https://github.com/janaghoniem/fruits-recognition-using-deep-learning-with-data-augmentation
A deep learning project for classifying 130+ fruits using EfficientNet, ResNet, and MobileNet with custom augmentations and SE blocks. Built on the Fruits-360 dataset.
https://github.com/janaghoniem/fruits-recognition-using-deep-learning-with-data-augmentation
background-replacement cnn data-augmentation deep-learning efficientnetb0 fruits360 image-classification image-recognition keras mobilenetv2 resnet-50 squeeze-and-excitation tensorflow transfer-learning
Last synced: 2 months ago
JSON representation
A deep learning project for classifying 130+ fruits using EfficientNet, ResNet, and MobileNet with custom augmentations and SE blocks. Built on the Fruits-360 dataset.
- Host: GitHub
- URL: https://github.com/janaghoniem/fruits-recognition-using-deep-learning-with-data-augmentation
- Owner: janaghoniem
- Created: 2025-06-03T14:11:59.000Z (about 1 year ago)
- Default Branch: main
- Last Pushed: 2025-06-03T14:27:03.000Z (about 1 year ago)
- Last Synced: 2025-06-04T01:09:26.713Z (about 1 year ago)
- Topics: background-replacement, cnn, data-augmentation, deep-learning, efficientnetb0, fruits360, image-classification, image-recognition, keras, mobilenetv2, resnet-50, squeeze-and-excitation, tensorflow, transfer-learning
- Language: Jupyter Notebook
- Homepage:
- Size: 31.7 MB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Fruits Recognition Using Deep Learning with Data Augmentation
A deep learning project for classifying fruit images using CNNs, custom data augmentation, and transfer learning. Built on the [Fruits-360 dataset](https://www.kaggle.com/datasets/moltean/fruits), this project focuses on improving robustness with background replacement, noise, blur, and attention mechanisms.
## Overview
Trained and evaluated the following models:
- **EfficientNetB0**
- **ResNet50**
- **MobileNetV2**
## Dataset
- **Fruits-360**: 130+ fruit classes with uniform backgrounds
- All images resized to **100x100**
- Split into training, validation, and test sets
## Key Features
- Transfer Learning using pretrained CNNs
- Background replacement with random textures to simulate real-world conditions
- Extra augmentations: blur, noise, flips, zoom
- SE block to enhance channel-wise attention
- Early stopping and model checkpointing
- Detailed evaluation: classification reports, confusion matrices
## Training Process
