Enhancing War on Drug Effort by Leveraging Remote Sensing Data and Machine Learning
Dedi Irawadi, Tuga Mauritsius
Abstract:
Traditional methods of eradicating illicit cannabis plantations that relies on informants are ineffective. Integrating remote sensing data with machine learning enhances the speed and accuracy of detection, improving eradication efforts. Several studies for smart cannabis detection have been developed in other regions of the world, however, no publications are recorded to implement the utilization of combination of remote sensing data with machine learning algorithms for cannabis plantation detection in Indonesia. This study proposes a fast methodology using remote sensing and machine learning to identify potential cannabis plantations in Indonesia. A Random Forest (RF) model outperforms a Gradient Boosting Tree (GBT) in detecting cannabis plantations, achieving higher accuracy (90.18% overall, Kappa = 0.8309) than GBT (84.89% overall, Kappa = 0.7712). The RF-based model also demonstrates superior spatial performance with less misclassification. Although the proposed methodology performs well, it is limited to homogeneous cannabis plantations. Future research should also address high cloud cover, which complicates detection and increases the risk of overfitting.
Keywords:
Remote sensing, machine learning, cannabis, smart detection, law enforcement