https://github.com/samiyaalizaidi/image-classification-with-rpi
This repository contains all the files for the computer vision based final project of the Neural Networks and Machine Learning class.
https://github.com/samiyaalizaidi/image-classification-with-rpi
classification computer-vision convolutional-neural-networks deep-learning keras-classification-models neural-networks python3 raspberry-pi-4
Last synced: 4 months ago
JSON representation
This repository contains all the files for the computer vision based final project of the Neural Networks and Machine Learning class.
- Host: GitHub
- URL: https://github.com/samiyaalizaidi/image-classification-with-rpi
- Owner: samiyaalizaidi
- Created: 2023-11-09T23:49:04.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2024-03-03T10:17:17.000Z (over 1 year ago)
- Last Synced: 2024-12-29T09:42:30.508Z (6 months ago)
- Topics: classification, computer-vision, convolutional-neural-networks, deep-learning, keras-classification-models, neural-networks, python3, raspberry-pi-4
- Language: Jupyter Notebook
- Homepage:
- Size: 12.1 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# CPE 4903: Neural Networks and Machine Learning
This project aims to train two Convolutional Neural Networks (CNNs): 'Hand Digit Recognition' and 'Cats and Dogs Classification.' The models will then be deployed on a Raspberry Pi to perform inference through PiCamera.
The Hand Digit Classifier was required to have an accuracy of $`\geq 99\%`$ whereas the Cats and Dogs Classification model must have an accuracy of $`\geq 85\%`$.
## Design and Implementation
### Hardware Specifications
Both models were deployed on the same device. The products and their specifications are:
- Raspberry Pi 4 - Model B - 2GB RAM - 64-bit quad-core Cortex-A72 Processor
- Raspberry Pi Camera Module - imx219
### Software Specifications
The software specifications along with the Python libraries are:
- Raspbian 32-bit OS, Bookworm
- Python 3.7.12
- TensorFlow 2.4.0
- Keras Preprocessing
- OpenCV
- Pyenv for creating a local environment for Python
### Training the Models
#### Hand-Digit Recognition
The dataset used for training the model MNIST. The dataset was split into training and testing, with 60000 images for training and 10000 for testing. The size of the input images was $`(28, 28, 1)`$.
The model was trained using TensorFlow and Keras on Google Colab. It's a custom model consisting of 4 convolutional layers and 3 dense layers. Max pooling was also used after every convolutional layer.
#### Cats and Dogs Classification
The dataset used for training consisted of images of cats and dogs. It was split into training, validation, and test subsets with a ratio of 60\%-15\%-25\% using `sklearn.model_selection.train_test_split()` function. A total of 10,000 images were used, with 5000 images of cats and 5000 images of dogs. After splitting the data, there were
- 6500 images in the training set
- 1506 images in the validation set
- 1994 images in the testing set
The model was created using TensorFlow and Keras on Google Colab. Ensemble learning was used to get more accurate results. VGG-16 was used and on top of that, 2 fully connected layers were added along with a Dropout layer of $`30\%`$. The pre-trained model, trained on the ImageNet dataset, was used and all the initial layers were frozen. All of the images were reshaped to $`(224, 224, 3)`$ for better results.
## Environment
#### Setting Up the Environment
To use the Raspberry Pi for image classification, a virtual environment with an older version of Python was used. The instructions are provided [here](setup/env.txt).
#### Other Requirements
After setting up the environment, run this command in the virtual environment.
```
pip install -r setup/requirements.txt
```
---
**Contributions**: All of the code was written by [Samiya Ali Zaidi](https://www.linkedin.com/in/samiya-ali-zaidi/).
Took help from this [video](https://www.youtube.com/watch?v=vekblEk6UPc) to set up the environment.