Researchers at the World Institute of Kimchi examine an image calibrated by an AI model that scans a kimchi specimen through hyperspectral imaging, at the institute in Gwangju, in this December 2023 photo. Courtesy of World Institute of Kimchi By Ko Dong-hwan Data-driven artificial intelligence (AI) is being used to guarantee the high quality of mass-produced kimchi for global consumers, according to a state-run kimchi laboratory, Tuesday. This technological approach marks a departure from the traditional method of pickled cabbage making, which has historically relied on preparing the staple side dish by relying on feel. The World Institute of Kimchi (WiKim) announced the successful development of this technology following a six-month joint project that concluded in December. The collaboration involved a domestic AI developer and a digital education consulting firm. The primary objectives of the project were to construct a dataset necessary for quality checks on kimchi and to create an AI model capable of analyzing the collected data. In collaboration with Seoul-based data solution provider Catalonix and edu-tech service developer SLI (Solution Learning Innovation), both based in Seoul, WiKim succeeded in constructing a comprehensive dataset. The dataset contained 270,000 records of RGB color and hyperspectral images, facilitating a detailed food quality evaluation at each stage of the kimchi-making process. The consortium extracted this data from various stages, including the seasoning of cabbages in salty brine, the mixing process with ingredients and spicy paste, and the subsequent fermentation. Using this dataset, the developed AI model is capable of scanning and analyzing images to determine the levels of sweetness, saltiness, and fermentation at each stage of kimchi production. These key factors predominantly influence the overall quality of the pickled side dish. The dataset and AI model together enable efficient and precise quality checks for kimchi at production factories. This involves assessing the qualities of ingredients, evaluating the mixing process, and monitoring the fermentation stages. Additionally, the system can assign grades to the produced kimchi based on its varying qualities, providing a comprehensive quality assessment of the final product. "By solely analyzing the image dataset, this technology can expedite the inspection of mass-produced kimchi from the manufacturing stage through distribution faster than previous methods, all the while ensuring a consistently high level of quality," WiKim President Chang Hae-choon said in a statement. This breakthrough is anticipated to enhance Korea's kimchi industry, which has traditionally lacked concrete standards for quality determination. The industry has largely depended on subjective judgments from individual makers, primarily based on their personal experiences. Moreover, the diminishing workforce in the country, attributed to an aging population, poses a significant risk to an industry that has been slow to adopt technological advancements and has relied heavily on human senses for quality control. Such problems loomed large especially when the country's kimchi products are becoming more popular worldwide thanks to the global expansion of Korean culture led by K-pop and social media. Chang emphasized the need to address the challenges inherent in the entire process of exporting kimchi, spanning from cultivating ingredients to shipping the final products overseas. Without a more accurate and systematic analysis, ensuring uniform quality for global kimchi consumers remains a challenge that needs to be overcome. "In order to ensure a consistent high quality of exported kimchi, it became imperative to abandon our traditional methods and embrace a new, innovative approach," Chang said. "The AI-based, non-destructive quality check model for kimchi has overcome the past limitations and raised production efficiency." WiKim's accomplishment under the Ministry of Science and ICT follows the National Information Society Agency's selection of the institute for a national R&D project last year. This project aimed to support the creation of a dataset for AI learning.
