Saved in:
| Main Author: | |
|---|---|
| Format: | Recurso digital |
| Language: | |
| Published: |
Zenodo
2025
|
| Online Access: | https://doi.org/10.5281/zenodo.15622599 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Table of Contents:
- <p>In this project, we study image classification using the Fashion MNIST data with 70,000 grayscale<br>images classed into t-shirts, dresses and sneakers. We want to create a model that can spot various clothing<br>types correctly, as this will support fashion recommendation systems, automate tagging and handle<br>inventory tasks.<br>The work is performed using Convolutional Neural Networks (CNNs) and is carried out using TensorFlow<br>and Keras. Normalisation of data is done so that the model’s training process will use consistent pixel<br>values. In CNN, layers called convolutional are used to detect features, max-pooling for decrease in data<br>and dense layers for classifying what’s found. Using ReLU activations helps the neural network better<br>learn, also stopping overfitting. A sparse categorical cross entropy loss is used to compile the model and it<br>is optimised using the Adam optimizer, trained periodically using a validation split.<br>It is clear from the model that both accuracy and loss decrease during the training process. Following<br>testing on the test set, its strength shows with a 91% accuracy. This result confirms that CNNs can be<br>applied to perform real-world image classification tasks.</p>