Pith. sign in

REVIEW 4 major objections 7 minor 29 references

Deep learning networks for selection of persistent scatterer pixels in multi-temporal SAR interferometric processing

T0 review · 4 major / 7 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A spatio-temporal deep network selects more reliable radar persistent scatterers than the standard StaMPS method.

desk verdict New application with a real speedup, but the headline accuracy is circular and the improvement claim needs independent validation before it can be believed. read the letter →

arxiv 1909.01868 v3 pith:5QPG5L7J submitted 2019-09-04 eess.IV cs.LGstat.ML

Ashutosh Tiwari , Avadh Bihari Narayan , Onkar Dikshit This is my paper · ORCID
classification eess.IVcs.LGstat.ML
keywords persistentscattererinterferometryPSpixelselectiondeeplearningconvolutionalLSTMsemanticsegmentationinterferometricphasestackSentinel-1deformationmonitoring
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper claims that a deep learning network can replace the slow, iterative PS pixel-selection step in multi-temporal InSAR with a fast pixel-wise classification. Two architectures are proposed, CNN-ISS and CLSTM-ISS, trained on roughly 10,000 real interferometric image patches labelled by the StaMPS algorithm, an open-access persistent-scatterer processing chain. On the unseen Kathmandu test site, CLSTM-ISS reports 93.50% validation accuracy, selects more PS pixels in urban and lengthening areas, selects fewer in forest and water, and completes selection in 8.2 minutes compared with 108.2 minutes for StaMPS. The central claim is that CLSTM-ISS, by learning both spatial and temporal phase behaviour, produces a higher density of reliable PS pixels than either StaMPS or the spatial-only CNN-ISS.

What carries the argument

The load-bearing component is the convolutional LSTM (convlstm) cell, which replaces the matrix multiplications inside an LSTM with convolutions over image neighbourhoods. This lets the network carry a cell state through the ten interferograms and learn spatial phase patterns and their temporal coherence together. The full CLSTM-ISS stacks two convlstm layers, one convolution layer, dropout, and a fully connected output, and it is trained with an f1-loss in which the PS class receives a weight of 200 against 1 for non-PS. Pooling layers are deliberately omitted because PS pixels are isolated and pooling would bias learning toward spatially correlated nuisance phase components.

What would settle it

Run CLSTM-ISS on a fresh non-urban site with corner reflectors or GPS-validated displacement, and compare phase noise and velocity error on the pixels CLSTM-ISS adds beyond StaMPS; if the added pixels are not more phase-stable, the improvement claim collapses.

Watch

Extended reading notes

Core claim

The central discovery claimed is that a convolutional long short-term memory network, CLSTM-ISS, which treats the interferogram stack as an image time series, classifies PS and non-PS pixels better than the StaMPS algorithm and better than the spatial-only CNN-ISS. On the unseen Kathmandu test set, CLSTM-ISS achieves 93.50% validation accuracy versus 89.21% for CNN-ISS, selects 192,177 PS pixels versus 38,286 for StaMPS, and 97.01% of its PS pixels pass the STIP greater-than-35 reliability threshold versus 92.49% for StaMPS and 80.10% for CNN-ISS. In area-wise terms it detects the highest PS density in man-made areas (52.9%) and lengthening areas (48.1%) and the lowest in forest and vegetation (3.4%). The authors interpret this as CLSTM-ISS learning the true spatio-temporal coherence of scatterers rather than simply reproducing its StaMPS training labels.

Load-bearing premise

The comparison rests on assuming that a pixel is good if it lies where the slope-orientation R-index and land-cover map say PS pixels should occur and if more than 35 similar-time-series neighbours surround it; if those proxies are wrong, the claim that CLSTM-ISS outperforms StaMPS is unsupported.

