Semantic Segmentation of Clouds in Satellite Imagery Using Deep Pre-trained U-Nets

Gonzales, Cindy; Sakla, Wesam

doi:10.1109/aipr47015.2019.9174594

Cited by 15 publications

(10 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The values in the confusion matrix were labeled as true positive (TP), true negative (TN), false positive (FP), and false negative (FN). Since our study had an imbalanced class distribution, it was necessary to select evaluation metrics which punish false predictions and therefore measure the relevance and completeness of the model predictions [12]. For this reason, we used the metrics precision and recall, called respectively the complement of commission errors and the complement of emission errors by the remote sensing community.…”

Section: Evaluation Metricsmentioning

confidence: 99%

Towards Comparative Physical Interpretation of Spatial Variability Aware Neural Networks: A Summary of Results

Gupta¹,

Molnar²,

Luo³

et al. 2021

Preprint

View full text Add to dashboard Cite

Given Spatial Variability Aware Neural Networks (SVANNs), the goal is to investigate mathematical (or computational) models for comparative physical interpretation towards their transparency (e.g., simulatibility, decomposability and algorithmic transparency). This problem is important due to important use-cases such as reusability, debugging, and explainability to a jury in a court of law. Challenges include a large number of model parameters, vacuous bounds on generalization performance of neural networks, risk of overfitting, sensitivity to noise, etc., which all detract from the ability to interpret the models. Related work on either modelspecific or model-agnostic post-hoc interpretation is limited due to a lack of consideration of physical constraints (e.g., mass balance) and properties (e.g., second law of geography). This work investigates physical interpretation of SVANNs using novel comparative approaches based on geographically heterogeneous features. The proposed approach on feature-based physical interpretation is evaluated using a case-study on wetland mapping. The proposed physical interpretation improves the transparency of SVANN models and the analytical results highlight the trade-off between model transparency and model performance (e.g., F1-score). We also describe an interpretation based on geographically heterogeneous processes modeled as partial differential equations (PDEs).

show abstract

Section: Evaluation Metricsmentioning

confidence: 99%

Towards Comparative Physical Interpretation of Spatial Variability Aware Neural Networks: A Summary of Results

Gupta¹,

Molnar²,

Luo³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…In the field of remote sensing, the use of new artificial intelligence techniques is also an active topic (Yuan et al 2020;Boukabara et al 2019). One prominent example is satellite-based cloud property retrieval (Weng et al 2018;Xie et al 2017;Gonzales and Sakla 2019;Lee et al 2020).…”

Section: Introductionmentioning

confidence: 99%

Convective-Scale Assimilation of Cloud Cover from Photographs Using a Machine Learning Forward Operator

Reinhardt

Schoger

Kurzrock³

et al. 2023

Artificial Intelligence for the Earth Systems

View full text Add to dashboard Cite

This paper presents an innovational way of assimilating observations of clouds into the ICOsahedral Nonhydrostatic weather forecasting model for regional scale, ICON-D2, which is operated by Deutscher Wetterdienst (DWD). A convolutional neural network (CNN) is trained to detect clouds in camera photographs. The network’s output is a greyscale picture, in which each pixel has a value between 0 and 1, describing the probability of the pixel belonging to a cloud (1) or not (0). By averaging over a certain box of the picture a value for the cloud cover of that region is obtained. A forward operator is built to map an ICON model state into the observation space. A three dimensional grid in the space of the camera’s perspective is constructed and the ICON model variable cloud cover (CLC) is interpolated onto that grid. The maximum CLC along the rays that fabricate the camera grid, is taken as a model equivalent for each pixel. After superobbing, monitoring experiments have been conducted to compare the observations and model equivalents over a longer time period, yielding promising results. Further we show the performance of a single assimilation step as well as a longer assimilation experiment over a time period of six days which also yields good results. These findings are a proof of concept and further research has to be invested before these new innovational observations can be assimilated operationally in any numerical weather prediction (NWP) model.

show abstract

“…The algorithm provides the image directly to the convolutionary neural networks, and the algorithm removes the most important features of the image [10]. In the findings indicate that CNN functionality extracted from profound learning must be taken into account in the most visual recognition tasks [11]. To identify cloud image classifications, priority knowledge is needed, which is learned through identified cloud image types with a similar composition.…”

Section: Introductionmentioning

confidence: 99%

“…To identify cloud image classifications, priority knowledge is needed, which is learned through identified cloud image types with a similar composition. The data sets of the CCSN (Cirrus Cumulus Stratus Nimbus) divides into 11…”

Section: Introductionmentioning

confidence: 99%

Convolutional Neural Network for object Identification and Detection

Andini

Rouza

Fimawahib

et al. 2022

J. Phys.: Conf. Ser.

View full text Add to dashboard Cite

The goal of this study is to use a Convolutional Neural Network to find the optimum architectural model for classifying cloud images. Cirrus Cumulus Stratus Nimbus uses a source dataset that includes 11 cloud classifications and 2545 cloud photos (CCSN). In this study, the best Convolutional Neural Network is retrained almost fast by transferring education from Google’s basic design. Based on the modified Googlenet architecture, the training and testing phases of the classification process are divided into two. The dataset is separated into three sections during the training phase: 70% of the training data, 15% of the validation data, and 15% of the test data. There are two trials to categorize cloud photographs during the test phase, one of which has ten cloud kinds that can be randomly chosen. The precision achieved throughout the training was 44.5%, according to the findings. The results of the two tests are 75%, with an average error of 0.2. In the testing phase, the percentage is 75%.

show abstract

Semantic Segmentation of Clouds in Satellite Imagery Using Deep Pre-trained U-Nets

Cited by 15 publications

References 10 publications

Towards Comparative Physical Interpretation of Spatial Variability Aware Neural Networks: A Summary of Results

Towards Comparative Physical Interpretation of Spatial Variability Aware Neural Networks: A Summary of Results

Convective-Scale Assimilation of Cloud Cover from Photographs Using a Machine Learning Forward Operator

Convolutional Neural Network for object Identification and Detection

Contact Info

Product

Resources

About