Zhang Boli's team and Yangzijiang Pharmaceutical release a large AI model for testing Chinese medicinal herbs

On October 16, at the Qihuang Town Convention and Exhibition Center in Zhangshu, Jiangxi, gardenia fruits were fed one by one into a testing device. An automatic loading mechanism sent them into the chamber, where an industrial vision system scanned their appearance, X-ray imaging looked through to the interior, and hyperspectral analysis examined their composition—and several dozen seconds later, a digitized "health check report" was generated.

That day, the 57th National Traditional Chinese Medicinal Materials and Pharmaceuticals Trade Fair opened in Zhangshu, and an AI (artificial intelligence) large model for testing traditional Chinese medicinal materials, jointly developed by the team of Chinese Academy of Engineering academician Zhang Boli and Yangzijiang Pharmaceutical Group, was officially released, with intelligent testing equipment carrying the technology unveiled at the same time.

Judging the quality of traditional Chinese medicinal materials has long relied on pharmacists' "observing with the eyes, touching with the hands, smelling with the nose, and tasting with the mouth," and sensory descriptions lack unified quantitative indicators, so judgments often differ among different masters and different batches. Moreover, manual inspection of goods is time-consuming, and laboratory physicochemical testing involves complex pretreatment and waiting times measured in hours or even days, while medicinal material trading is on-site work conducted in batches, multiple batches, and at high frequency, so manual inspection of goods affects work efficiency.

Traditional Chinese medicinal materials come in many varieties and from wide sources, making it difficult to distinguish genuine from fake and good from poor. Coupled with problems such as long circulation chains and difficult full-chain traceability, insufficient standardization and a lack of high-level evidence-based evidence remain problems, and the phenomenon of "many standards, little use" is widespread.

This intelligent testing equipment is led by a dedicated multimodal AI large model and integrates three major technologies: industrial vision, hyperspectral analysis, and X-ray penetration imaging. Industrial vision identifies shape, color, and texture, helping to detect appearance abnormalities and surface foreign matter; hyperspectral analysis provides reference for estimating component content, quality grading, and authenticity identification; X-ray penetration looks through the interior to identify structural abnormalities such as cavities and insect damage. The three types of signals are handed together to the large model, which "reads" large numbers of qualified and adulterated samples and can comprehensively judge the quality of traditional Chinese medicinal materials.

A single-sample test with the equipment takes only several dozen seconds and requires no complex sample preparation. When medicinal materials arrive, a sample is taken and put into the chamber and results come out immediately—warehouse quality inspection can use this to determine whether there is adulteration or deterioration, and reject nonconforming goods on the spot without waiting idly; random sampling on production lines provides timely warnings when counterfeit products are mixed in or mold appears, avoiding scrapping entire batches; at trading markets, medicines are tested on the spot when purchased, and both buyer and seller see the results simultaneously, reducing disputes caused by "judging goods by experience."

Liu Changqing, senior manager of Tianjin Modern Innovative Traditional Chinese Medicine Technology Co., Ltd., introduced that the equipment's comprehensive testing accuracy can reach over 90%, and modeling has been completed for more than 50 commonly used traditional Chinese medicinal materials, covering bulk varieties prone to adulteration such as gardenia, ginseng, angelica, red ginseng, and astragalus.

Liu Changqing introduced that AI testing does not replace statutory inspection and professional judgment. It distills the skills of veteran pharmacists—feeling its texture, weighing it in the hand, observing its shape, tasting its flavor, and smelling its aroma—into callable and traceable data assets: images, spectra, and judgment results from testing are automatically saved, forming an electronic archive of medicinal materials from purchase to factory entry, which can be used both for incoming material ledgers and quality reviews and can connect to regulatory traceability platforms. More importantly, AI testing transforms "color, texture, and appearance quality" into quantifiable indicators, helping to implement digital standards for geo-authentic medicinal materials.

At the same time, trace chemical components such as pesticide residues and heavy metals are difficult to quantify accurately using vision plus hyperspectral analysis alone, and the team is studying multi-technology integration solutions to achieve breakthroughs. In addition, the equipment will in the future be iterated into a portable version and deployed down to production areas and markets.