Locating disease hotspots in Tea fields using Geo-referenced aerial imagery

Abstract

Clustering techniques serve a function in spatial analysis in detecting disease-affected regions via geographic information. In this research, large-scale images are obtained from drones and segmented into small tiles, from which geospatial coordinates are extracted and processed with clustering methods. The aim is to enhance pesticide application in tea estates, specifically in terms of controlling the Tea Mosquito Bug (TMB), hence boosting efficiency in pest management and lowering operational expenses. Four clustering methods-K-means, KNN (K-Nearest Neighbors), DBSCAN (Density-Based Spatial Clustering of Applications with Noise), and OPTICS (Ordering Points to Identify the Clustering Structure) are compared on latitude–longitude coordinates extracted from
infected regions. Their effectiveness is quantified using Silhouette scores and DaviesBouldin Index (DBI), which assess clustering quality and disease density representation. KNN yields notable accuracy (Silhouette: 0.42, DBI: 0.61), making it useful for hotspot identification, whereas DBSCAN and OPTICS show weaker results due to variations in disease density. A hybrid strategy integrating KNN for proximity-based pre-clustering followed by K-means refinement significantly improves performance compared to standalone methods. Traditional K-means alone achieves a Silhouette score of 0.53 and DBI of 0.63, while the hybrid KNN–K-means method enhances results considerably (Silhouette: 0.83, DBI: 0.31). These findings demonstrate that hybrid clustering provides a
more precise spatial representation of disease hotspots, offering a reliable solution for targeted pest management. The study emphasizes the importance of selecting appropriate algorithms in geospatial analysis to ensure accurate disease mapping in agricultural applications.