Food Image Classification Model

This project was an introduction for me to multi-layer neural networks. Here, I learned about Convolutional Neural Networks, its model architecture, hyperparamter tuning and regularization techniques.

Project Overview

In this project, I created a convolutional neural network (CNN) for food image classification from scratch. The project follows my thought process during the fine tuning of the model to allow for better model generalisation. I later introduced and fine tuned pre-trained models such as InceptionV3 and MobileNet before choosing the best model for my task.

Model Building

Model from scratch

1. Develop Baseline Model
2. Scale till overfitting
3. Introduce data augmentation
4. Regularization techniques such as L1/L2 regularizer and dropout layers
5. Batch Normalization
6. Tune hyperparameters such as learning rate, batch size
7. Try optimizers such as SGD, reLU and RMSprop
8. Additional experimentation (Sparse Categorical cross-entropy and spatial dropout 2D)

Pretrained models

1. Import pre-trained model
2. Data Augmentation
3. Fine-tuning by unfreezing
4. Try Global Average Pooling

Model Evaluation

• Confusion Matrix
• Classification Report