Machine Learning in Remote Sensing

This repository provides tools and resources for applying machine learning to remote sensing data, enabling the generation of qualitative and quantitative data products. It covers tasks such as data cleaning, analysis, feature extraction, feature engineering, and outcome modeling across various remote sensing modalities, including multispectral, hyperspectral, and lidar data. The repository will be regularly updated with new tools and resources, as well as modifications to existing ones. Each directory below focuses on specifics task of spectral data processing and Machine Learning and includes all necessary resources.

Current Resources:

• EDA: Exploratory Data Analysis (EDA) of Sentinel2 based 13 band Multispectral imagery collected over Rochester, New York.

• ML: Analysis of Hyperspectral Imagery and application of Principal Component Analysis(PCA) for dimensionality reduction and using the data for K-means clustering based Unsupervised classification.

• HSI_classification_regression: This directory contains two main subdirectories:

  HSI_classification_regression/
  │── Classification/    # Binary classification (Logistic Regression) & Multi-class classification (XGBoost)  
  │                      # using the Pavia University Hyperspectral dataset  
  │── Regression/        # Chlorophyll content prediction using Linear Regression, PLSR, and MLP  
  │                      # based on spectral band values as features  
  

• GEE: Historical Trend Analysis of Lake Temperatures using LANDSAT Data obtained through Google Earth Engine (GEE).

• Deep-Learning: This directory contains two main subdirectories:

  Deep-Learning/
  │── CNN-based-Chlorophyl-Estimation/    #  1D convolutional neural network (CNN) for leaf chlorophyll estimation  
  │                                       # using the 425 bands hyperspectral leaf reflectance data  
  │── Transfer-Learning/                  # Using Pretrained Resnet18 Model to classify UCMerced Data 
  │                                       # based on Transfer learning principle
  

Suggestions are highly appreciated.

For resource usage and potential collaborations, contact: rb1005@rit.edu.