<![CDATA[Land cover classification is an important element in spatial analysis for environmental planning and management. This study aims to evaluate the effectiveness of the YOLOv8 algorithm in classifying land cover objects using high-resolution aerial imagery data from drones. Data collection was carried out using a DJI Mavic 1 Pro drone at an altitude of 100 metres, followed by annotation with bounding boxes through the Roboflow platform, division of the dataset into training, validation, and test data, and training of the YOLOv8 model with 50 epochs to measure performance and detection accuracy. The results showed that YOLOv8 was able to detect land cover objects such as buildings, vegetation, and trees with good accuracy, especially in images with a single object. Accuracy decreased in images with two to three objects due to increased spatial complexity and overlap between objects. The bounding box-based approach is effective for detecting individual objects, but it is not yet capable of classifying the entire image area as semantic segmentation methods do. Thus, YOLOv8 has the potential to be applied in monitoring land use change and sustainable environmental management, and can be further developed through integration with segmentation models such as U-Net or Mask R-CNN to improve spatial classification accuracy.]]>
Copyrights © 2026
