40 lines
1.5 KiB
Markdown
40 lines
1.5 KiB
Markdown
# ☕ Coffee Brewing Level Prediction Based on Spectroscopy Data Using Deep Learning
|
|
|
|
This repository contains the complete pipeline for a thesis project that predicts **coffee brewing levels** using **spectroscopy data** and **deep learning models**, later deployed on a **Raspberry Pi** with a real-time interface using **OLED display** and a **physical button**.
|
|
|
|
---
|
|
|
|
## 📚 Description
|
|
|
|
The goal of this project is to classify brewed coffee into categories such as **Underdeveloped**, **Strong**, **Weak**, **Bitter**, or **Ideal** based on spectral data obtained using the **AS7265x spectroscopy sensor**. A deep learning model is trained using Python and TensorFlow, then converted into **TensorFlow Lite** and deployed on a physical prototype for real-time prediction.
|
|
|
|
---
|
|
|
|
## 🧪 Data Collection
|
|
|
|
- **Sensor**: [AS7265x Triad Spectral Sensor](https://www.sparkfun.com/sparkfun-triad-spectroscopy-sensor-as7265x-qwiic.html)
|
|
- **Platform**: Raspberry Pi
|
|
- **Categories**: Coffee brewing levels
|
|
- **Format**: Raw 18-channel spectral data saved in `.csv` format
|
|
|
|
---
|
|
|
|
## 🧠 Modeling
|
|
|
|
### Main Libraries for Modeling:
|
|
- `pandas`: data manipulation and preprocessing
|
|
- `seaborn`: data visualization and EDA
|
|
- `tensorflow`: deep learning model training and conversion to `.tflite`
|
|
- `numpy`: numerical operations
|
|
|
|
---
|
|
|
|
## 📟 Prototype Deployment
|
|
|
|
A physical prototype is built using:
|
|
|
|
- **Raspberry Pi**
|
|
- **AS7265x Spectroscopy Sensor**
|
|
- **SSD1306 OLED Display**
|
|
- **Push Button**
|
|
- **TensorFlow Lite Interpreter** |