ClouDet: A Dilated Separable CNN-Based Cloud Detection Framework for Remote Sensing Imagery

Guo, Hongwei; Bai, Hongyang; Qin, Weiwei

doi:10.1109/jstars.2021.3114171

Cited by 13 publications

(3 citation statements)

References 45 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Tao Ye et al [38] proposed a lightweight fusion detector (LFD) based on YOLOv3 and successfully deployed it on an embedded device for obstacle avoidance in railway transit systems. Hongwei Guo et al [39] applied a multiple features fusion strategy, dilated separable convolution and context pooling into cloud detection. This achieved lightweight but fairly good performance under various complex conditions.…”

Section: Related Workmentioning

confidence: 99%

Lightweight-aware mixed MobileNetv2-YOLOv4 model for helmet detection on construction sites

Song,

Wang,

Wang

et al. 2023

Preprint

View full text Add to dashboard Cite

Helmet detection based on video monitoring often encounters challenges such as large memory requirements, high complexity of models and difficulty in applying on ubiquitous terminals on construction sites. To address the above problems, a lightweight-aware safety helmet detection method for multi-platforms based on mixed MobileNetv2-YOLOv4 is proposed. First, a lightweight MobileNetv2 backbone is introduced to reduce the number of parameters. Then, a simple model pipeline is achieved by combining the lightweight backbone and widely used YOLOv4 detection head with Feature Pyramid Network (FPN). Also, modifications are made to both the backbone and detection head for a better trade-off between lightweight and performance (speed and accuracy). The whole model is coded under the Darknet framework in the C language and is successfully deployed on a mobile device. Finally, massive experiments and a case study in the real condition are performed to validate the feasibility of the model. The model is lightweight with only 10.4 MB and 3.39 BFLOPs while reaching comparable performance against other detection methods. The model reaches a mAP of 73.1% on the PASCAL VOC 2007 + 2012 dataset and 89.3% on the GDUT-HWD dataset, outperforming YOLOv4, YOLOv5 and Faster R-CNN in speed and model size. The results show that the model has the universality to be implemented in various scenes and the flexibility to be deployed on various devices including workstations, normal PCs and mobile devices. And the safety management level on construction sites is also improved through the successful deployment.

show abstract

Section: Related Workmentioning

confidence: 99%

Lightweight-aware mixed MobileNetv2-YOLOv4 model for helmet detection on construction sites

Song,

Wang,

Wang

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…With respect to the cloud detector algorithm, s2cloudless was chosen for being commonly deployed, easily applicable and performing well [37], [38]. However, many alternative approaches exist [39], [40], [41], [42], [35], whose variable sensitivity thresholds may result in qualitatively different cloud masks and thus different downstream analysis results. The chosen s2cloudless algorithm is reported to show a fair "balance (within 10%) between commission and omission errors" [38], which may avoid any one-sided biases to either false alarms or misses of clouds in our subsequent analysis.…”

Section: ) Outliers As Distractorsmentioning

confidence: 99%

Explaining the Effects of Clouds on Remote Sensing Scene Classification

Gawlikowski

Ebel

Schmitt

et al. 2022

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Most of Earth is covered by haze or clouds, impeding the constant monitoring of our planet. Preceding works have documented the detrimental effects of cloud coverage on remote sensing applications and proposed ways to approach this issue. However, up-to-now, little effort has been spent on understanding how exactly atmospheric disturbances impede the application of modern machine learning methods to Earth observation data. Specifically, we consider the effects of haze and cloud coverage on a scene classification task. We provide a thorough investigation of how classifiers trained on cloud-free data fail once they encounter noisy imagery -a common scenario encountered when deploying pretrained models for remote sensing to real use cases. We show how and why remote sensing scene classification suffers from cloud coverage. Based on a multi-stage analysis, including explainability approaches applied to the predictions, we work out four different types of effects that clouds have on the scene prediction. The contribution of our work is to deepen the understanding of the effects of clouds on common remote sensing applications, and consequently guide the development of more robust methods.

show abstract

“…With the development of deep learning in the field of computer vision, networks used for semantic segmentation of remote sensing images of clouds and cloud shadows have also made significant progress. Guo et al proposed a lightweight fully convolutional neural network (ClouDet) [25], which uses atrous separable convolution to improve the efficiency and accuracy of the network, and introduces multi-scale feature fusion to deal with cloud shadows of different scales. Yan et al used the pyramid pooling module to extract context information.…”

Section: Introductionmentioning

confidence: 99%

HyCloudX: A Multibranch Hybrid Segmentation Network With Band Fusion for Cloud/Shadow

Hu,

Weng,

Xia

et al. 2024

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Semantic segmentation of cloud and shadow is an important task in remote sensing and atmospheric science. However, the complexity of cloud/shadow shapes, and noise disturbances (such as snow and ice, buildings, complex backgrounds, and atmospheric optics) make this task challenging. The traditional deep network has good details and generalization due to its local feature extraction ability and spatial invariance, but it is relatively weak in dealing with global context information, which leads to misjudgment and missed judgment in complex scenes. The Transformer can effectively capture long-distance dependencies through the self-attention mechanism, but it may have challenges in extracting local image features and maintaining spatial consistency, resulting in loss of detail information and insufficient generalization. This paper proposes a hybrid branch semantic segmentation network composed of convolutional network and Transformer in parallel. A series of modules are designed to solve the problems of lack of multiscale feature extraction and insufficient fusion in some of the Convolution-Transformer hybrid networks. In particular, the network utilizes the rich information in auxiliary bands such as near-infrared to improve the segmentation performance, so that the network can process a wider range of data and improve generalization. Experimental results on CloudSEN-12, 38-Cloud, and SPRCS-Val show that our network outperforms existing methods. After introducing the band fusion branch (HyCloudX), the network improves the segmentation performance and generalization, especially in the case of complex noise interference.

show abstract

ClouDet: A Dilated Separable CNN-Based Cloud Detection Framework for Remote Sensing Imagery

Cited by 13 publications

References 45 publications

Lightweight-aware mixed MobileNetv2-YOLOv4 model for helmet detection on construction sites

Lightweight-aware mixed MobileNetv2-YOLOv4 model for helmet detection on construction sites

Explaining the Effects of Clouds on Remote Sensing Scene Classification

HyCloudX: A Multibranch Hybrid Segmentation Network With Band Fusion for Cloud/Shadow

Contact Info

Product

Resources

About