Editorial extensions

If this is right

  • A trained CLSTM-ISS could cut PS selection from hours or days to minutes, making near-real-time deformation monitoring feasible for repeated satellite acquisitions.
  • The higher density of reliable PS pixels should improve phase unwrapping and produce clearer velocity maps, since CLSTM-ISS retains the StaMPS-like displacement pattern while adding coherent points.
  • Because CLSTM-ISS includes 78.44% of the pixels StaMPS selects, it is unlikely to lose the information current processing chains rely on, while adding new coherent pixels in man-made and lengthening terrain.
  • The method is not tied to StaMPS's proprietary logic: since labels came from an open-access algorithm and the input is a standard interferometric stack, the same training scheme can be adapted as better training labels become available.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The reliability comparison rests on STIP, a coherence measure introduced by the same group; an independent check using corner reflectors or GPS-validated deformation is needed before generalising the "more reliable" claim.
  • The paper demonstrates ten-interferogram stacks from Sentinel-1; generalisation to other stack lengths, sensors, or orbital geometries is plausible but not demonstrated, and would need its own training data.
  • The clean separation of forest, water, and uncropped land suggests the same spatio-temporal architecture could segment other decorrelated terrain types, such as snow, cropland with seasonal cycles, or wetlands, if labelled stacks were available.
  • If the speed advantage holds, the practical bottleneck in MT-InSAR would shift from PS selection to phase unwrapping and time-series inversion, which the paper does not replace.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 7 minor

Summary. The paper proposes two deep-learning architectures, CNN-ISS and CLSTM-ISS, for pixel-wise classification of persistent scatterer (PS) and non-PS pixels in multi-temporal InSAR interferograms. The networks are trained on roughly 10,000 100-by-100 interferometric image patches from three Sentinel-1 study sites, using StaMPS-derived PS labels as ground truth, and tested on an unseen Kathmandu dataset. The authors report that CLSTM-ISS achieves 93.50% validation accuracy versus 89.21% for CNN-ISS, produces a higher PS density than StaMPS, and runs in minutes compared with 108 minutes for StaMPS. The claimed superiority over StaMPS is based on a qualitative comparison using R-index and land-cover maps, and on a reliability analysis using the STIP metric introduced in the authors' prior work.

Significance. If the central claim were established, the contribution would be practically valuable: a trained network that selects PS pixels in near real time could remove a major computational bottleneck in MT-InSAR processing and support time-critical deformation monitoring. The paper also addresses a real problem, uses real-world multi-site training data, and explicitly acknowledges the absence of true labels. However, the evaluation does not support the central claim of improved classification over StaMPS: the headline validation accuracies measure agreement with the same StaMPS-generated labels used for training, while the external checks are either heuristic (R-index plus land-cover) or based on the authors' own STIP metric with an unvalidated threshold. The reported gains in PS density and STIP counts could plausibly be explained by an over-permissive classifier rather than by better phase-stability selection. No code, trained weights, or repeated experimental runs are provided, limiting reproducibility.

major comments (4)
  1. [Section 4, Table 5] The reported validation accuracy of 93.50% for CLSTM-ISS measures agreement with StaMPS-generated training labels, not correctness of PS selection. Since the same labels define the training target, high validation accuracy cannot support the claim that CLSTM-ISS improves classification over StaMPS. The paper itself acknowledges this in Section 4: 'with the lack of true labels, using these metrics for quality evaluation would provide an incorrect estimate of the classifier performance.' An independent ground truth or a strictly separate reference method is required before any claim of improvement can be made.
  2. [Section 4, Figures 6-8 and Table 4] The qualitative evaluation uses a combination of the R-index and a classified land-cover image as the reference for where PS pixels 'should' occur. This is a heuristic proxy, not an independent validation of phase stability. The assumption that more PS pixels in man-made and lengthening areas and fewer in forest and river areas necessarily indicates better classification is not quantitatively justified, and no uncertainty or sensitivity analysis is given. Consequently, statements such as 'CLSTM-ISS outperformed the other two methods' (Section 4, after Table 4) are not supported by the evidence presented.
  3. [Section 5, STIP analysis] The STIP reliability check relies on a metric introduced by the same authors in Narayan et al. (2018a, 2018b), and the STIP>35 threshold is asserted without independent validation for the Kathmandu data. Given that CLSTM-ISS selects roughly five times as many PS pixels as StaMPS (192,177 versus 38,286 in Table 5), the larger absolute number of STIP>35 pixels (186,435 versus 35,413) could simply reflect a more permissive selection threshold. Without an analysis of false positives against known non-PS targets, the higher STIP count does not establish that CLSTM-ISS selects more true PS pixels.
  4. [Section 5, Figures 9-11] The velocity maps and time-series displacement comparisons are evaluated qualitatively by visual pattern matching. The similarity of CLSTM-ISS to StaMPS is not a meaningful benchmark because StaMPS is the source of the training labels. No quantitative metric (e.g., RMS difference against independent deformation measurements), no error bars, and no repeated experimental runs are reported. This leaves the central claim that CLSTM-ISS improves 'reliable PS density' without a rigorous, independent basis.
minor comments (7)
  1. [Section 3] The text states that 'random sampling was used to select test samples (images) from the training data,' which conflicts with the description of the Kathmandu dataset as an unseen test set in Table 1 and the following paragraph. Please clarify whether random sampling refers to validation samples only.
  2. [Table 2] The filter counts in Table 2 appear inconsistent: layer '(conv+BN)4+relu' is listed with 32 filters but an output dimension of 64 channels, and similar inconsistencies appear in Table 3 (including a duplicated row label '(convlstm+BN)2+relu'). Please verify the architecture tables and the corresponding text.
  3. [Equation (11)] Equation (11) is garbled in the manuscript, making it impossible to verify the ConvLSTM gate equations. A clean, correctly typeset version of the equations is needed.
  4. [Section 4.3] The description of the f1-loss and 'probabilistic' accuracy is confusing: stating that a true non-PS pixel with predicted probability 0.4 counts as 0.6 false positive and 0.4 true negative does not match standard definitions of accuracy or loss. Please define the loss exactly.
  5. [Table 6 and Figure 10] Table 6 appears to contain no visible entries in the manuscript, although the text gives the key numbers (92.49%, 80.10%, 97.01%). The numerical values should be presented in the table itself.
  6. [Section 5, STIP threshold] The statement that STIP>35 is 'a threshold generally used to define a coherent PS pixel' is made without a citation. Please provide a reference or supporting analysis for this threshold.
  7. [Sections 3 and 4.3] The observation that more than 95% of pixels are non-PS is repeated nearly verbatim in Section 3 and Section 4.3; consider keeping it in one place.

