Monitoring Visual Properties of Food in Real Time During Food Drying

• Published in: Food engineering reviews 2023-06, Vol.15 (2), p.242-260
• Main Authors: Iheonye, Anthony C., Raghavan, Vijaya, Ferrie, Frank P., Orsat, Valérie, Gariepy, Yvan
• Format: Article
• Language: English
• Subjects: Chemistry; Chemistry and Materials Science; Chemistry/Food Science; Computer vision; Control systems; Deep learning; Driers; Drying; Fluidized beds; Food; Food quality; Food Science; Food waste; Image segmentation; Monitoring; Real time; Shelf life; Vision systems; Visual discrimination learning
• ISSN: 1866-7910; 1866-7929
• DOI: 10.1007/s12393-023-09334-6

Annually , one-third of the food produced globally is lost or wasted. A considerable portion of global food waste comprises dry foods that are rejected due to their unattractive appearance. One effective technique to solve this problem is by developing dryers that consistently produce dry foods that are visually appealing and have a long shelf life. The beating heart of such dryers is a computer vision (CV) system that monitors the visual attributes of the food, in real time, during the drying process. Unfortunately, there are currently no real-time CV systems for monitoring the visual attributes of food during fluidized bed drying. This setback is linked to figure-ground separation challenges encountered while segmenting real-time images of the food. Sadly, when current CV systems are used to monitor visual attributes of food during fluidized bed drying, these CV systems fail miserably because they are not designed to account for three major dryer-dependent determinants—the layout, the state and pattern of motion, and the behavior of food materials within the image captured during fluidized bed drying. To solve this lingering problem, this paper reviewed various computer vision systems based on the three determinants. This study revealed that input images for the different CV systems can be categorized as being either static-type images or chaotic-type images. The CV systems were grouped into “ Static-input offline CV systems ,” “ Static-input online CV systems ,” and “ Chaotic-input online CV systems .” Building on the insight gained while reviewing the three classes of CV systems, two novel AI-driven solutions for monitoring visual attributes of food, in real time, during fluidized bed drying were proposed. The first solution was a “two-pass” deep learning system that predicts visual attributes from segmented results. While the second solution was a “single-pass” deep learning system that by-passes the segmentation step, thus saving computational cost. When such AI-driven solutions are merged with a control system and then integrated with fluidized bed dryers, this union could open the gateway to intelligent drying, where dryers consistently produce high-quality dry foods. By extension, consistency in product quality could reduce global food losses and waste significantly.