ACLNet: an attention and clustering-based cloud segmentation network

Abstract

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

Infrared Radiometric Image Classification and Segmentation of Cloud Structure Using Deep-learning Framework for Ground-based Infrared Thermal Camera Observations

“…CloudDeepLabV3+ (Li et al, 2023) designs a lightweight ground-based cloud image adaptive segmentation method that integrates multi-scale feature aggregation and multi-level attention feature enhancement. ACLNet (Makwana et al, 2022) uses EfficientNet-B0 as the backbone, "à trous spatial pyramid pooling" (ASPP see Chen et al 2017) to learn at multiple receptive fields, and global attention module (GAM see Liu et al 2021) to extract fine-grained details from the image. It provides a lower error rate, higher recall, and higher F1-score than state-of-the-art cloud segmentation models.…”

“…Therefore, to determine cloud presence in infrared images, deep convolutional neural networks (CNNs) appear to be a viable approach to process images in real-time. Multiple models relying on CNNs have been developed such as CloudSegnet (Dev et al, 2019a), CloudU-Net (Shi et al, 2021b) CloudU-Netv2 (Shi et al, 2021a), SegCloud (Xie et al, 2020), TransCloud-Seg (Liu et al, 2022), CloudDeepLabV3 (Li et al, 2023), ACLNet (Makwana et al, 2022), DeepCloud (Ye et al, 2017), CloudRaednet (Shi et al, 2022), DMNet (Zhao et al, 2022) and DPNet Zhang et al (2022). Nonetheless, these methodologies exclusively address RGB-colored images (Li et al, 2011;Dev et al, 2016).…”