AI for Pest & Disease Detection: Build an Image Classifier

• Introduce participants to the fundamentals of AI-based pest and disease detection using leaf images.
• Enable participants to explore, organise and assess the quality of plant health image datasets in Google Colab.
• Guide participants in building a transfer learning–based image classifier for crop pest/disease classes.
• Develop the ability to interpret model predictions (class + confidence) and link them to simple advisory messages for farmers.
• Create awareness about the risks, limitations and biases of image-only AI diagnosis, and the need for human expert validation.
• Encourage participants to design a practical deployment flow (app/WhatsApp bot + agronomist) for real-world extension use.

📅 Day 1 – Data & Image Classification Basics (Hands-on)

• Problem framing: farmer takes a picture → app gives “likely disease + advice”.
• Hands-on: mount Google Drive in Colab, list images, show sample images with matplotlib.
• Image quality discussion: lighting, angle, noise and their effect on model performance.
• Mini dataset tour: folder structure (healthy/, diseased/ or 3–4 classes).
• Quick reflection: limitations of AI purely from images.
• Hands-on take-home: Colab notebook 01_pest_image_dataset_exploration.ipynb + folder with labelled training images.

📅 Day 2 – Build & Train a Simple Classifier (Transfer Learning, Hands-on Lab)

• Concept input: what is transfer learning and why it helps with small datasets.
• Hands-on: load a pre-trained model (e.g., MobileNet/ResNet) and freeze base layers.
• Model head: add a small dense classification layer for pest/disease classes.
• Training lab: train for a few epochs and view accuracy/loss curves.
• Hands-on quick test: upload a new leaf image and see the predicted class.
• Hands-on take-home: Colab notebook 02_pest_classifier_training.ipynb + first working “AI diagnosis” model (even if rough).

📅 Day 3 – Responsible Use & Deployment Ideas (Hands-on Design & Reporting)

• Discussion: error, bias & impact – misclassification → wrong spraying advice and farmer risk.
• Hands-on: add a prediction report cell (predicted class, confidence, advisory template text).
• Hands-on design exercise: sketch an app/WhatsApp bot flow using this model.
• Human in the loop: integrate agronomist/extension worker validation into the flow.
• Closing: how to extend – more classes, multilingual advisory, integration with helplines.
• Hands-on take-home: Colab notebook 03_pest_classifier_reporting_and_flow.ipynb + app-flow diagram for project documentation.

• Understand how pest and disease image datasets are structured (classes, folders, data quality).
• Use Google Colab to explore, visualise and manage plant health images (dataset exploration notebook).
• Build and train an image classifier for crop pest/disease detection using transfer learning (e.g., MobileNet/ResNet).
• Evaluate model performance using accuracy/loss curves and simple test images.
• Generate a basic prediction report including class, confidence and advisory template text.
• Design a responsible deployment flow (app or WhatsApp bot) with human agronomist validation in the loop.

• Undergraduate and postgraduate students in Agriculture, Agricultural Engineering, Plant Pathology, Biotechnology, Data Science or related fields.
• Researchers and faculty members interested in applying AI/ML or computer vision to crop health and pest/disease management.
• Professionals from agri-input companies, agri-tech startups, FPOs, NGOs and extension services working with farmers.
• AI/ML enthusiasts looking for a practical, agriculture-focused image-classification project.
• Preferred background: Basic familiarity with any programming language (Python is a plus) and comfort with using Google Colab.