https://github.com/farshidnooshi/fruits-360-dataset-classifier
Classifying fruits on the Fruit-360 dataset by creating a fully connected artificial neural network from scratch.
https://github.com/farshidnooshi/fruits-360-dataset-classifier
deep-learning image-classification machine-learning neural-network
Last synced: 18 days ago
JSON representation
Classifying fruits on the Fruit-360 dataset by creating a fully connected artificial neural network from scratch.
- Host: GitHub
- URL: https://github.com/farshidnooshi/fruits-360-dataset-classifier
- Owner: FarshidNooshi
- License: mit
- Created: 2022-02-01T08:41:08.000Z (over 3 years ago)
- Default Branch: master
- Last Pushed: 2022-07-07T09:01:37.000Z (almost 3 years ago)
- Last Synced: 2025-04-01T01:51:08.647Z (about 2 months ago)
- Topics: deep-learning, image-classification, machine-learning, neural-network
- Language: Python
- Homepage:
- Size: 14.5 MB
- Stars: 5
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
In The Name Of GOD
# Fruit-360-classification
In this project, a fully connected Artificial Neural Network(ANN) is implemented from scratch.
## Neural network architecture and details
This ANN was implemented to classify 4 classes of fruits. Feedforward algorithm was implemented in vectorized form using softmax as activation function for each layer. Back propagation was implemented in both *iterative* and *vectorized* forms with *sum of squared errors (SSE)* as cost function. *Stochastic Gradient Descent* algorithm was used to train the network.
### Additional parts included:
- Hyperparameter tuning
- Improving SGD using momentum algorithm
- Adding more classes of fruits and hyperparameter tuning
- Using softmax as output layer's activation function
## Dataset
* The [Kaggle 360-Fruits dataset](https://www.kaggle.com/moltean/fruits) was used.
Also a [feature extraction and size reduction technique](https://github.com/FarshidNooshi/Fruit-Detector/blob/master/Implementation/project_assets/Feature_Extraction_Train.py) was used on train and test dataset to simplify the problem.