TFLite_HSI_CNN

Repository for TFLite HSI CNN classification on the RP2040 microcontroller (Raspberry Pi Pico).

---

🛠️ Required Software

• Arduino IDE version 2.1.0 or later
• Web browser with internet access (for Google Colab)

---

🧪 How to Run the Python Part

1. Open Google Colab
2. Load the notebook: HSI_Training_2D.ipynb
3. Run the "Initialization" section
4. Grant Google Colab access to your Google Drive
5. In the "Global" section, set the variables for execution
6. Run the remaining sections

📁 Output Files (inside /Models folder)

• db_CNN.h (where db depends on the selected dataset)
• tensor_correct.h
• tensor_input.h

---

⚙️ Arduino IDE Setup (First-Time Configuration)

1. Open Arduino IDE
2. Connect the Raspberry Pi Pico
3. In the left panel, go to Boards Manager (second icon)
4. Search for Arduino Mbed RP2040 Boards and install version 4.0.2 or newer
5. Add TensorFlow Lite library:
   Go to Sketch → Include Library → Add .ZIP Library, and load:
   Libs/Arduino_TensorFlowLite-2.4.0-ALPHA.zip

---

🚀 Inference on the Board

1. Load the sketch:
   Sketch/sketch_cnn_in_i8_out_i8_db
2. In the same folder as the sketch, add the previously generated data files
3. Click "Upload" to flash the board
4. To monitor execution:
   Go to Tools → Serial Monitor

Cite this:

J. V. S. Hütner, F. Viel, C. A. Zeferino and E. A. Bezerra, "TinyML Applied in Hyperspectral Image Classification on COTS Microcontroller," 2024 XIV Brazilian Symposium on Computing Systems Engineering (SBESC), Recife, Brazil, 2024, pp. 1-6, doi: 10.1109/SBESC65055.2024.10771925.

@INPROCEEDINGS{hutner2024tinyML, author={Hütner, João Victor Santos and Viel, Felipe and Zeferino, Cesar A. and Bezerra, Eduardo Augusto}, booktitle={2024 XIV Brazilian Symposium on Computing Systems Engineering (SBESC)}, title={TinyML Applied in Hyperspectral Image Classification on COTS Microcontroller}, year={2024}, volume={}, number={}, pages={1-6}, keywords={Deep learning;Adaptation models;Satellites;Microcontrollers;Tiny machine learning;Computational modeling;Convolutional neural networks;CubeSat;Hyperspectral imaging;Image classification;Hyperspectral Image;TinyML;CNN;CubeSat;COTS}, doi={10.1109/SBESC65055.2024.10771925}}