This paper describes a mock-up of an automatic sorting machine designed to detect external defects in potato tubers, perform automatic inspection, and remove defective tubers from the processing flow using a jet of compressed air. The stationary machine analyzes a single working flow, ensuring uniform tuber distribution through a controlled feeding mechanism. The developed vision and inspection system is based on regulatory standards for food potato quality control. The recognition process consists of three main modules: segmentation, tracking, and classification. Segmentation is performed using color threshold analysis, which distinguishes defective areas from the conveyor belt background. This method enhances real-time defect detection by comparing tubers with reference samples. Tracking relies on a centroid tracking algorithm, which continuously identifies tuber coordinates and predicts movement, ensuring precise timing for defect removal. Classification is conducted using an artificial neural network trained on a proprietary dataset containing images of both marketable and defective tubers. The deep learning model effectively identifies various defects, including mechanical damage, rot, and disease. The combination of these methods enhances sorting accuracy, reduces errors, and improves overall inspection efficiency, making the system a reliable solution for automated potato quality control.