Smart Drying Technologies: A Case Study on Red Flesh Apple (Kissabel®) Using Computer Vision and Deep Learning

Food drying is a crucial process that helps in enhancing the shelf-life of food at ambient temperature, also lowering transportation costs. However, er-roneous applications can degrade product quality and lead to a significant energy consumption as common dryers use fossil fuels. The present study aims to de-velop a “smart” prototype hot-air dryer augmented with Computer Vision (CV) and Deep Learning (DL) technologies to monitor quality and drying status of red-flesh sliced apples (Malus domestica B. - cultivar Kissabel®) during the process. Both CV and DL represent Process Analytical Technology (PAT) tools that, combined with a Quality by Design (QbD) approach, can lead to real-time data from the process, which are potentially vital for understanding and controlling the drying dynamics of products. The experimentation was conducted on apple slices (approx. 5-mm thick) using the following drying parameters: drying time, 24 hr; temperature, 35 °C; R.H., 35 %; and airflow, 3 m s-1. The “smart” drying system consisted of i) a CMOS digital camera; ii) a LED illumination setup (4200 K); iii) a load cell; iv) an Arduino UNO microcontroller (master); v) a Jumo Di-con Touch controller (slave); vi) a temperature and R.H. sensor; and vii) a Rasp-berry Pi 4 microcomputer. The open-source software Node-Red was used as an orchestrator for controlling devices, as well as acquiring and storing product fea-tures in real-time, such as weight, color, size, and shape of slices. All acquired images of apple slices were subjected to semantic segmentation using a deep learning model. With the slice shrinkage data obtained from semantic segmenta-tion and the moisture data, time-independent moisture ratio prediction models were developed. The approach opens new perspectives for the use of CV and DL technologies in food drying.