Bayesian U-Net: Estimating Uncertainty in Semantic Segmentation of Earth Observation Images

Dechesne, Clément; Lassalle, Pierre; Lefèvre, Sébastien

doi:10.3390/rs13193836

Cited by 28 publications

(11 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Samples from the posterior distribution approximated by such a model can be recovered quite easily by enabling the dropout layers not only at training time, but also at inference time. It has been shown that MC dropout can quantify model uncertainties well for remote sensing tasks like aerial image segmentation [38]. Recently, Hartmann et al [39] also successfully applied a Bayesian UNet for the segmentation of glaciers in SAR imagery.…”

Section: Quantifying Uncertainty With Contour Modelsmentioning

confidence: 99%

Deep Active Contour Models for Delineating Glacier Calving Fronts

Heidler

Mou

Loebel

et al. 2022

IGARSS 2022 - 2022 IEEE International Geoscience and Remote Sensing Symposium

View full text Add to dashboard Cite

This work has been accepted by IEEE TGRS for publication in a future issue. Choosing how to encode a realworld problem as a machine learning task is an important design decision in machine learning. The task of glacier calving front modeling has often been approached as a semantic segmentation task. Recent studies have shown that combining segmentation with edge detection can improve the accuracy of calving front detectors. Building on this observation, we completely rephrase the task as a contour tracing problem and propose a model for explicit contour detection that does not incorporate any dense predictions as intermediate steps. The proposed approach, called "Charting Outlines by Recurrent Adaptation" (COBRA), combines Convolutional Neural Networks (CNNs) for feature extraction and active contour models for the delineation. By training and evaluating on several large-scale datasets of Greenland's outlet glaciers, we show that this approach indeed outperforms the aforementioned methods based on segmentation and edgedetection. Finally, we demonstrate that explicit contour detection has benefits over pixel-wise methods when quantifying the models' prediction uncertainties.The project page containing the code and animated model predictions can be found at https://khdlr.github.io/COBRA/.

show abstract

Section: Quantifying Uncertainty With Contour Modelsmentioning

confidence: 99%

Deep Active Contour Models for Delineating Glacier Calving Fronts

Heidler

Mou

Loebel

et al. 2022

IGARSS 2022 - 2022 IEEE International Geoscience and Remote Sensing Symposium

View full text Add to dashboard Cite

show abstract

“…12 In general, solutions fall under two approaches: measuring DNN model-related uncertainty (i.e., epistemic uncertainty) and measuring data-related uncertainty (i.e., aleatoric uncertainty). 13 A representative technique for model-related uncertainty evaluation uses Bayesian neural networks, 12,14 which replace a single weight in the DNN with a probability distribution to produce a probabilistic prediction.The variation in model parameters can be translated to the variation in segmentation predictions, yielding an error margin for each pixel. Monte Carlo dropout 15,16 represents an alternative technique.…”

Section: Introductionmentioning

confidence: 99%

Quantifying U‐Net uncertainty in multi‐parametric MRI‐based glioma segmentation by spherical image projection

