|
|
|
|
|
--- |
|
|
license: mit |
|
|
tags: |
|
|
- asl |
|
|
- sign-language |
|
|
- cnn |
|
|
- tensorflow |
|
|
- mediapipe |
|
|
- computer-vision |
|
|
--- |
|
|
|
|
|
# 🧠 ASL Sign Language Detection with CNN (Conv1D) |
|
|
|
|
|
Model ini dikembangkan untuk mendeteksi huruf American Sign Language (ASL) dari koordinat tangan (landmark) secara real-time menggunakan CNN (Convolutional Neural Network). |
|
|
|
|
|
## 📌 Deskripsi |
|
|
|
|
|
- Input: 63 fitur (x, y, z) dari 21 titik tangan yang diekstraksi dengan [MediaPipe Hands](https://google.github.io/mediapipe/solutions/hands.html) |
|
|
- Model: CNN dengan `Conv1D`, dilatih menggunakan TensorFlow |
|
|
- Dataset: ASL Alphabet (87.000+ gambar, 29 kelas) |
|
|
|
|
|
## 🏋️ Training |
|
|
|
|
|
``` |
|
|
Epoch 27/50 |
|
|
accuracy: 98.80% - val_accuracy: 98.78% - loss: 0.0386 - val_loss: 0.0636 |
|
|
``` |
|
|
|
|
|
## 📊 Evaluasi Model |
|
|
|
|
|
- Akurasi Uji: **98.92%** |
|
|
- Total dataset : 63.676 sampel |
|
|
- Jumlah data latih : 50.940 sampel |
|
|
- Jumlah data uji : 12.736 sampel |
|
|
|
|
|
### 📋 Classification Report |
|
|
|
|
|
| Label | Precision | Recall | F1-score | Support | |
|
|
| ---------------- | --------- | ------ | -------- | --------- | |
|
|
| A | 0.98 | 1.00 | 0.99 | 437 | |
|
|
| B | 1.00 | 1.00 | 1.00 | 441 | |
|
|
| C | 1.00 | 1.00 | 1.00 | 398 | |
|
|
| D | 0.99 | 0.99 | 0.99 | 493 | |
|
|
| E | 1.00 | 0.99 | 0.99 | 462 | |
|
|
| F | 1.00 | 0.99 | 1.00 | 575 | |
|
|
| G | 0.99 | 0.99 | 0.99 | 488 | |
|
|
| H | 0.99 | 0.99 | 0.99 | 479 | |
|
|
| I | 0.99 | 0.99 | 0.99 | 477 | |
|
|
| J | 0.99 | 0.99 | 0.99 | 516 | |
|
|
| K | 1.00 | 0.99 | 0.99 | 540 | |
|
|
| L | 1.00 | 1.00 | 1.00 | 505 | |
|
|
| M | 0.96 | 0.93 | 0.94 | 313 | |
|
|
| N | 0.91 | 0.96 | 0.94 | 255 | |
|
|
| O | 0.98 | 0.99 | 0.99 | 453 | |
|
|
| P | 1.00 | 0.99 | 1.00 | 408 | |
|
|
| Q | 0.99 | 1.00 | 0.99 | 419 | |
|
|
| R | 1.00 | 0.97 | 0.98 | 508 | |
|
|
| S | 0.99 | 0.99 | 0.99 | 510 | |
|
|
| T | 1.00 | 0.99 | 0.99 | 470 | |
|
|
| U | 0.96 | 0.99 | 0.98 | 503 | |
|
|
| V | 0.99 | 1.00 | 0.99 | 510 | |
|
|
| W | 0.99 | 0.98 | 0.98 | 491 | |
|
|
| X | 0.99 | 0.99 | 0.99 | 432 | |
|
|
| Y | 1.00 | 1.00 | 1.00 | 517 | |
|
|
| Z | 0.99 | 0.99 | 0.99 | 470 | |
|
|
| DEL | 0.98 | 0.99 | 0.99 | 340 | |
|
|
| NOTHING | 0.00 | 0.00 | 0.00 | 1 | |
|
|
| SPACE | 0.98 | 0.99 | 0.99 | 325 | |
|
|
| **Accuracy** | | | **0.99** | **12736** | |
|
|
| **Macro Avg** | 0.95 | 0.95 | 0.95 | 12736 | |
|
|
| **Weighted Avg** | 0.99 | 0.99 | 0.99 | 12736 | |
|
|
|
|
|
|
|
|
> *Note: Label "NOTHING" memiliki support sangat kecil (1 sampel), sehingga metriknya tidak signifikan.* |
|
|
|
|
|
## 📦 Dependencies |
|
|
|
|
|
- TensorFlow |
|
|
- MediaPipe |
|
|
- scikit-learn |
|
|
- NumPy |
|
|
|
|
|
## 📹 Demo |
|
|
|
|
|
Model ini bisa digunakan untuk prediksi real-time langsung dari webcam di browser setelah dikonversi ke TensorFlow.js. |
|
|
|
|
|
## 👤 Penulis |
|
|
|
|
|
- Ade Maulana |
|
|
- IG: [@ademaulana_](https://instagram.com/ademaulana_) |
|
|
- TikTok: [@ademaulana_4](https://tiktok.com/@ademaulana_4) |
|
|
|
|
|
## 📄 License |
|
|
|
|
|
MIT License |
|
|
|
