The basis of every forecast for weather is dataand lots of it. Although the majority of the atmospheric data collection is now fully automated, the process of determining cloud types and volumes remain a manual process. This issue is what prompted Swapnil Verma start an initiative which uses machine learning to identify six distinct classes of cloud.
The hardware that makes up this system is the Arduino Portenta H7 due to its high-performance processor and numerous connectivity features, and the Portenta Vision Shield to capture crisp images. Both boards were attached to an individual base that was placed on the top of a tripod, and powered by a battery bank via USB-C.
The MicroPython software that is installed within the Portenta H7 relies on the OpenMV library to capture pictures from the Vision Shield and performing a little amount of processing on the images. Then, Verma trained an image classification model with more than 2100 images from various cloud types including clear sky, pattern cloud thin white cloud thick white cloud dark cloud and veil cloudby using Edge Impulse and deployed it back to the board. While the Portenta is running, it scans images, categorizes them locally, then transmits the final result through MQTT to the client devices which allow them to take in the information received. Verma has even added an option to take images at a low speed and sleeps during the process to conserve the battery.
The post This system classifies different types of clouds using tinyML appeared first on Arduino Blog.