Circularity Check

2 steps flagged · score 6.0 of 10

The 93.50% accuracy is measured against StaMPS-derived training labels, and the STIP reliability check is taken from the authors' own prior work; the claimed improvement over StaMPS is therefore not independently established.

  1. fitted input called prediction [Abstract; Section 3 (training labels); Section 4, Table 5]
    "The networks were trained using training labels obtained from the Stanford method for Persistent Scatterer Interferometry (StaMPS) algorithm. However, pixel selection results, when compared to a combination of R-index and a classified image of the test dataset, reveal that CLSTM-ISS estimates improved the classification of PS and non-PS pixels compared to those of StaMPS and CNN-ISS. The predicted results show that CLSTM-ISS reached an accuracy of 93.50%, higher than that of CNN-ISS (89.21%)."

    The 93.50% validation accuracy is computed against StaMPS-generated labels, which are the same labels used as training targets for the network. It therefore measures how closely CLSTM-ISS reproduces StaMPS's pixel decisions, not whether those decisions are correct. Since the paper's central claim is that CLSTM-ISS 'improved' PS classification over StaMPS, reporting this fit-to-StaMPS accuracy as a 'predicted result' is a fitted input called a prediction: agreement with the training-label source is presented as evidence of superiority over that source.

  2. self citation load bearing [Section 5, 'Qualitative analysis of PS selection' (STIP analysis and Table 6)]
    "In order to statistically assess the quality of PS pixels detected using the proposed architectures, a reliable index, known as similar time series interferometric pixel (STIP) was employed. STIP is a measure of noise in phase time series (Narayan et al., 2018a; 2018b). ... It is found that 92.49% of the total PS pixels selected by the StaMPS method have more than 35 STIP (a threshold generally used to define a coherent PS pixel) in the neighbourhood defined by a window of 25×5 pixels (Table 6)."

    The quantitative support for the claim that 'CLSTM-ISS detected the most number of coherent PS pixels' rests entirely on the STIP metric, whose source is prior work by the same authors (Narayan, Tiwari, Dwivedi, Dikshit). The paper supplies no independent validation of STIP as ground truth for phase stability; it simply applies a threshold 'generally used to define a coherent PS pixel.' Thus the reliability/density advantage of CLSTM-ISS over StaMPS is load-bearing on a self-citation rather than on an external benchmark. If STIP is not accepted as a valid proxy, the outperformance claim is unsupported.

