Automatic Extraction of Road Networks by using Teacher-Student Adaptive Structural Deep Belief Network and Its Application to Landslide Disaster

Shin Kamada; Takumi Ichimura

REVIEW 3 major objections 6 minor 49 references

Automatic Extraction of Road Networks by using Teacher-Student Adaptive Structural Deep Belief Network and Its Application to Landslide Disaster

T0 review · 3 major / 6 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read The paper claims that replacing RoadTracer's convolutional decision network with a Teacher-Student Adaptive Deep Belief Network and adding a taboo list to the graph search improves average road-detection accuracy on seven difficult cities f

desk verdict Credible niche application of the authors' own Adaptive DBN/teacher-student ensemble to RoadTracer, but the headline improvement is computed on selected hard cities with an under-specified protocol. read the letter →

arxiv 2511.05567 v2 pith:VWESSVPU submitted 2025-11-04 cs.CV cs.AIcs.LG

Shin Kamada , Takumi Ichimura This is my paper · ORCID

classification cs.CVcs.AIcs.LG

keywords roadnetworkextractionTracerDeepBeliefAdaptivestructurallearningTeacher-Studentensembletaboosearchlandslidedisasterlightweight

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

This paper argues that the road-extraction system RoadTracer can be made substantially more accurate by swapping its convolutional neural network for an Adaptive Deep Belief Network trained with a Teacher-Student ensemble, and by adding a taboo list to its graph search to prevent premature stopping. On seven cities where the earlier model failed, the reported detection accuracy rises from 40.0% to 89.0% on average, and the method reaches 86.29% on the RoadTracer benchmark, above recent CNN-based competitors. The authors further show that a pruned version of the model runs in near real time on a CPU-only embedded device and can flag roads that became unavailable after a landslide disaster in Japan. If these results hold under proper held-out evaluation, the approach offers a lightweight, graph-based alternative for rapid road mapping and disaster response.

What carries the argument

Three components carry the argument. (1) Adaptive DBN: a Deep Belief Network that grows/removes hidden neurons via Walking Distance and energy thresholds, and adds layers automatically, so the network fits the data without manual architecture search. (2) Teacher-Student ensemble: one parent DBN is trained on all data; several child DBNs are trained on the parent's confusion cases; neurons from the best child are copied into the parent when KL divergence exceeds a threshold, transferring knowledge about ambiguous features. (3) Taboo search: if the graph search finds no new vertex for n consecutive steps, the position is added to a forbidden list so the walk continues elsewhere rather than ter

What would settle it

Train the teacher and student models using data only from cities other than the seven test cities, then evaluate on the seven; if the average detection accuracy drops back below about 50%, the headline improvement was an artifact of in-sample selection rather than a general capability. Alternatively, run a controlled comparison where the same held-out cities are evaluated under the original RoadTracer, Adaptive DBN alone, and Adaptive DBN+TB+TS, with model selection performed only on training cities.

Watch

Extended reading notes

Core claim

The central claim is that RoadTracer's decision function, which decides whether to extend a road graph by walking to a new vertex, can be powered by a Teacher-Student Adaptive DBN instead of a 17-layer CNN, and that doing so fixes the failure cases where the CNN-based model stops early. The Adaptive DBN self-organizes its hidden neurons and layers during training; the student models are trained on mis-classified confusion cases and their neurons are copied into the teacher via KL-divergence-based knowledge distillation. The taboo search keeps the walk from getting stuck in local loops. On the seven hard cities, the combination raises detection accuracy from roughly 40% to 89% average, and on

Load-bearing premise

The reported 40% to 89% improvement assumes the seven evaluation cities were held out from training the teacher and student DBNs; the paper never states the train/test split, and the student models were built from confusion cases whose source split is not described.

Editorial extensions

If this is right

If the accuracy gain holds, graph-based road extraction no longer needs a heavy CNN; a self-structuring DBN with an ensemble can outperform recent CNN-based road-topology methods.
The taboo search means the algorithm trades longer search time for broader coverage, which matters for disaster scenarios where the road graph is partially destroyed and local loops are common.
The pruned model runs at about 4.6 FPS on CPU (0.86 s per 1024x1024 image) versus 0.63 FPS for the original RoadTracer, making real-time aerial-photo road detection feasible on embedded hardware.
The before/after comparison on the 2018 Hiroshima rainfall disaster shows that subtracting detected road vertices before and after the event can automatically flag disconnected road segments, matching the roads residents reported as unavailable.

