https://github.com/ratikaewkam/physics-classification

This project was developed as a prototype for future advanced video classification applications in schools or laboratories.
https://github.com/ratikaewkam/physics-classification

neural-networks physics python

Last synced: about 1 month ago
JSON representation

This project was developed as a prototype for future advanced video classification applications in schools or laboratories.

• Host: GitHub
• URL: https://github.com/ratikaewkam/physics-classification
• Owner: ratikaewkam
• License: mit
• Created: 2025-07-29T13:41:50.000Z (11 months ago)
• Default Branch: main
• Last Pushed: 2025-07-31T09:06:07.000Z (11 months ago)
• Last Synced: 2025-08-07T16:48:29.351Z (11 months ago)
• Topics: neural-networks, physics, python
• Language: Jupyter Notebook
• Homepage:
• Size: 826 KB
• Stars: 0
• Watchers: 0
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# physics-classification

This project was developed as a prototype for future advanced video classification applications in schools or laboratories.

---

We tested the model using two classification classes:

#### Free Fall

![Free Fall](https://raw.githubusercontent.com/ratikaewkam/physics-classification/refs/heads/main/img/Freefall.png)

#### Sound vs Fire

![Sound vs Fire](https://raw.githubusercontent.com/ratikaewkam/physics-classification/refs/heads/main/img/SoundvsFire.png)

We designed three different neural network models by varying the activation functions, the number of layers, and the number of neurons.

All three models achieved 100% accuracy, but the third model had the fewest parameters, making it the fastest in terms of computation and prediction speed.

### Results

We divided the dataset into four folders: training, testing, validation, and unseen.

- During training and testing, all models performed without any issues.

- However, when we used unseen data—especially videos captured from different camera perspectives (e.g., different zoom levels)—the models struggled to make accurate predictions.

- If the perspective and setup of the unseen video matched the training or testing conditions, the models could still predict correctly, even if that exact video had never been used before.

### Next Steps

In the next phase, I will collaborate with three high school students. I will build the base model, and they will apply it in real-world scenarios. For example, they will integrate the model with a Raspberry Pi.

As for me, if time permits, I plan to explore new approaches to neural network design.

### Credits

- Model Development & Owner of the Sound vs. Fire Dataset: ([Rati Kaewkam](https://github.com/ratikaewkam))

- Raspberry Pi Integration & Owners of the Free Fall Dataset: (Three high school students)