full rationale

The paper's training procedure is self-consistent but its central evaluation is not independent. The network is trained on StaMPS labels, so the reported validation accuracy (93.50%) measures agreement with StaMPS, not correctness against true PS phase stability; the paper itself concedes in Section 4 that 'with the lack of true labels, using these metrics for quality evaluation would provide an incorrect estimate of the classifier (network) performance.' The subsequent qualitative evaluation substitutes a combination of R-index and a classified image, which is a heuristic reference, and the quantitative STIP reliability analysis is taken from the authors' own prior work (Narayan et al. 2018a/b). None of these constitutes an independent ground truth, so the abstract's claim that CLSTM-ISS 'improved the classification of PS and non-PS pixels compared to those of StaMPS' is not established by an external benchmark. This is partial circularity rather than full equivalence: the network architectures, the training on real interferograms, and the computational-time comparison are independent contributions, and STIP, while self-cited, is at least an explicit and reproducible quantitative criterion. Score 6 reflects that the central performance claim partially reduces to agreement with StaMPS training labels and to self-cited reliability criteria.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

Central evaluation relies on three inputs beyond the paper's own experiments: StaMPS-generated labels (Section 3), R-index and land-cover heuristics (Section 4), and STIP from the authors' prior work (Section 5). The first creates a circular baseline, the second is qualitative, and the third is self-referential.

free parameters (4)
  • PS class weight = 200
    Hand-set weight for the PS class in the f1-loss; directly controls predicted PS density and therefore the central comparison.
  • Non-PS class weight = 1
    Hand-set, paired with the PS class weight of 200.
  • StaMPS coherence threshold for training labels = not stated (described as 'higher coherence threshold')
    Chosen to minimize false positives in training labels; this choice determines what the network learns as PS.
  • STIP reliability threshold = 35
    Threshold from the authors' prior work used to declare a PS pixel reliable; used in the claim that 97.01% of CLSTM-ISS pixels are coherent.
assumptions (4)
  • domain assumption StaMPS selections are a sufficient training signal for true PS pixels.
    Section 3 uses StaMPS output as ground-truth labels; if StaMPS misses true PS pixels, the network inherits that bias.
  • domain assumption R-index values between 0.5 and 0.9 and man-made land cover indicate high PS probability.
    Section 4 uses this heuristic as the qualitative reference for judging where PS pixels should appear.
  • domain assumption STIP > 35 in a 25 by 5 neighborhood identifies a coherent PS pixel.
    Section 5 uses STIP from Narayan et al. 2018a/b, authored by the same investigators, to declare CLSTM-ISS pixels reliable.
  • standard math Probabilistic approximation of the non-differentiable F1 loss is a valid training objective.
    Section 4.3 converts discrete labels into probabilities to compute a differentiable F1-like loss, a standard approximation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Deep learning networks for selection of persistent scatterer pixels in multi-temporal SAR interferometric processing." pith.science (2026). https://pith.science/paper/5QPG5L7J

@misc{pith2026190901868,
  author       = {Pith},
  title        = {Pith review of: Deep learning networks for selection of persistent scatterer pixels in multi-temporal SAR interferometric processing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5QPG5L7J}},
  note         = {Machine review of arXiv:1909.01868}
}
read the original abstract

In multi-temporal SAR interferometry (MT-InSAR), persistent scatterer (PS) pixels are used to estimate geophysical parameters, essentially deformation. Conventionally, PS pixels are selected on the basis of the estimated noise present in the spatially uncorrelated phase component along with look-angle error in a temporal interferometric stack. In this study, two deep learning architectures, namely convolutional neural network for interferometric semantic segmentation (CNN-ISS) and convolutional long short term memory network for interferometric semantic segmentation (CLSTM-ISS), based on learning spatial and spatio-temporal behaviour respectively, were proposed for selection of PS pixels. These networks were trained to relate the interferometric phase history to its classification into phase stable (PS) and phase unstable (non-PS) measurement pixels using ~10,000 real world interferometric images of different study sites containing man-made objects, forests, vegetation, uncropped land, water bodies, and areas affected by lengthening, foreshortening, layover and shadowing. The networks were trained using training labels obtained from the Stanford method for Persistent Scatterer Interferometry (StaMPS) algorithm. However, pixel selection results, when compared to a combination of R-index and a classified image of the test dataset, reveal that CLSTM-ISS estimates improved the classification of PS and non-PS pixels compared to those of StaMPS and CNN-ISS. The predicted results show that CLSTM-ISS reached an accuracy of 93.50%, higher than that of CNN-ISS (89.21%). CLSTM-ISS also improved the density of reliable PS pixels compared to StaMPS and CNN-ISS and outperformed StaMPS and other conventional MT-InSAR methods in terms of computational efficiency.

