Einsatz von Bildverarbeitungstechnologien zur Reduktion von Schlupf in der Lebensmittelfertigung

Abstract

This Master's thesis investigates the integration of image processing systems for the reduction of undetected defective products within quality assurance processes in the food industry. Undetected defects describes the unintentional passing of defective products through quality control systems, which can have serious consequences in a safety- and health-sensitive sector such as food production, ranging from product recalls and reputational damage to legal ramifications. The objective of this thesis was to analyze the causes and consequences of undetected defects, identify relevant influencing factors, and demonstrate how modern technologies, particularly image processing in combination with artificial intelligence (AI) and machine learning (ML), can significantly mitigate this risk. Various image processing approaches were introduced, and their importance for food safety was discussed. A central element of the analysis was the evaluation of misclassifications using the confusion matrix, which enables a detailed measurement of classification system performance. Particular attention was given to the false negative rate (FNR), representing the proportion of undetected defective products, which directly correlates with undetected defects risk. The thesis also examined the differences between random sampling inspections and 100% quality control, critically assessing their respective advantages and limitations. It was shown that high sensitivity is essential to reliably detect defective products and thereby minimize classic undetected defects. Furthermore, the study addresses practical challenges in the implementation of image processing systems, such as limitations in real-time processing, ambiguous defect patterns, incompatibilities between different system providers, and the need for skilled personnel. The deployment of IoT, edge computing, and data-driven process control increasingly enables a proactive quality strategy, not only reducing undetected defects but also enhancing maintenance planning, sustainability, and resource efficiency. In conclusion, the thesis finds that image-based quality assurance systems offer significant benefits for food safety and operational efficiency. In addition to reducing external failure costs, they lay the foundation for a forward-looking, data-driven quality management approach. The targeted reduction of misclassifications, particularly false negatives, is a key factor in improving product quality and minimizing risks across the entire value chain.

Date of Award	2025
Original language	German (Austria)
Supervisor	Herbert Jodlbauer (Supervisor)