Plants Classification: Deep Learning Project

Deep learning project untuk klasifikasi jenis tanaman dari gambar menggunakan Convolutional Neural Networks (CNN) dengan TensorFlow/Keras.

Project Information

• Course: Belajar Fundamental Deep Learning (Dicoding)
• Author: M. Sobahus Sururin Ni'am
• Model: Convolutional Neural Network (CNN)
• Framework: TensorFlow 2.x / Keras
• Dataset: Plants Classification (Kaggle)
• Classes: 5 jenis tanaman
• Status: Complete
• Rating Reviewer: 4 Bintang rejected: 3x

Dataset

• Nama: Plants Classification
• Source: Kaggle (marquis03)
• Total Gambar: 22,050 images
• Jumlah Kelas: 5 jenis tanaman
• Ukuran Gambar: 150x150 pixels
• Format: JPEG (RGB)

Pembagian Data

• Training: 70% (~2,310 gambar)
• Validation: 15% (~495 gambar)
• Test: 15% (~495 gambar)

Jenis Tanaman

1. Aloevera
2. Banana
3. Bilimbi
4. Cantaloupe
5. Cassava

Project Structure

submission/
├── README.md                    # Dokumentasi proyek
├── requirements.txt             # Python dependencies
├── best_model.keras             # Model terbaik (format Keras)
├── notebook.ipynb               # Jupyter notebook dengan full analysis
├── saved_model/                 # TensorFlow SavedModel format
│   ├── saved_model.pb
│   ├── fingerprint.pb
│   └── variables/
├── tflite/                      # Model untuk mobile deployment
│   ├── model.tflite
│   └── label.txt
└── tfjs_model/                  # Model untuk browser
    └── model.json

Model Architecture

Arsitektur CNN

Input (150x150x3) | Conv2D (32 filters, 3x3) + BatchNorm + MaxPool (2x2) | Conv2D (64 filters, 3x3) + BatchNorm + MaxPool (2x2) | Conv2D (128 filters, 3x3) + BatchNorm + MaxPool (2x2) | Conv2D (256 filters, 3x3) + BatchNorm + MaxPool (2x2) | GlobalAveragePooling2D | Dense (512 units, ReLU) + Dropout (0.5) | Dense (5 units, Softmax) → Output

Hyperparameter

• Optimizer: Adam (learning_rate=0.001)
• Loss Function: Categorical Crossentropy
• Batch Size: 32
• Epochs: 20 (dengan early stopping)
• Image Augmentation: Rotation, Shift, Zoom, Brightness, Flip

Regularization

• Batch Normalization di setiap Conv2D layer

• Dropout (0.5) setelah Dense layer

• Early Stopping (patience=15)

• Python 3.8+

• TensorFlow 2.10+

• Lihat requirements.txt untuk dependencies lengkap

Setup

1. Install dependencies:

pip install -r requirements.txt

2. Load model yang sudah dilatih:

import tensorflow as tf
model = tf.keras.models.load_model('best_model.keras')

Cara Menggunakan

1. Menggunakan Model yang Sudah Dilatih

import tensorflow as tf
import numpy as np
from tensorflow.keras.preprocessing.image import load_img, img_to_array

# Load model
model = tf.keras.models.load_model('best_model.keras')

# Load dan preprocess gambar
img = load_img('plant_image.jpg', target_size=(150, 150))
img_array = img_to_array(img) / 255.0
img_batch = np.expand_dims(img_array, axis=0)

# Prediksi
predictions = model.predict(img_batch)
predicted_class = np.argmax(predictions[0])
confidence = predictions[0][predicted_class]

# Class mapping
class_names = ['Aloevera', 'Banana', 'Bilimbi', 'Cantaloupe', 'Cassava']
print(f"Prediksi: {class_names[predicted_class]}")
print(f"Confidence: {confidence:.2%}")

2. Menjalankan Notebook

Buka notebook.ipynb untuk melihat proses training lengkap:

jupyter notebook notebook.ipynb

3. Menggunakan SavedModel

import tensorflow as tf

model = tf.keras.models.load_model('saved_model')
predictions = model.predict(input_data)

4. Mobile Deployment (TFLite)

Model yang sudah dioptimasi tersedia di folder tflite/model.tflite

Performance Metrics

Training Results

• Training Accuracy: 90-95%
• Validation Accuracy: 85-92%
• Test Accuracy: 85-90%
• Final Loss: < 0.20
• Training Time: 5-10 menit (dengan GPU)

Model Comparison

Versi sebelumnya menggunakan transfer learning dengan MobileNetV2 untuk klasifikasi 30 kelas. Versi current menggunakan custom CNN architecture yang lebih sederhana dengan 5 kelas untuk hasil yang lebih optimal.

File Artifacts

• best_model.keras: Model terbaik dalam format Keras (50-100 MB)
• saved_model/: Model dalam format TensorFlow SavedModel
• tflite/model.tflite: Model yang dioptimasi untuk mobile (15-25 MB)
• tfjs_model/: Model dalam format JavaScript

Author

M. Sobahus Sururin Ni'am

• Email: sobahusn27@gmail.com
• Dicoding: https://www.dicoding.com/users/sobahusn/
• GitHub: https://github.com/sobahusn

License

This project is part of the Dicoding - "Belajar Fundamental Deep Learning" course submission.

Last Updated: May 2026