Figures

Figures reproduced from arXiv: 1909.01868 by the authors.

Figure 2
Figure 2. Example images of the Training dataset used during network training. ( [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. shows the CNN-ISS network architecture. It contains an input layer, four convolutional (conv) layers (each followed by a batch normalization (BN) layer), a dropout layer and a fully connected (FC) layer as the final layer for pixel-wise classification (or segmentation). The output is a labelled image corresponding to the stack of interferograms, with each pixel showing a semantic label. The label is either PS or Non… view at source ↗
Figure 4
Figure 4. Basic structure of a convlstm cell. The convolution operation is shown by ‘*’ symbol.           11 11 11 1 ˆ ˆ ˆ tanh ˆ ˆ tanh t xi t yi t si t i t xf t yf t sf t f t t t t xs t ys t s t xo t yo t so t o t t t i W X W Y W S b f W X W Y W S b S f S i W X W Y b o W X W Y W S b Y o S                                    (11) The CLSTM-ISS architecture can be derived from th… view at source ↗
Figures from the paper (7 more)
Figure 5
Figure 5. Figure 5: Complete labelled images corresponding to the study sites used for training. (a), (b) and (c) represent labelled images for New Delhi, Nainital and Ahmedabad respectively. Point objects in each image represent the PS pixels [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]
Figure 6
Figure 6. Figure 6: Results of PS selection. (a) Study area containing different features used for performance evaluation and two test sites TS-I and TS-II for detailed analysis. Red dot indicates the location used for analysis of time series velocity estimates (b) computed R-index (c) Cl…
Figure 7
Figure 7. Figure 7: Detailed view of test site TS-I shown in [PITH_FULL_IMAGE:figures/full_fig_p018_7.png]
Figure 8
Figure 8. Figure 8: Detailed view of test site TS-II shown in [PITH_FULL_IMAGE:figures/full_fig_p020_8.png]
Figure 9
Figure 9. Figure 9: Number of STIP selected in the neighbourhood of PS pixels selected by (a) StaMPS, (b) CNN-ISS and (c) CLSTM-ISS. Estimated 1D LOS velocity of the PS pixels selected by (d) StaMPS, (e) CNN-ISS and (f) CLSTM-ISS [PITH_FULL_IMAGE:figures/full_fig_p023_9.png]
Figure 10
Figure 10. Figure 10: Histogram showing the frequency of the number of STIP detected by PS pixels for StaMPS, CNN-ISS and CLSTM-ISS methods. Green line marks the threshold STIP value [PITH_FULL_IMAGE:figures/full_fig_p024_10.png]
Figure 11
Figure 11. Figure 11: Time series displacement plot of the Kathmandu city generated using StaMPS, CNN-ISS and CLSTM-ISS. The location marked as a red dot in [PITH_FULL_IMAGE:figures/full_fig_p025_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 23 canonical work pages

  1. [5]

    Hooper, A., Segall, P., and Zebker, H.,

    Deep learning approach for automatic classification of ocular and cardiac artifacts in MEG Data, Journal of Engineering, DOI: 10.1155/2018/1350692. Hooper, A., Segall, P., and Zebker, H.,

    work page doi:10.1155/2018/1350692 2018

  2. [6]

    Hossain, M.D., and Chen, D.,

    Advances in Joint CTC-Attention based End-to- End Speech Recognition with a Deep CNN Encoder and RNN-LM, Accessible from: https://arxiv.org/pdf/1706.02737.pdf [Last accessed: 31/8/2019]. Hossain, M.D., and Chen, D.,

    arXiv 2019

  3. [7]

    Hu, Y., Li, X., Zhou, N., Yang, L., Peng, L., Xiao, S.,

    Segmentation for Object-Based Image Analysis (OBIA): A review of algorithms and challenges from remote sensing perspective, ISPRS Journal of Photogrammetry and Remote Sensing, 150, 115-134, https://doi.org/10.1016/j.isprsjprs.2019.02.009. Hu, Y., Li, X., Zhou, N., Yang, L., Peng, L., Xiao, S.,

    work page doi:10.1016/j.isprsjprs.2019.02.009

  4. [8]

