MSc Research Thesis
2025
AI-Powered Food Recognition Using CNNs
Designed and evaluated a computer vision solution to identify perishable food items in restaurant environments, with the aim of reducing pre-consumption food waste through smarter inventory visibility.
39,962
food images prepared
10,000
images trained per model
2 CNNs
ResNet50 & InceptionV3
What I did
Problem Framing
Identified a practical sustainability problem in restaurants: food being wasted before consumption due to poor storage visibility.
Dataset Preparation
Worked with a large food image dataset, cleaned metadata, validated image files, mapped class labels and created a usable training dataset.
Model Development
Built transfer learning pipelines using TensorFlow/Keras with ResNet50 and InceptionV3 for multi-class food image classification.
Training Strategy
Used ImageDataGenerator, data augmentation, batch-size tuning, Google Colab GPU resources, ModelCheckpoint and EarlyStopping.
Performance Evaluation
Compared models using top-1 accuracy, top-3 accuracy, training loss and validation loss to understand accuracy and deployability trade-offs.
Business Value
Positioned the solution as a foundation for AI-enabled fridge monitoring, stock awareness and waste reduction in food service operations.
Python
TensorFlow
Keras
ResNet50
InceptionV3
Google Colab
Computer Vision
Data Preprocessing