Electronic Theses and Dissertation

Deteksi objek kecil pada citra udara merupakan tantangan dalam computer vision karena resolusi rendah dan kepadatan objek tinggi pada citra drone. penelitian ini mengusulkan integrasi yolov8 dengan slicing aided hyper inference (sahi) untuk meningkatkan akurasi deteksi objek kecil. yolov8 digunakan sebagai model dasar, sedangkan sahi memotong citra menjadi potongan kecil untuk meningkatkan sensitivitas deteksi terhadap objek kecil. penelitian ini menggunakan dataset visdrone2019, dengan tahapan pelatihan model yolov8, implementasi sahi, dan evaluasi performa menggunakan metrik map, precision, recall, dan f1-score. hasil penelitian menunjukkan bahwa kombinasi yolov8 dan sahi dapat meningkatkan performa deteksi objek kecil pada citra udara secara signifikan. kata kunci: deteksi objek kecil, yolov8, sahi, citra udara, visdrone.

Small object detection in aerial imagery is a major challenge in computer vision due to low object resolution and high background complexity in drone-captured images. This study proposes the integration of YOLOv8 and Slicing Aided Hyper Inference (SAHI) to enhance the accuracy of small object detection. YOLOv8 serves as the base model, while SAHI improves detection sensitivity by slicing high-resolution images into smaller patches. The VisDrone2019 dataset is used to evaluate the system, with model performance measured using mAP, precision, recall, and F1-score metrics. The results show that the combination of YOLOv8 and SAHI significantly improves small object detection performance in aerial imagery. Keywords: small object detection, YOLOv8, SAHI, aerial imagery, VisDrone.