Towards a CNN-Based Method for Improving Long-Term Storability of Apples: A Socio-Technical Approach to Pre-Warehouse Fault Detection

Efficient post-harvest management of apple supply chains is critical to minimize waste and preserve fruit quality, particularly in the storage phase. This study addresses the inefficiency and inaccuracy of defect and disease assessment in Controlled Atmosphere (CA) storage intake, where rapid and reliable evaluation of large quantities is imperative. Existing literature focuses on machine learning applications in pre-harvest crop monitoring, but practical, post-harvest solutions remain underexplored. We propose the development of a Convolutional Neural Network (CNN) for detecting apple defects that seamlessly integrates into existing CA storage warehousing operations. Employing Action Design Research, this interdisciplinary study collaborates with a medium-sized fruit storage enterprise in Styria, Austria, iteratively refining a CNN model through an Action Design Research (ADR) team. This research not only aims to fill the gap between academic methodologies and practical application but also seeks to enhance operational efficiency, demonstrating the real-world utility of advanced machine learning in agriculture.