potato disease classification

This project classifies potato leaf diseases from uploaded images using a deep learning model trained with TensorFlow.

Rather than focusing only on model training, the project demonstrates a complete AI deployment pipeline by integrating a React frontend, FastAPI backend, TensorFlow Serving, and Docker.

The application predicts one of the following classes:

Early Blight

Late Blight

Healthy

Along with the prediction, it provides:

Confidence score

Disease description

Treatment recommendations

Prevention tips

Features

Potato leaf disease classification

Transfer Learning with MobileNetV2

Fine-Tuned deep learning model

Confidence score prediction

Disease information

Treatment recommendations

Prevention guidelines

FastAPI REST API

TensorFlow Serving for inference

Docker & Docker Compose support

Responsive React interface

System Architecture

React Frontend

│

▼

FastAPI Backend

│

▼

TensorFlow Serving

│

▼

MobileNetV2 SavedModel

Model Development

The project initially started with a custom CNN model. Although the results were acceptable, the overall performance and generalization were not satisfactory.

To improve the model, the architecture was redesigned using Transfer Learning.

The final model uses MobileNetV2 pretrained on ImageNet as the feature extractor, followed by custom classification layers.