Rain detection in image using convolutional neural network

Weather is a phenomenon occurs in the earth's atmosphere. Weather affects human daily activities, especially outdoor activities. Weather observations including rainfall observation in Indonesia conducted by Meteorology, Climatology, and Geophysical Agency (BMKG). BMKG facing a major problem in terms of rainfall data spatial density. The insufficient amount and unevenly distributed rainfall measurement instrument, are two main factors contributing to rainfall data special density problems. One of the very prominent methods to gain a larger amount of rainfall measurement location is using the image obtained from existing Closed Circuit Television (CCTV) spread over vast areas, especially in the Jakarta region. The approach to recognize and classify the rainfall in a certain area from the CCTV image used in this research is the Convolutional Neural Network (CNN) method. The image data was taken from CCTV located in Kamal, Kalideres, West Jakarta. The images taken is split into two categories, the one that shows a rainy day and the one that shows a clear day. These two categories of images will be used as sample data to train CNN, an effort to obtain a suitable model. By using the CNN method, it's possible to recognize and classify the rainfall condition within an image based on the model. Python is an open-source programming language that widely used nowadays to run CNN. The image classification using this CNN, scored approximately 98.30% of accuracy, which means that the model is optimal to recognize and classify rainfall conditions in a certain area based on the CCTV images.