Detection of Compost Piles in Agricultural Area Using Unmanned Aerial Vehicle Imagery and Artificial Intelligence

Non-point sources of water pollution, particularly in agricultural areas, are increasingly concerning. Among these, compost piles are receiving heightened attention as a source of pollution. This study utilized unmanned aerial vehicle (UAV) imagery for the efficient detection and management of compost. The basic data consisted of multispectral UAV imagery, including Red, Green, Blue, Rededge, and NIR bands, as well as the Normalized Difference Vegetation Index (NDVI). The analysis compared object-based image analysis with a random forest classifier (OBIA-RF) and deep learning-based instance segmentation using Mask R-CNN. In the accuracy evaluation based on Intersection Over Union (IoU) 75 %, OBIA-RF was analyzed with precision of 0.4356, recall of 0.7273, and F1 score of 0.5128, while Mask R-CNN was analyzed with precision of 0.6923, recall of 0.9818, and F1 score of 0.8120. While OBIA is useful, it shows limitations in handling spatial and spectral data, leading to partial misclassifications within the compost boundaries. Mask R-CNN, in contrast, effectively maintains spatial and contextual data, ensuring more precise identification and delineation of compost piles. This research underscores the potential of UAV-based remote sensing combined with advanced image analysis techniques in environmental management. These methods offer a more efficient and comprehensive alternative to conventional field surveys for monitoring agricultural non-point source pollution.