Geospatial Intelligence and Artificial Intelligence for Detecting Potential Coca Paste Production Infrastructure in the Border Region of Venezuela and Colombia

Abstract

Abstract Cocaine production has reached record levels in recent years. Latin America and the Caribbean are the primary sources of all cocaine consumed globally, thus there are indications that cocaine production processes could spread to countries of transit and consumption, becoming a threat to the security of states. In this article, we address the challenge of detecting potential primary infrastructures to produce coca paste in the border region of Venezuela and Colombia. We use geospatial intelligence and artificial intelligence to detect these objects in remote sensing images and identify their geographic location. We generated a dataset of 16,778 training samples that we named CocaPaste-PI-DETECTION, constructed from PlanetScope satellite imagery rated at NIIRS level 3, ground truth data, and A1, A2, and B2 information sources. An advanced deep learning model, specialized for object detection tasks, was trained. A mean Average Precision (mAP) score of 90.07% was obtained, and we analyzed generalization capabilities and conducted different experiments that demonstrated how the proposed methodology could strengthen intervention strategies against drug trafficking.