Honey Adulteration Detection

ISEF Category: Biochemistry

Subcategory: Analytical Biochemistry  ·  Difficulty: Intermediate  ·  Setup: School Lab  ·  Time: 1 to 2 Months

The Hook

A jar labeled honey can still hide cheap syrup. Your eyes will not catch that. You can test the label against the chemistry and train a CNN to spot the pattern. That turns a food fraud problem into a clean image-classification project.

What Is It?

Honey is mostly sugars, but not all honey has the same sugar mix. On paper chromatography, different sugars move at different speeds, so the sample leaves a spot pattern that works like a fingerprint. When someone adds corn syrup or rice syrup, that fingerprint changes.

A CNN, short for convolutional neural network, is a pattern detector for images. You give it standardized photos of the chromatograms, and it learns which spot shapes and positions match authentic honey and which ones point to adulteration.

Why This Is a Good Topic

This is a strong science fair topic because you can change one variable at a time and measure a clear result, like syrup type, image setup, or model design. It connects to food fraud, labeling, and consumer trust. You can learn separation science, image standardization, and basic machine learning without needing a university lab.

Research Questions

• How does the adulterant type, corn syrup versus rice syrup, change the chromatography fingerprint of honey?
• What is the effect of image standardization on CNN accuracy when classifying honey chromatograms?
• Does using grayscale images versus RGB images improve the model's ability to detect adulteration?
• To what extent does paper type change spot separation and classification performance?
• Which image preprocessing method, cropping, background subtraction, or brightness correction, gives the highest accuracy?
• What is the effect of honey brand diversity on false positives in the classifier?

Basic Materials

• Chromatography paper strips
• Small glass jars with lids
• Distilled water
• Pure honey samples
• Corn syrup
• Rice syrup
• Plastic transfer pipettes
• Digital kitchen scale with 0.1 g accuracy
• Ruler
• Pencil
• Smartphone camera
• Phone tripod or stand
• LED desk lamp or light box
• White poster board backdrop
• Disposable gloves
• Graph paper notebook.

Advanced Materials

• TLC plates or high-quality chromatography paper
• TLC chamber or sealed developing tank
• Micropipette set
• Analytical balance
• Reference sugar standards for glucose, fructose, maltose, and sucrose
• Flatbed scanner or fixed camera rig
• UV-visible spectrophotometer
• HPLC or LC-MS access for ground-truth sugar profiles
• Calibration target for image correction
• Computer with GPU access.

Software & Tools

• Google Colab: Runs your CNN in a browser with free notebook access for training and testing.
• Python: Cleans images, trains models, and calculates accuracy metrics.
• ImageJ: Measures spot area and intensity from chromatogram photos.
• OpenCV: Crops, aligns, and standardizes images before training.
• scikit-learn: Compares your CNN with simpler baseline classifiers.

Experiment Steps

1. Define your sample classes and decide which honey, corn syrup, and rice syrup groups you will compare.
2. Standardize your strip layout, camera position, and lighting so every image starts from the same setup.
3. Build a labeled image set and decide how you will split it into training, validation, and test groups.
4. Choose the image features your CNN will learn from, such as whole-strip photos or cropped lane images, and plan a baseline method for comparison.
5. Set your evaluation rules before modeling, including accuracy, confusion matrix, and per-class recall.

Common Pitfalls

• Letting strip length, paper type, or solvent front differ between runs, which changes the sugar pattern more than adulteration does.
• Photographing wet and dry strips in the same dataset, which teaches the CNN to read moisture artifacts instead of the chromatogram.
• Using too few honey brands, which makes the model memorize one product instead of detecting adulteration.
• Cropping images inconsistently, which shifts spot positions and breaks the fingerprint the model needs.
• Skipping a clean baseline method, which leaves you unable to tell whether the CNN beats a simple rule-based classifier.

What Makes This Competitive

A stronger version of this project tests more than one adulterant, more than one honey source, and more than one imaging setup. You can push past a class-level project by comparing CNN results with a simple spot-matching baseline and reporting confusion matrices, not just overall accuracy. The best entries also hold out brands the model never saw, so you know whether it generalizes beyond your training set.

Project Variations

• Compare raw honey with honey cut using corn syrup, rice syrup, and invert syrup to see which adulterant leaves the clearest fingerprint.
• Test whether grayscale, RGB, or edge-enhanced images give the CNN the best separation between authentic and adulterated samples.
• Replace the CNN with a simpler classifier built on measured spot positions and intensities, then compare accuracy and interpretability.

Learn More

• PubMed: Search review articles on honey adulteration, food fraud, and sugar profiling.
• NCBI Bookshelf: Read free biochemistry chapters on carbohydrates and chromatography.
• USDA FoodData Central: Compare carbohydrate profiles for syrups and sweeteners.
• MIT OpenCourseWare: Review free lessons on machine learning, classification, and overfitting.
• OpenStax Chemistry 2e: Refresh separation methods, mixtures, and polarity in a free textbook.