ACLNet: an attention and clustering-based cloud segmentation network

We propose a novel deep learning model named ACLNet, for cloud segmentation from ground images. ACLNet uses both neural network and machine (ML) algorithm to extract complementary features. Specifically, it EfficientNet-B0 as the backbone, "`a trous spatial pyramid pooling" (ASPP) learn at multiple receptive fields, "global attention module" (GAM) finegrained details image. also k-means clustering boundaries more precisely. is effective daytime nighttime It provides lower error rate, higher recall F1-score than state-of-art models. The source-code of available here: https://github.com/ckmvigil/ACLNet.