    A sample update -based convolutional neural network framework for object detection in large-area remote sensing images, IEEE Geoscience and Remote Sensing Letters, DOI:10.1109/LGRS.2018.2889247. Intel,

    arXiv 2018

  5. [9]

    Jefkine,

    CNN: Back-propagation, Lecture 3, Available from: http://courses.cs.tau.ac.il/Caffe_workshop/Bootcamp/pdf_lectures/Lecture%203%20CNN%20- %20backpropagation.pdf, [Last accessed: 31/8/2019]. Jefkine,

    work page 2019

  6. [10]

    Johnson and Karpathy,

    Backpropagation In Convolutional Neural Networks, DeepGrid Organic Deep Learning, Available from: https://www.jefkine.com/general/2016/09/05/backpropagation-in- convolutional-neural-networks/ [Last accessed: 31/8/2019]. Johnson and Karpathy,

    work page 2016

  7. [11]

    Johnson, J.M

    Convolutional Neural Networks for Visual Recognition, Stanford University: Stanford CS231n notes, Available from: http://cs231n.github.io/ [Last accessed: 31/8/2019]. Johnson, J.M. and Khoshgoftaar, T.M.,

    work page 2019

  8. [13]

    Subsidence in the Kathmandu Basin, before and after the 2015 Mw 7.8 Gorkha Earthquake, Nepal Reveal ed from Small Baseline Subset -DInSAR Analysis

    “Subsidence in the Kathmandu Basin, before and after the 2015 Mw 7.8 Gorkha Earthquake, Nepal Reveal ed from Small Baseline Subset -DInSAR Analysis” GIScience & Remote Sensing, 55 (4), 604-621. Kumar, B., Pandey, G., Lohani, B., Mishra, S.C.,

    work page 2015

Show all 29 references
  1. [14]

    American Journal of Theoretical and Applied Statistics

    Minimax Estimation of the Parameter of Maxwell Distribution under Different Loss Functions. American Journal of Theoretical and Applied Statistics. 5, (4), 2016, 202-207. Li, M., Hu, Y., Zhao, N., Guo, L.,

    2016

  2. [15]

    Lotter, W., Kreiman, G., and Cox, D.,

    LPCCNet: A Lightweight Network for Point Cloud Classification, IEEE Geoscience and Remote Sensing Letters, DOI: 10.1109/LGRS.2018.2889472. Lotter, W., Kreiman, G., and Cox, D.,

    2018

  3. [16]

    Persistent Scatter Identification and Look-Angle Error Estimation Using Similar Time-Series Interferometric Pixels

    Deep predictive coding networks for video prediction and unsupervised learning, ICLR, Available from: https://arxiv.org/abs/1605.08104 [Last accessed: 31/8/2019]. Narayan, A. B., Tiwari, A., Dwivedi, R., and Dikshit, O., 2018a. “Persistent Scatter Identification and Look-Angle...

    2019 arXiv

  4. [17]

    Notti, D., Meisina, C., Zucca, F., Colombo, A.,

    F1 score, Evaluate classification models using F1 score, Machine Learning, Available from: https://www.ritchieng.com/machinelearning-f1-score/, [Last accessed: 31/8/2019]. Notti, D., Meisina, C., Zucca, F., Colombo, A.,

    2019

  5. [19]

    Sarkar, S.,

    Dealing with Unbalanced Classes in Machine Learning, deep ideas a blog on artificial intelligence, deep learning and cognitive science, Available from: http://www.deepideas.net/unbalanced-classes-machine-learning/ [Last accessed: 31/8/2019]. Sarkar, S.,

    2019

  6. [20]

    Schmitt, M., Hughes, L.H., Zhu, X.X.,

    Convolutional Neural Networks, Kharagpur: Indian Institute of Technology Kharagpur, Available from: http://cse.iitkgp.ac.in/~sudeshna/courses/DL17/CNN-22mar2017.pdf [Last accessed: 31/8/2019]. Schmitt, M., Hughes, L.H., Zhu, X.X.,

    2019

  7. [21]

    The Sentinel 1-2 dataset for deep learning in SAR- optical data fusion, ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume IV-1, 2018, Germany,

    2018

  8. [22]

    Skalski, P.,

    About Train, Validation and Test Sets in Machine Learning, Towards Data Science [Available from]: http://tarangshah.com/blog/2017 -12-03/train-validation-and-test-sets/ [Last accessed: 31/8/2019]. Skalski, P.,

