<![CDATA[Weeds are one of the major constraints in lowland rice cultivation because they compete with rice plants for essential growth resources and may reduce crop productivity. Rapid and accurate weed identification and mapping are therefore needed to support effective weed management. This study aimed to analyze the reflectance patterns of weeds and rice plants and to evaluate the performance of pixel-based classification using Random Forest (RF) and Support Vector Machine (SVM) algorithms on multispectral Unmanned Aerial Vehicle (UAV) imagery. The study was conducted in a 1,300 m² lowland rice field in Ciherang Village, Ciomas District, Bogor Regency, using multispectral UAV imagery acquired on October 4, 2023, before weed cleaning, and October 13, 2023, after weed cleaning. The imagery consisted of green, red, red-edge, and near-infrared (NIR) bands. Pixel samples representing weed and non-weed classes, including rice, soil, and water bodies, were selected using purposive sampling. The pixel data were then divided into training and validation datasets at a 70:30 ratio and classified using RF and SVM with default parameters in the Dzetsaka plugin. The results showed that weeds generally had higher reflectance values than rice plants, particularly in the red-edge and NIR bands, providing spectral separability that could be exploited for pixel-based classification. RF achieved 100% Overall Accuracy (OA), User's Accuracy (UA), and Producer's Accuracy (PA) on both observation dates. SVM achieved an OA of 100% on October 4, 2023, which decreased to 92.13% on October 13, 2023, with a PA of 81.65% for the weed class. Weed area estimates also differed between the two algorithms, with the average weed area estimated by RF decreasing from 5.75% to 2.50%, while that estimated by SVM decreased from 3.06% to 2.76% after weed cleaning. These findings indicate that multispectral UAV imagery combined with pixel-based classification can be used to identify and map weeds in lowland rice fields, with RF showing more consistent performance than SVM under the conditions of this study.]]>
