Plant Doctor: AI-Powered Plant Disease Detector

Inspiration

Over 50% of India’s population relies on agriculture for their livelihood. Yet, according to FAO estimates, plant diseases cause 20–40% yield losses worldwide, with some regions in India facing even higher rates. Early detection is critical, but access to expert plant pathologists is limited—especially in rural areas. I wanted to create an accessible, AI-powered tool to help farmers, gardeners, and students identify plant diseases quickly and accurately.

What it Does

Plant Doctor is an AI-based plant disease detection system capable of identifying over 38 plant diseases from leaf images with 99% accuracy, trained on a dataset of 60,000+ images. Users can simply upload a photo or use their device camera to get instant diagnosis results, along with brief treatment suggestions.

Features include:

Real-time plant disease detection and diagnosis
Treatment recommendations and prevention tips
Interactive plant health quizzes for agricultural education
Disease encyclopedia for learning about plant illnesses

How I Built It

Collected and cleaned a large-scale dataset of plant leaf images, creating consistent metadata and captions for each class.
Trained a deep learning model using PyTorch and ResNet18, optimized with transfer learning for high accuracy.
Built the UI using Streamlit, combining image/camera inputs, instant predictions, educational tools, and chatbot support.

Challenges We Ran Into

Handling a massive dataset within limited computing resources.
Managing time for metadata creation, training, and feature integration within the hackathon deadline.
Fixing UI while the backend training was still in progress.

Accomplishments That We're Proud Of

Successfully trained a deep learning model on several classes of plant diseases.
Designed a feature-rich, educational, and interactive platform.
Created a tool that has real-world impact potential for both farmers and hobbyists.

What We Learned

Large datasets require not only computing power but also careful data labeling and cleaning.
Balancing model accuracy with speed is essential for real-time applications.
Hackathon timelines demand prioritizing critical features over perfection.