Plant Disease Classification

CNN-based leaf disease classifier (38 classes) with a Streamlit inference UI and recorded training history.

Companion model weights also live in plant-disease-detector (trained_plant_disease_model.keras).

Overview

Upload a plant leaf image and get a disease class prediction plus a short care recommendation. Training was done in Train_Plant.ipynb; inference UI is main2.py (Streamlit + TensorFlow/Keras).

Tech stack

• TensorFlow / Keras
• NumPy
• Streamlit
• Training notebook + training_hist.json

Features

• Multi-crop disease recognition (Apple, Corn, Grape, Potato, Tomato, … — 38 labels)
• Confidence from model softmax
• Rule-based treatment tips per class
• Team project (Ribhu S, Saketh, Varun) per UI about section

Results (from training_hist.json)

Three logged epochs:

Epoch	Train acc	Val acc
1	~57.8%	~84.1%
2	~85.9%	~91.4%
3	~91.5%	~91.0%

Final logged validation accuracy ≈ 91.0%. No additional held-out test metrics are recorded in-repo beyond this history file.

Getting started

1. Obtain trained_plant_disease_model.keras (see plant-disease-detector or your training output).
2. pip install tensorflow streamlit numpy
3. streamlit run main2.py

Project structure

Train_Plant.ipynb          training
main2.py                   Streamlit app
training_hist.json         logged accuracy/loss
*.JPG / pptx               samples / slides

Future improvements

• Publish eval on a fixed test split with confusion matrix
• Slimmer dependency story and model card
• Merge classifier + detector repos for a single recruiter entry point