Analyzing Urban Environments through Satellite Imagery
Christine Straub
Project Overview:
In this project, I utilized satellite imagery data from the Landsat series to analyze urban area environments. By combining traditional algorithms with artificial intelligence techniques, I developed a system that can process and interpret large-scale satellite images effectively.
Libraries:
Tensorflow, Keras, Scikit-image, GeoTiff, Pillow, OpenCV, Numpy, Pandas(geopandas), rioxarray, earthpy
Key Achievements:
- Data Preparation:
- Successfully imported and preprocessed multi-band satellite images into the Python environment.
- Seamlessly stitched together multiple satellite images to create a comprehensive view of the urban landscape.
- Performed initial calibration using the metadata associated with the geo-imagery data.
- Enhanced the precision of the stitched images by applying image transformation techniques, such as keypoint matching using the SURF algorithm.
- Traditional Algorithms and Machine Learning:
- Implemented classic machine learning and image processing techniques, including the Sliding Window algorithm and a combination of Markov Tree and Decision Tree models, to analyze satellite imagery.
- Developed a Deep Neural Network (DNN) based on the Mask-RCNN architecture, which was trained on customized data. This DNN replaced traditional band arithmetic algorithms, enabling more accurate and efficient analysis of the satellite imagery.
- Visualization and Output:
- Generated pixel-classified segmentation images, allowing for clear visualization of the analyzed results using the matplotlib library.
- Exported the processed images in the GeoTIFF format, ensuring compatibility with geographic information systems (GIS) and other geospatial applications.
- Vegetation Analysis:
- Calculated the Normalized Difference Vegetation Index (NDVI) to assess the health and distribution of vegetation in the urban environment.
- Produced comprehensive landcover classification results based on the NDVI analysis.
- Automated Data Pipeline:
- Constructed an automated data pipeline that streamlines the processing of geo-images, enabling efficient handling of large datasets and facilitating rapid analysis of urban environments.
By leveraging cutting-edge technologies and algorithms, this project demonstrates my expertise in satellite image analysis, machine learning, and data visualization. The developed system provides valuable insights into urban environments, aiding in urban planning, environmental monitoring, and decision-making processes.
Relevant Information
About Landsat Data
At over 40 years, the Landsat series of satellites provides the longest temporal record of moderate resolution multispectral data of the Earth’s surface on a global basis. The Landsat record has remained remarkably unbroken, proving a unique resource to assist a broad range of specialists in managing the world’s food, water, forests, and other natural resources for a growing world population. It is a record unmatched in quality, detail, coverage, and value. Source: USGS
Landsat data are spectral and collected using a platform mounted on a satellite in space that orbits the earth. The spectral bands and associated spatial resolution of the first 9 bands in the Landsat 8 sensor are listed below.
Segmentation using CNN
Applying the segmentation (MaskRCNN) to multi-band imageshttps://www.tensorflow.org/tutorials/images/segmentation
