EcoScan is a desktop application designed to assess coral health using advanced AI technology. It provides accurate classification of coral images as healthy or bleached, offering insights into marine ecosystem health.
Key Features:
AI-Powered Diagnosis: Utilizes a fine-tuned ResNet18 model for reliable binary classification of coral health.
Explainability with Grad-CAM: Visualizes the model's focus areas to help users understand its decisions.
Modern Dark UI: Features a sleek, user-friendly interface designed for clarity and efficiency.
Real-time Inference: Delivers fast, local processing on CPU or GPU for immediate results.
Ideal for marine biologists, researchers, and conservationists, EcoScan enables early detection of coral bleaching, supporting efforts to protect marine ecosystems. It can be installed via winget.
README
EcoScan: Coral Bleaching Detection
EcoScan is a desktop application designed to assess coral health. Using deep learning (ResNet18) and computer vision (Grad-CAM), it classifies coral images as Healthy or Bleached and provides visual explainability for its diagnoses.
Features
Desktop Application: A standalone Windows Desktop App built with Tkinter.
AI-Powered Diagnosis: Utilizing a fine-tuned ResNet18 model for accurate binary classification.
Explainability: Integrated Grad-CAM (Gradient-Weighted Class Activation Mapping) to visualize which parts of the coral the model is focusing on.
Modern Dark UI: Sleek, dark-themed interface.
Real-time Inference: Fast, local inference on CPU or GPU.
Tech Stack
Core: Python, PyTorch, Torchvision
GUI: Tkinter
Utilities: Pillow, Matplotlib, Numpy
Installation (via Winget)
winget install lainx86.EcoScan
Dataset
The model was trained using the Coral Reefs Images dataset by Asfar Hossain Sitab on Kaggle.