← Projects/Deep Learning · Climate Tech

Coastal Guardian

Deep Learning for Coastal Threat Classification

2025·Archived·Deep Learning · CNN · ResNet50 · TensorFlow · Computer Vision
Coastal Guardian

Coastal Guardian

Deep Learning for Coastal Threat Classification using satellite imagery and transfer learning.

Approach

  • Designed a CNN on NASA satellite imagery to classify 4 coastal threat categories (erosion, pollution, algal bloom, clear)
  • Used ResNet50 transfer learning and augmentation on imbalanced classes
  • Achieved 87%+ validation accuracy on 2,400 image patches
  • Evaluated with confusion matrix, ROC-AUC and precision-recall

Status

Completed project.

On this page