This research presents a system designed to control both the quantity and quality of tube caps in the manufacturing process of blood tubes. The system performs several tasks, including verifying the presence of 100 tubes on each tray and detecting defects on tube caps such as scratches or cracks. For tube counting, the Hough Transform technique is used to count the number of tubes on each tray based on the foam tray image. To detect defective caps, we propose a novel method, FeatureExtracted, which leverages CNN-based feature extraction followed by Principal Component Analysis (PCA) and K-Nearest Neighbors (KNN) classification. Experimental results show that the system achieves 99% accuracy with an average inspection time of 0.11 s per cap ( \(\approx \) 11.8 s per tray of 100 caps), satisfying the real-time throughput requirements of industrial production lines. These results highlight the reliability of the proposed approach in separating good and defective caps, making the system a practical solution for automated large-scale quality control.