health Through Canopy Density Mapping Using Very High-Resolution Aerial Photographs

The health of Indonesian rubber plantations has recently been compromised by rubber leaf fall disease, prompting a need for effective monitoring techniques. This study explores the use of high-resolution aerial photographs to assess rubber plant health through canopy density analysis. We employed the Random Forest machine learning algorithm for this purpose, focusing on two classification systems: [low, medium, high] and [low, high] canopy densities. Our findings reveal contrasting levels of accuracy between the two classification systems. The three-tier classification ([low, medium, high]) resulted in moderate accuracy (Overall Accuracy: 0.50, Kappa Value: 0.24), suggesting that this approach might be too detailed for the task. In contrast, the binary classification ([low, high]) demonstrated significantly better performance, with satisfactory accuracy (Overall Accuracy: 0.76, Kappa Value: 0.33). This improvement indicates that a simpler classification system with fewer categories is more effective for identifying the health of rubber plants using aerial photographs and machine learning techniques. This study underscores the importance of selecting an appropriate level of classification detail in machine learning models for agricultural monitoring. The results suggest that less complex models, with fewer canopy density categories, are more suitable for accurately assessing the health of rubber plants in situations like the rubber leaf fall disease outbreak in Indonesia.