NM-AIST researchers develop artificial intelligence and recognition of images technology to make tobacco leaf grading faster, more consistent and transparent across the tobacco value chain in Tanzania

The Problem: Grading’s Value Depends on When It’s Done
The market value of tobacco for the tobacco farmer is greatly affected by the quality grade given to the tobacco.


Tobacco grade labeling in Tanzania is highly dependent on manual assessment by experienced graders, who determine the physical characteristics of cured tobacco leaves before assigning the appropriate grade.


This method relies on the expertise of humans which is good but people can vary in their judgments which might lead to inconsistencies . This method is also slow and subjective .


These challenges affect not only farmers but also buyers, processors and other stakeholders who rely on accurate grading information to evaluate quality and make business decisions.


Recognizing the need for greater consistency and transparency, researchers at the Nelson Mandela African Institution of Science and Technology (NM-AIST) have explored how Artificial Intelligence could support a more efficient grading process.

From Industry Challenge to Scientific Innovation
The research journey began in 2023 under the leadership of Faith Nguleni together with Shubi Felix Kaijage, Devotha Godfrey Nyambo and Jacob Bulenga Lisuma of NM-AIST’s School of Computational and Communication Science and Engineering (CoCSE).

The team was inspired by the challenges they saw in the tobacco industry in Tanzania and investigated how computer vision and deep learning could be used to recognize the features of cured tobacco leaves and establish the proper grade labels for them.

Instead of developing a classification system in which each tobacco grade was an independent class, the researchers proposed an approach based on the real logic used in tobacco grading.

They built a huge image dataset, had it reviewed by experts and ran experiments with AI models to come up with a three-step classification process that assesses the quality, grade level and color of leaves.


This work resulted in the development of “A System and Method for Grade Labeling of Flue-Cured Tobacco Leaves” in2026.

What’s new: Teaching artificial intelligence to read a tobacco leaf
The innovation is an AI system that identifies cured tobacco leaf grade labels from photographs.

It begins with an image of a tobacco leaf, which is analyzed using specially developed deep-learning models, trained to recognize its visible characteristics.

Next, the system takes three steps:
1. Assessment of quality
The AI analyzes the leaf image and assigns its quality class according to the features it learned during model training.

2. Determine grade level
The system identifies the appropriate grade level based on the L1–L5 classification categories used in the research.


3. Color classification
Finally, the system evaluates the leaf color and integrates the outcomes of all three stages to produce the corresponding tobacco grade label.

The technology is based on a sequence that mirrors the current logic of tobacco grading in Tanzania, and is designed to support more consistent and efficient identification of grades.

Importantly, the innovation is meant to support human graders, not replace their expertise, offering another digital tool for quality assessment and decision-making.


Why It Matters: Improving Fairness and Transparency in Grading
Tobacco grading is a factor in important commercial decisions across the value chain.

More consistent grade identification could build farmers’ confidence in the assessment of their produce.

For graders, the technology may provide an additional tool to assist quality assessment and reduce reliance on subjective judgment.

More rapid and standardized grade identification could improve efficiency and consistency in handling and quality assurance for both buyers and processors.

The innovation offers opportunities for agricultural regulatory institutions and researchers interested in strengthening digital systems for the assessment of agricultural quality.

NM-AIST shows how locally developed digital technologies can solve practical industry challenges by applying Artificial Intelligence to a well-established agricultural process. ## From Research to Promising AI Results

The innovation has gone through dataset development, AI model training, and technical testing.

One of its main contributions is a dataset of 49,779 tobacco leaf images which provides a solid basis for training and evaluation of the system.

According to the reported results of the research team, the developed AI models achieved:
– 95% accuracy in classifying tobacco leaf quality.
– 82% accuracy on grade-level classification over L1–L5 categories.
– 94.7% accuracy on color classification.

The results show promising performance in the three stages of classification, but further validation is required to prove the reliability of the system in different practical grading conditions.

The innovation has been granted a patent with reference number TZ/P/2026/000114 filed in 2026.

The development is an important step toward the use of decision-support technologies based on artificial intelligence in the Tanzanian agricultural sector as the system is still at the testing stage. Extending the System’s Functionality

Now, the team of researchers is working to improve the system by increasing the number of tobacco leaf images in its data set.

Another priority is to capture more images of the different remaining tobacco leaf stalk positions, so that the AI models are trained on a broader spectrum of leaf features.

The expansion will support further model development and testing, helping the technology become more representative of the different tobacco leaves encountered in grading.

The researchers are looking for a strategic partner to help develop the dataset and reinforce the next step of the innovation.

Further research, industry collaboration and practical validation could see the system contribute to more efficient, consistent and transparent tobacco grading practices in Tanzania.

This innovation is yet another example of how Artificial Intelligence and locally-led research can help support agricultural transformation and address real challenges facing industry, a trend NM-AIST continues to demonstrate.

Innovation at a Glance

InnovationAI-Powered Tobacco Leaf Grade Labelling System
Lead InnovatorFaith Nguleni
Research TeamShubi Felix Kaijage, Devotha Godfrey Nyambo and Jacob Bulenga Lisuma
InstitutionNelson Mandela African Institution of Science and Technology (NM-AIST)
SchoolComputational and Communication Science and Engineering (CoCSE)
Research Started2023
Year Developed2026
SectorAgriculture, Artificial Intelligence, Computer Vision and Digital Technology
TechnologyDeep learning and image-based hierarchical classification
Dataset Developed49,779 tobacco leaf images
Quality Classification AccuracyApproximately 95%
Grade-Level Classification AccuracyApproximately 82%
Colour Classification AccuracyApproximately 94.7%
IP ReferenceTZ/P/2026/000114
IP Year Reported2026
Development StageTechnical testing and dataset expansion
Potential BeneficiariesTobacco farmers, graders, buyers, processors and regulatory institutions



Collaborate With Us
NM-AIST is open for collaboration with tobacco industry actors, agriculture research institutions, technology companies and development partners who are interested in taking this innovation further.

Current partnership opportunity: Helping to create and grow tobacco leaf image data sets on more stalk positions and for further AI model improvement and validation.

Artificial Intelligence to enhance agricultural quality assessment and encourage more transparent decision making.