Cloud Detection from AWiFS Imagery using Deep Learning

Cloud detection is an essential pre-processing step to delineate cloud-affected regions for many applications involving optical remote sensing data. This paper presents a deep learning segmentation approach using attention-based U-net architecture, which is used to detect clouds for the Indian Remote Sensing (IRS) Resourcesat series Advance Wide Field Sensor (AWiFS) imagery. The network can identify cloud cover from the top-of-atmosphere reflectance scene using an Encoder and Decoder along with an attention mechanism. The attention mechanism optimizes the computational resources by minimizing activations, and the Decoder can pay attention only to the relevant part of the scene. This approach results in an overall accuracy of 87.3%. The result demonstrates that the proposed method can be considered a benchmark for AWiFS imagery.