R&D Demo | Cloud masking
Contact Us
Massive storm with intense winds, torrential rains
and severe flooding
Cold ocean waters reach deeply into the mountainous coastline of northern Norway
The Camp Fire grew to become the destructive wildfire
Crop fields somewhere in Vietnam
Applications Area
Insurance
Type measurements
Burn delineation
Deforestation
etc...
Agriculture
THE SOLUTION
SoftServe developed an algorithm based on machine learning techniques to rapidly process a batch of images to detect only high-informative ones for further processing.
challenge
Difficulty
Overuse
Value
67% of the Earth's surface is typically covered by clouds at any one time.
Therefore, a significant part of satellite images contain only partial data.
This results in considerable increases in time and costs to sort cloudy for non-cloudy images.
how it works
U-Net
Human
SoftServe employed a popular deep learning architecture called U-Net for the task of recognizing different labels in a sky/cloud image.
We trained an architecture on 256x256 px cropped images to reach higher accuracy. The output of the U-Net model is a probabilistic mask, wherein each pixel has a softargmax value in the range [0, 1]. This value indicates the degree of cloudiness of the pixel in the image. However, the success of any machine learning application depends heavily on the quality and size of the training data.
SoftServe’s R&D team prepared the high-quality labeled dataset from scratch. Clouds and their cast shadows were then accurately labeled on the satellite images with different landscapes. It consists of a thousand hand-labeled 512x512 patches of Sentinel-2 images. Each labeled with one of the following three classes: clear (land), cloud, shadow. Additionally, the following were used for training: Red, Green, Blue and NIR channels.
VALUE PROPOSITION
SoftServe’s machine learning algorithm can sort satellite images based on the level of cloud/shadow obstruction to provide the most relevant and data rich images with little needed human effort.
Universal
Effective
Adjustable
Can be used with different satellites and platforms
Detects both clouds and their cast shadows
Easy to re-train on any landscape or environment
Let's work
together
Thank you
for your request!
We appreciate you contacting us. We have sent you an email from hello@connect.softserveinc.com
Have a great day ahead!
Contact us
Follow us