Human-In-The-Loop Pest Identifier (YOLOv8 & Streamlit)

This project isn't just an object detector; it's a complete, self-improving system. It uses YOLOv8 to identify pests and a Streamlit web app to create a Human-in-the-Loop (HITL) feedback cycle.

When the model makes a low-confidence prediction, a human (the "farmer") provides the correct label. This feedback is then used to automatically retrain and improve the model, creating a system that learns from its own mistakes.

Demo: The Loop in Action

This is the core value of the project: the model gets smarter with user feedback.

• Model v1 (Initial Guess): Identifies a pest with 74% confidence.
• Human Feedback: The user confirms or corrects the label.
• Retraining: The user clicks "Retrain," and a v2 model is automatically trained in the background using this new, verified data.
• Model v2 (After Correction): The same image is now identified with 91% confidence.

---

🚀 How to Replicate This Project

This entire project is designed to be run from a single Google Colab notebook.

Project Structure

Your repository should be organized in Google Drive like this:

/PestDetector/
├── app.py                 # The Streamlit web app
├── pest_data.yaml         # YOLO dataset configuration file
├── production_model.pt    # The starting "v1" model (you must create this)
├── datasets/              # Folder for your images and labels
│   ├── images/
│   │   ├── train/
│   │   └── val/
│   ├── labels/
│   │   ├── train/
│   │   └── val/
└── Pest_Identifier_Colab.ipynb  # Main Colab notebook

---

Part 1: Initial Setup & Training (Model v1)

Before you can run the app, you need a starting "v1" model.

1. Set Up Your Colab Environment

Mount your Google Drive and install the YOLOv8 library.

from google.colab import drive
drive.mount('/content/drive')

!pip install ultralytics

# Navigate to your project directory
%cd /content/drive/MyDrive/PestDetector/

2. Prepare Your Dataset

This project requires a dataset in YOLO format.

• Get Data: You can find a "pest" dataset on Roboflow or Kaggle.

• Organize: Place your images and labels in the datasets/ folder as shown in the project structure.

• Configure YAML: Create a pest_data.yaml file that points to your data. It must contain:

  • The absolute path to your datasets folder in Google Drive.
  • The relative paths to your train and val folders.
  • The number of classes (nc) and their names.

  Example pest_data.yaml:

  path: /content/drive/MyDrive/PestDetector/datasets
  train: images/train
  val: images/val
  nc: 3
  names:
    - 'aphids'
    - 'fruit_flies'
    - 'stink_bugs'

3. Train Your "v1" Model

Run the YOLOv8 training command.

!yolo train model=yolov8n.pt data=pest_data.yaml epochs=50 imgsz=640

4. Create the "Production" Model

After training:

1. Find runs/detect/train/weights/best.pt
2. Copy it to /content/drive/MyDrive/PestDetector/
3. Rename it to production_model.pt

---

Part 2: Running the Human-in-the-Loop App

1. Install App Dependencies

!pip install streamlit ultralytics pyngrok

2. Create the app.py File

%%writefile app.py
# Paste your full app.py code here

3. Add Your ngrok Token

!ngrok config add-authtoken YOUR_PRIVATE_TOKEN_HERE

4. Launch the App

from pyngrok import ngrok
public_url = ngrok.connect(8501)
print(public_url)
!streamlit run app.py

---

How to Use the App

1. Upload an image → model predicts.
2. Give feedback if prediction is wrong.
3. Click "Retrain & Activate Model" to generate a smarter model.
4. Re-upload the same image to see confidence improve.