Weather Prediction using Cloud’s Images

Traditional weather prediction operates on a large scale, as it relies on professional devices, complicated computer algorithms, sophisticated meteorologists, and more. Some also based on radiation of satellite image or spatial textures [1]. Though more precise, it cannot be achieved on one's own. This project aims to build a portable mobile App that forecasts weather conditions running locally via photos taken on a phone. It runs by classifying clouds and their types with the help of the existing cloud shape database, adopting techniques such as CNN (convolutional neural network)/deep learning. The development process includes collecting photos of clouds, finding out their relationship with weather conditions, writing python code to train the model with given data, testing the model, and then building the mobile App. We divided the photos gathered into four types: rain clouds, fair-weather clouds, tornadoes clouds, and fog, and we mainly focused on the shape of the clouds, through testing by whether the testing set or photos of our own afterward, the accuracy was acceptable with rains and fair-weather, especially fogs, except for tornados. However, as a mobile App, its recognition rate needs to be improved, and there are high demands for resolution, angle, and brightness, so the training model we got is not yet ready for use to customers, though adding photos into the training set could improve it.