<![CDATA[This study examines the performance of two artificial intelligence models—VGG16 and Yolov8n—in sorting plastic bottles using computer vision. The objective is to individually assess the classification and detection performance of these models with constrained computational resources and training data. The dataset consists of 300 original images for two classes (bottle and other), and is split into 210 training, 45 validation and 45 test images. The images were taken in different lighting conditions and orientations to simulate the real waste sorting situation . Both models were trained and evaluated on CPU based hardware to simulate a constrained computing environment. The VGG16 was evaluated using classification metrics like accuracy, precision, recall, and F1-score, while the YOLOv8n was evaluated using object detection metrics like precision, recall, F1-score, mAP@0.5, and frame processing speed (FPS). The accuracy of the VGG16 model was 91% on the test set. The mAP@0.5 of YOLOv8n was 0.56 with an average processing speed of 47.12 FPS, while the average processing speed of VGG16 was 5.17 FPS. These results indicate that VGG16 had a good performance on image-level classification, while YOLOv8n had a higher processing efficiency and better object-localization performance in the studied conditions. Further evaluation on embedded hardware is required to establish the suitability of YOLOv8n for practical real-time waste-sorting applications.]]>
Copyrights © 2026
