Overview
This project is a web-based wild plant recognition system that allows users such as hikers and campers to upload photos of plants they encounter in nature and receive instant predictions about the species. A Convolutional Neural Network (CNN) was trained using TensorFlow/Keras on a Kaggle dataset containing around 16,500 images from 35 different wild edible plant classes, split into training, validation, and test sets. The goal is to provide an informative and accessible tool that helps users identify plants correctly instead of relying on guesswork.
The trained CNN model was exposed through a FastAPI backend that loads the model, processes incoming images, and returns prediction results as JSON. On the frontend side, a lightweight interface built with HTML, CSS, and JavaScript lets users upload an image, sends it to the API, and displays the predicted plant name together with additional information. The system achieved solid accuracy on unseen images and demonstrates how deep learning models can be integrated into real-world web applications for computer vision tasks.
Key Features
- CNN-based classification of 35 wild edible plant species
- Real-time predictions from uploaded plant photos via web interface
- TensorFlow/Keras model served through a FastAPI backend
- Clean HTML/CSS/JavaScript UI for image upload and result display
- Architecture designed for easy scaling and future deployment to cloud platforms
Gallery