Reading between the lines

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

The seven evaluation cities were chosen because the prior Adaptive DBN performed worst there; if the student models were trained on confusion cases drawn from these same cities, the 40% to 89% number would be an in-sample improvement, not a prediction on unseen cities. The paper does not state the train/test split.
The reported 86.29% comparison to NodeConnect may not be apples-to-apples, since the models were likely trained on different splits of the RoadTracer dataset; a shared benchmark evaluation with identical train/test masks would settle this.
The taboo search's large increase in search time (e.g., London from 8310 to 37498 steps) suggests the accuracy gain comes partly from exploring more area, not just from better recognition; ablations separating the two effects would clarify the contribution.
The disaster experiment uses only six small areas and matches known manual reports; a natural next step is to test on a larger, independently annotated disaster dataset to measure false positives, since the method flags 'available roads' by differencing two noisy detections.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Quick take: this is a legitimate incremental application of the authors' prior Adaptive DBN and teacher-student ensemble to RoadTracer's decision function, plus a small taboo-search tweak to the graph walk. The disaster-response demo is a nice real-world test. But the headline '40% to 89%' number is computed on seven cities deliberately chosen because the baseline did worst there, and the paper never spells out the train/test split for the road data. So treat that number as a hard-case demonstration, not a general accuracy claim.

What's new: using the TS Adaptive DBN as the decision function in RoadTracer, and adding a taboo list to avoid early termination. The CIFAR results are a re-run of their earlier method; the road extraction experiments are the new part. The internal comparison in Table III shows a clear monotonic improvement from the baseline to +TB to +TB+TS across all seven cities, with 10-trial averages and standard deviations—that's evidence the components do something. The Jetson speed test is also useful for the claimed deployment scenario.

Where it gets soft: first, the seven cities were selected from a prior failure analysis of the Adaptive DBN baseline. That's fine if you want to show the fixes address known weaknesses, but then the 89% average is not representative of general performance. Second, the paper never states whether the teacher and student DBNs were trained on patches from these seven cities or from a disjoint set, and the student models are built on the teacher's confusion cases—if those confusion cases came from the same cities being tested, part of the gain could be memorization. Table IV complicates this further: the reported 86.29% on the RoadTracer dataset doesn't match the 89.0% average from Table III, and the evaluation set for the SOTA comparison isn't described. Third, there's no code or data, so none of this can be checked externally. The disaster application is qualitative—vertex counts before/after and a hand-check against local knowledge—so it's a demo, not a validated product.

Who this is for: people working on road extraction from aerial imagery, especially those interested in non-CNN backbones or edge deployment. The paper deserves a serious referee because the method is plausible and the problem is relevant, but a referee should push for a clear split, a proper random-city evaluation, and release of at least the trained models or the evaluation script.

Referee Report

3 major / 6 minor

Summary. The paper proposes a road-network extraction system built on RoadTracer, replacing the original CNN decision function with an Adaptive DBN whose structure is learned by neuron generation/annihilation and layer generation, and augmenting it with a Teacher-Student (TS) ensemble to handle mis-classified road features. A taboo-list mechanism is also added to the RoadTracer graph search to prevent early termination of the search. The authors report large gains on seven cities (40.0% to 89.0% average detection accuracy), a comparison with recent road-extraction methods (86.29% vs. NodeConnect's 77.96%), and an application to detecting available roads before/after a 2018 landslide in Hiroshima, with deployment on Jetson Xavier NX.

Significance. If the evaluation were performed on genuinely held-out cities, the result would be significant: it would demonstrate that a learnable RBM/DBN decision function plus explicit search modification can outperform CNN-based road-graph extraction, and the embedded deployment would be practically useful. The manuscript has real strengths: the algorithms are described in detail, Table III reports 10-trial statistics, and the internal comparison shows monotonic gains from taboo search and TS. However, the evidence is currently insufficient: the seven-city evaluation set appears to be selected from the authors' prior failure analysis, no train/test split is specified, the state-of-the-art comparison is under-specified and numerically inconsistent, and the disaster application lacks quantitative ground-truth validation. These issues must be resolved before the headline claims can be accepted.

major comments (3)

[§IV-D, Table III] The reported 40.0%→89.0% improvement is not established as a held-out result. The seven cities were selected because the authors' previous Adaptive DBN performed worst on them ('the detection accuracy ... was worse for 7 of 40 cities'), and the TS student models are trained on mis-classification ('confusion cases'). The manuscript never states (a) whether these seven cities/patches were excluded from teacher pre-training, structure selection, and student-model training; (b) how the road-data confusion subsets were constructed; or (c) whether thresholds such as θKL, T, and the number of student models were tuned on the same cities. Without this information, the improvement may reflect test-set adaptation rather than generalization. Please report the exact split and provide results on a randomly selected held-out set or on all 40 cities.
[Table IV] The comparison with recent methods is not reproducible. The text calls it 'the RoadTracer dataset' but does not specify the city set, the ground-truth vertex-matching criterion, the walk threshold, or the hyperparameter settings for RoadTracer, D-LinkNet, CoANet, and NodeConnect. It is also unexplained why 'Our model' is 86.29% in Table IV while the seven-city average in Table III is 89.0%. The claim of exceeding NodeConnect (77.96%) requires an identical evaluation protocol and a clear statement of which images were used for training and testing.
[§V-B, Tables V–VII] The landslide application is not quantitatively validated. The paper counts detected vertices before and after the disaster and infers unavailable connections, but it gives no ground-truth comparison—e.g., official road-closure records, manual labeling, or precision/recall. The statement in Section VI that the detected roads 'matched' resident-shared information is anecdotal. Moreover, before/after satellite images can differ in acquisition conditions and registration; these should be controlled or acknowledged. Please add quantitative evaluation or explicitly frame the section as a qualitative proof of concept.

