Product quality in many Agri-enabled FMCG businesses such as tea, tobacco, eucalyptus etc. depends heavily on the chemical composition of various compounds in the bales of leaves they purchase as the raw materials. In India, several factors such as weather, soil conditions, agricultural processes, curing processes etc. influence the chemical composition of these compounds in the leaves. Today, human experts manually grade the bales of leaves at the time of buying, making this process laborious, time-consuming, biased and highly subjective.

However, industry players have started leveraging Near Infrared and Hyperspectral Imaging to automate this process, with the help of advanced AI, Chemometrics, and Computer Vision.

The chemistry of leaves

The chemical and morphological properties, and hence the sensory grades of leaves depend largely on factors such as stalk position, leaf color, and the uniformity and oiliness of the leaves. The tenderness and freshness of leaves can be estimated from the humidity content. None of these characteristics can be extracted from normal imaging techniques.

How non-visible light is helping machines see better

Near Infrared (NIR) generally refers to the non-visible spectrum of light corresponding to the wavelengths from 800 to 2500 nm. Hyperspectral Imaging (HSI) is a technique that analyzes a wide spectrum of light instead of just assigning primary colors (red, green, blue) to each pixel. Unlike other optical technologies that can only scan for a single color, HSI is able to distinguish the full color spectrum in each pixel. Special-purpose cameras are used to capture NIR-HSI images to extract spatial (morphological) and chemical information about the sample in a single measurement.

The role of AI and Computer Vision in leaf grading

The color information is not available in non-visible spectrum of light. However, advanced image processing techniques can be used to extract features to differentiate between leaf bundles of different grades, and all these data points can be fed to machine learning to train AI models to be able to classify them in accordance to those grades. To this end, various traditional classification techniques (e.g., SVM or Decision Tree based methods) as well as deep learning approaches can be leveraged.

Conclusion

Research results in this area from within the industry seems very encouraging. Some of the large tobacco manufacturers in Europe and USA have published results pointing at evidences to have been able to accomplish this task with high enough accuracy in real time to warrant adoption of AI-based leaf grading automation into their leaf buying process.

It is high time that the Indian manufacturers take notice, and leverage this technique to enhance their supply chain efficiency too.

Photo by Alex Plesovskich on Unsplash