🌳 Tree Species Classifier

A two-part deep learning and data science project combining image-based tree species classification using a custom CNN with location-aware tree species recommendation powered by KNN on the Open Urban Tree Inventory — a dataset of 5 million street trees across 63 US cities.

---

📂 Repository Structure

tree-species-classifier/
├── Final_Week_Project.ipynb   → CNN image classifier (custom architecture)
├── 5M_trees.ipynb             → Urban tree EDA + KNN species recommender
├── amla/                      → Sample leaf/tree images (33 images)
├── 5M_trees/                  → Open Urban Tree Inventory CSVs (63 US cities)
│   ├── NewYork_Final_2022-06-18.csv
│   ├── LosAngeles_Final_2022-06-18.csv
│   └── ... (63 city files)
└── streamlit_integrated.py    → Streamlit app (integration layer)

---

📓 Notebook 1 — CNN Image Classifier (Final_Week_Project.ipynb)

Problem Statement

Manual identification of tree species from images is time-consuming and requires expert botanical knowledge. This notebook automates species identification using a custom CNN trained on leaf images.

Model Architecture

Input (150×150×3)
    → Conv2D(32) + ReLU + MaxPool
    → Conv2D(64) + ReLU + MaxPool
    → Conv2D(128) + ReLU + MaxPool
    → Flatten
    → Dense(128) + ReLU + Dropout(0.5)
    → Dense(num_classes) + Softmax

Training Details

Parameter	Value
Input size	150 × 150 px
Batch size	32
Epochs	50
Optimizer	Adam
Loss	Categorical Crossentropy
Data split	80% train / 20% validation
Augmentation	Rescaling (1/255)

Results

• Test accuracy: ~90%
• Model saved as .keras format to Google Drive

---

📓 Notebook 2 — Urban Tree EDA + KNN Recommender (5M_trees.ipynb)

Dataset

Open Urban Tree Inventory — 5,132,890 street trees across 63 of the 150 largest US cities, standardized from city-specific inventories into 28 common columns.

Source: Dryad — 5 Million Trees Dataset

Key columns used: common_name, scientific_name, city, state, latitude_coordinate, longitude_coordinate, condition, native, diameter_breast_height_CM

What it does

1. Data Cleaning & EDA

• Merged 63 city CSVs into a single dataframe
• Dropped columns with >60% missing values
• Filtered species with fewer than 3 samples
• Removed rows with missing critical fields

2. KNN-Based Species Recommender Given a location (lat/lon), tree diameter, native status, and city — recommends the most ecologically appropriate tree species based on what's already thriving nearby.

recommendation = recommend_species(
    lat=38.2274,
    lon=-85.8009,
    diameter_cm=1.2,
    native='naturally_occurring',
    city='Louisville',
    state='Kentucky',
    top_n=5
)
# → Bur oak (seen 42 times nearby), Silver maple (seen 38 times nearby) ...

3. Location Lookup Given a tree species name, returns the top cities/states where it is most commonly found.

---

🛠️ Tech Stack

Component	Technology
Deep Learning	TensorFlow, Keras
Data Processing	Pandas, NumPy
ML	Scikit-learn (KNN, StandardScaler)
Visualization	Matplotlib
Environment	Google Colab
App Layer	Streamlit
Model Persistence	joblib, .keras

---

🚀 Setup & Run

CNN Classifier

# Open in Google Colab
# Mount Google Drive and set dataset path:
data_dir = '/content/drive/MyDrive/Tree_Species_Dataset'

# Run all cells — model saved to Drive automatically

Urban Tree Recommender

pip install pandas scikit-learn numpy joblib tqdm

# Run 5M_trees.ipynb in Colab or Jupyter
# Dataset CSVs are in the 5M_trees/ folder

Streamlit App

pip install streamlit
streamlit run streamlit_integrated.py

---

🔮 Future Scope

• Add transfer learning (EfficientNetB0 / ResNet50) for higher accuracy
• Integrate Grad-CAM for visual model explainability
• Deploy as a mobile app for real-time field identification
• Extend recommender with climate and soil condition features
• Add Hindi/regional language support for Indian tree species

---

📊 Dataset Credit

Chamberlain, S. et al. (2022). A 5 million tree inventory of urban forests across the contiguous USA. Open Urban Tree Inventory. Dryad. https://doi.org/10.5061/dryad.2jm63xsrf

---

🔗 Resources

• Dataset & CNN Model (Google Drive)
• Project Presentation (Canva)

---

👩‍💻 Author

Asmita — @techAsmita
B.E. Computer Engineering | AI/ML Enthusiast
LinkedIn