.jpg)
With the rapid development of deep learning, the analysis of satellite images is mainly moving towards the use of convolutional neural networks (CNNs). This complex process requires a large volume of images to be collected, labelled and analysed in order to train these models.
The requirement to cover a wide variety of geographies and environmental conditions is in practice a major obstacle in the process of creating training datasets. This is a key issue in the analysis of satellite images using deep learning techniques, given the cost and difficulty of accessing the images of interest.
• Development of a procedural 3D engine for image generation, adapted to the analysis of satellite and airborne images
• Generation of images respecting specific constraints in terms of resolution, area covered, de-pointing, etc.
• Simulation of a wide range of environmental conditions and scenarios
• Automatic labelling of generated images
• Automatic detection of different classes of objects of interest.
• Camouflage synthesis and impact on identification/recognition performance.
