Image Semantic Segmentation of Satellite Imagery using U-Net. Image segmentation using U-Net architecture on Satellite Imagery with the help of fast.ai framework
Read more about it in this blog
Recent years has shown us the benefits of studying and modelling earth and climate change, we also witnessed the importance of satellite images. Public and private organisation are spending lots of resources over satellites (some are made public for the use of entrepreneurs and research scientists). Satellite images can provide as a spatial frame for modelling and understanding earth from above. With proper data processing and analytics incorporated with cutting edge technologies like deep learning can bring great insights.
The idea of semantic segmentation is to label each pixel with a corresponding class of what it is representing. In previous image classification story we made the neural network to find out the content of an image but here it is little more deep because here along with finding what is in the image we are locating the pixel points of the contents in the image. This type of task is commonly called dense prediction. Here same class objects are represented under same labels but there is one another similar type of task called instance segmentation which can find out separate objects of same class.