Skripsi
PERBANDINGAN PERFORMA MODEL YOU ONLY LOOK ONCE VERSION 11 (YOLOV11) DAN SWIN TRANSFORMER DALAM DETEKSI PENYAKIT PADA BUAH TOMAT
The global agricultural sector faces significant challenges in meeting the demand for food production, with plant diseases being a major factor in reducing yields. In tomato plants, fruit diseases are often manually detected by farmers, but this method is time-consuming and prone to misdiagnosis. This study aims to compare the performance of the YOLOv11 and Swin Transformer models in detecting tomato fruit diseases using Computer Vision techniques. This study began with dataset preparation by conducting a preprocessing process through the stages of cleaning, class balancing, resizing, and denoising. The YOLOv11 model was trained with the YOLO annotation format, while the Swin Transformer used the COCO JSON format. Evaluation was carried out based on Precision, Recall, mean Average Precision (mAP), and inference speed metrics measured by Latency and FPS values. The test results showed that the YOLOv11s model (M4 scenario). In the testing phase, YOLOv11s-M4 achieved a Precision value of 0.890, Recall 0.936, mAP50 0.935, and mAP50-95 of 0.818. This model also boasts a significant inference speed of 5.34 FPS. Overall, YOLOv11s provides an optimal balance between high accuracy and processing time efficiency, making it highly potential for implementation in real time agricultural monitoring systems.