Detection of Buildings from Satellite Images with Convolutional Neural Networks

Project Overview

This project focuses on detecting buildings from satellite images using Convolutional Neural Networks (CNNs). We trained two types of models: an object detection model and two classifiers, using the Functional Map of the World (FMoW) dataset to detect certain types of buildings in satellite images.

Models

Object Detection Model

• Architecture: Faster R-CNN
• Purpose: Identify and locate buildings within satellite images.

Classifiers

We utilized four different convolutional neural network architectures for classification:

• DenseNet
• ResNet
• InceptionV3
• Xception

Technologies Used

• Programming Languages: Python
• Libraries: PyTorch, Keras, TensorFlow
• Platform: Trained on Google Colab