    2017

  9. [23]

    Stack Overflow,

    Gentle Dive into Math Behind Convolutional Neural Networks, Towards Data Science, Medium Daily Digest, Available from https://towardsdatascience.com/gentle-dive-into-math- behind-convolutional-neural-networks-79a07dd44cf9 [Last accessed: 31/8/2019]. Stack Overflow,

    2019

  10. [24]

    Tiwari, S.,

    Imbalanced Classes in Convolutional Neural Networks, Available from: https://stackoverflow.com/questions/44504963/imbalanced-classes-in-convolutional-neural-networks [Last accessed: 31/8/2019]. Tiwari, S.,

    2019

  11. [25]

    Teunissen, P

    Activation functions in Neural Networks, GeeksforGeeks, Available from: https://www.geeksforgeeks.org/activation-functions-neural-networks/ [Last accessed: 31/8/2019]. Teunissen, P. J. G.,

    2019

  12. [26]

    Vincent, D and Visin, F.,

    Basic Overview of Convolutional Neural Network (CNN), Medium Daily Digest , Available from: https://medium.com/@udemeudofia01/basic -overview-of-convolutional-neural- network-cnn-4fcc7dbb4f17 [Last accessed: 31/8/2019]. Vincent, D and Visin, F.,

    2019

  13. [27]

    Wurm, M., Stark, T., Zhu, X.X., Weigand, M., Taubenbock, H.,

    A guide to convolution arithmetic for deep learning, Available from: https://arxiv.org/pdf/1603.07285.pdf, [Last accessed: 31/8/2019]. Wurm, M., Stark, T., Zhu, X.X., Weigand, M., Taubenbock, H.,

    2019 arXiv

  14. [28]

    Zhong, L., Hu, L., Zhou, H.,

    Semantic segmentation of slums in satellite images using transfer learning on fully convolutional neural networks, ISPRS Journal of Photogrammetry and Remote Sensing, 150, 59–69, https://doi.org/10.1016/j.isprsjprs.2019.02.006. Zhong, L., Hu, L., Zhou, H.,

    2019 doi

  15. [29]

    Zhu, X.X., Tuia, D., Mou, L., Xia, G.S., Zhang, L., Xu, F., Fraundorfer, F.,

    Deep learning based multi -temporal crop classification, Remote Sensing of Environment 221 (2019) 430–443, https://doi.org/10.1016/j.rse.2018.11.032. Zhu, X.X., Tuia, D., Mou, L., Xia, G.S., Zhang, L., Xu, F., Fraundorfer, F.,

    2019 doi

  16. [2005]

    Krishnan, P

    The STUN algorithm for Persistent Scatterer interferometry , Fringe 2005 Workshop, Frascati. Krishnan, P. V. S., Kim, D. J., Jung, J.,

    2005

  17. [2011]

    Sabinasz, D.,

    Models to predict Persistent Scatterers data distribution and their capacity to register movement along the slope, Fringe 2011 Workshop, 19–23, Ferrata, Italy: ESA/ESRIN. Sabinasz, D.,

    2011

  18. [2016]

    Ferretti, A., Prati, C., and Rocca, F.,

    How to set class weights for imbalanced classes in Keras? Available from: https://datascience.stackexchange.com/questions/13490/how-to-set-class-weights-for- imbalanced-classes-in-keras [Last accessed: 31/8/2019]. Ferretti, A., Prati, C., and Rocca, F.,

    2019

  19. [2017]

    Data Science Stack Exchange,

    What is the Difference Between Test and Validation Datasets?, Machine Learning Mastery, Available from: https:// machinelearningmastery.com/difference-test-validation-datasets/ [Last accessed: 31/8/2019]. Data Science Stack Exchange,

    2019

  20. [2018]

    Brownlee, J.,

    Detecting Volcano Deformation in InSAR using Deep learning, Available from: https://www.researchgate.net/publication/323510354_Detecting_Volcano_Deformation_in_InSAR_u sing_Deep_learning, [Last accessed: 31/8/2019]. Brownlee, J.,

    2019

  21. [2019]

    Brownlee, J.,

    A Gentle Introduction to Transfer Learning for Deep Learning, Machine Learning Mastery, Available from: https://machinelearningmastery.com/transfer-learning-for-deep-learning/ [Last accessed: 31/8/2019]. Brownlee, J.,

    2019

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.