Yang,

Lafata,

Vaios

et al. 2023

Medical Physics

View full text Add to dashboard Cite

BackgroundUncertainty quantification in deep learning is an important research topic. For medical image segmentation, the uncertainty measurements are usually reported as the likelihood that each pixel belongs to the predicted segmentation region. In potential clinical applications, the uncertainty result reflects the algorithm's robustness and supports the confidence and trust of the segmentation result when the ground‐truth result is absent. For commonly studied deep learning models, novel methods for quantifying segmentation uncertainty are in demand.PurposeTo develop a U‐Net segmentation uncertainty quantification method based on spherical image projection of multi‐parametric MRI (MP‐MRI) in glioma segmentation.MethodsThe projection of planar MRI data onto a spherical surface is equivalent to a nonlinear image transformation that retains global anatomical information. By incorporating this image transformation process in our proposed spherical projection‐based U‐Net (SPU‐Net) segmentation model design, multiple independent segmentation predictions can be obtained from a single MRI. The final segmentation is the average of all available results, and the variation can be visualized as a pixel‐wise uncertainty map. An uncertainty score was introduced to evaluate and compare the performance of uncertainty measurements.The proposed SPU‐Net model was implemented on the basis of 369 glioma patients with MP‐MRI scans (T1, T1‐Ce, T2, and FLAIR). Three SPU‐Net models were trained to segment enhancing tumor (ET), tumor core (TC), and whole tumor (WT), respectively. The SPU‐Net model was compared with (1) the classic U‐Net model with test‐time augmentation (TTA) and (2) linear scaling‐based U‐Net (LSU‐Net) segmentation models in terms of both segmentation accuracy (Dice coefficient, sensitivity, specificity, and accuracy) and segmentation uncertainty (uncertainty map and uncertainty score).ResultsThe developed SPU‐Net model successfully achieved low uncertainty for correct segmentation predictions (e.g., tumor interior or healthy tissue interior) and high uncertainty for incorrect results (e.g., tumor boundaries). This model could allow the identification of missed tumor targets or segmentation errors in U‐Net. Quantitatively, the SPU‐Net model achieved the highest uncertainty scores for three segmentation targets (ET/TC/WT): 0.826/0.848/0.936, compared to 0.784/0.643/0.872 using the U‐Net with TTA and 0.743/0.702/0.876 with the LSU‐Net (scaling factor = 2). The SPU‐Net also achieved statistically significantly higher Dice coefficients, underscoring the improved segmentation accuracy.ConclusionThe SPU‐Net model offers a powerful tool to quantify glioma segmentation uncertainty while improving segmentation accuracy. The proposed method can be generalized to other medical image‐related deep‐learning applications for uncertainty evaluation.

show abstract

“…In recent years, with its unique deep feature expression capacity, deep learning has provided new ideas for remote sensing image processing including semantic segmentation [12,13], object detection [14,15], image matching [16,17], etc. Many remote sensing image change detection methods based on deep learning have been proposed.…”

Section: Introductionmentioning

confidence: 99%

Feature Decomposition-Optimization-Reorganization Network for Building Change Detection in Remote Sensing Images

Zhou

Zhu

et al. 2022

Remote Sensing

View full text Add to dashboard Cite

Building change detection plays an imperative role in urban construction and development. Although the deep neural network has achieved tremendous success in remote sensing image building change detection, it is still fraught with the problem of generating broken detection boundaries and separation of dense buildings, which tends to produce saw-tooth boundaries. In this work, we propose a feature decomposition-optimization-reorganization network for building change detection. The main contribution of the proposed network is that it performs change detection by respectively modeling the main body and edge features of buildings, which is based on the characteristics that the similarity between the main body pixels is strong but weak between the edge pixels. Firstly, we employ a siamese ResNet structure to extract dual-temporal multi-scale difference features on the original remote sensing images. Subsequently, a flow field is built to separate the main body and edge features. Thereafter, a feature optimization module is designed to refine the main body and edge features using the main body and edge ground truth. Finally, we reorganize the optimized main body and edge features to obtain the output results. These constitute a complete end-to-end building change detection framework. The publicly available building dataset LEVIR-CD is employed to evaluate the change detection performance of our network. The experimental results show that the proposed method can accurately identify the boundaries of changed buildings, and obtain better results compared with the current state-of-the-art methods based on the U-Net structure or by combining spatial-temporal attention mechanisms.

show abstract

Bayesian U-Net: Estimating Uncertainty in Semantic Segmentation of Earth Observation Images

Cited by 28 publications

References 28 publications

Deep Active Contour Models for Delineating Glacier Calving Fronts

Deep Active Contour Models for Delineating Glacier Calving Fronts

Quantifying U‐Net uncertainty in multi‐parametric MRI‐based glioma segmentation by spherical image projection

Feature Decomposition-Optimization-Reorganization Network for Building Change Detection in Remote Sensing Images

Contact Info

Product

Resources

About