SYNOPSIS
AI, or Artificial Intelligence have become particularly trendy for the last years. It mixes applied mathematics and computer science, and find a direct application in image analysis for classification, object detection or semantic segmentation. In particular, in GIS context, aerial imagery or street-views are canonical use-cases.
By considering Python libraries dedicated to numeric analysis, raster management as well as AI model design, the trainees will be guided until conceiving a convolutional neural network architecture that fits the image analysis purpose.
This training will first focus on AI concept definitions, in order to demystify their usage and their usefullness in GIS projects. Then the trainees will be encouraged to practice in a large part of the training: from the beginning of a Python project to the definition of a simple semantic segmentation model for georeferenced image analysis.
GOALS
With this training, you will acquire the following skills:
- Understand the interest of AI in the imagery context, and be aware of the requirements for succeeding in a AI project.
- Handle rasters in Python.
- Design convolutional neural networks applied to 2D imagery, and implement them with Python.
PROGRAM
This program is indicative. It could be adapted to your specific needs.
- AI and machine learning : presentation of concepts
- History and definition
- Machine learning, supervised and unsupervised learning, deep learning and examples of use cases
- Artificial Intelligence and neural networks for 2D imagery
- Short state-of-the-art in imagery
- Some recalls on linear algebra
- Build a convolutional neural network
- Workspace configuration
- Install Python, ipython interpreter and jupyter-notebook
- Install Python libraries dedicated to imagery (numpy, GDAL, …
- Python workspace configuration
- First Python programs
- Python in a nutshell (data types, control structures, functions, classes)
- Handle data with Python
- Handle images with Python
- Read an image with GDAL, OpenCV, pillow, pyvips
- Georeferencement
- Towards a data pipeline for images
- Convolutional neural networks for imagery : application in Python
- Build training and testing datasets
- Build a model, train it, infer results
- Results post-processing, how to extract geo-referenced data
DURATION
3 days
PRE-REQUISITE
- Base knowledges in linear algebra and statistics
- Ease in Python
- Ease in Unix-like environment
The next courses (Lyon or Paris):
.
Contact us for on-site trainings (dates are flexible to your needs).
You want to participate in this training ?
Please give us the details below if you can:
* Training
Place of training, Number of people involved, Initial level of participants, Time constraints, Specific expectations
* Contact details
Organization, Address, Contact, Email, Intracommunity VAT