minor comments (6)

[Abstract] The final sentence, 'This version of the article was improved the search algorithm at the border around image,' is an ungrammatical version-note and should be removed from an abstract.
[Abstract and Section VII] The wording 'improved from 40.0% to 89.0%' collapses two distinct changes: Table III shows taboo search alone raises the seven-city average from 40.0% to 75.0%, and TS raises it further to 89.0%. Please attribute the gain to both components and state that the 40.0% baseline is the previous Adaptive DBN.
[Table III] The column label 'Searching time' is misleading because the text defines it as the number of steps/epochs until the search finishes, not wall-clock time. Rename the column accordingly.
[§III-B, Tables I–II] The CIFAR-10/100 results (99.7% and 95.5%) are exceptionally high and need a statement of data preprocessing, augmentation, and whether the same protocol and splits were used for all baselines. The sentence '10-fold cross validation test was conducted for only the Adaptive DBN' also leaves unclear whether the TS results were cross-validated.
[§V-A] In the description of the six disaster areas, 'about 750 square mm' appears to be the wrong unit; please correct to an appropriate area unit such as km² or m².
[General] There are several typos and inconsistencies: 'V ancouver' in Table III, 'aliment' in §IV-A, 'coping' for 'copying' in §III-A. A data/code availability statement would also improve reproducibility.

Circularity Check

1 steps flagged · score 6.0 of 10

Seven-city road-detection improvement is an in-sample fit: the TS student models are trained on the same failure cases/city features that are then reported as test accuracy.

fitted input called prediction [Section IV-D, Table III (and Abstract); cf. Algorithm 1 and Section IV-C]
"In the previous research, the Adaptive DBN was evaluated on 40 cities ... However, the detection accuracy of the Adaptive DBN was worse for 7 of 40 cities, which are London, Louisville, New York, Tokyo, Amsterdam, Vancouver, and Montreal ... Therefore, the seven cities are evaluated by the proposed taboo search and TS model in this paper. ... the TS model was able to detect many roads over the wide roads and bridges in Louisville because such features were trained in the student models."

The seven cities are the same failure set that motivated the TS students: Algorithm 1 constructs students from 'mis-classified cases' of the teacher, and the text explicitly explains Louisville's bridge detection by saying those features 'were trained in the student models.' These same seven cities are then used to compute the headline improvement from 40.0% to 89.0% (Table III) and the SOTA comparison value 86.29% (Table IV). The paper does not state any train/test split or that the confusion cases came from outside the seven evaluation cities. As written, the student DBN was fit to the very failure modes and city-specific features on which detection accuracy is reported, so the claimed prediction is an in-sample fit rather than an independent test.

full rationale

The localized circularity is in the road-network evaluation: the reported 40.0%→89.0% improvement is not an independent prediction because the TS students are trained on teacher mis-classifications, and the manuscript says the Louisville bridges were detected because such features 'were trained in the student models.' The same seven cities are the source of the failure analysis and the accuracy table. This reduces the central road-detection claim by construction for those cities. The CIFAR-10/100 experiments use 10-fold cross-validation and provide independent evidence for the TS mechanism, and the landslide-disaster application is a qualitative real-image demonstration, so not every claim in the paper is circular. The Adaptive DBN neuron/layer generation, KL-divergence distillation, and taboo search are algorithmically defined and not tautological; no uniqueness theorem is imported. However, because the headline road-network result rests on fitting to the evaluation cities, the paper earns a 6 rather than a lower score. The lack of code/data and unspecified split are protocol concerns, but the circularity itself is grounded in the explicit statement that test-city features were trained in the student models.

Assumptions & free parameters 7 free parameters · 6 assumptions · 0 invented entities

The central claim rests on the authors' prior self-cited Adaptive DBN/TS model, on an assumed (but unstated) train/test split, and on the validity of vertex-recall as a cross-method accuracy metric. The model's thresholds are inherited from CIFAR experiments or from the authors' earlier papers, and the taboo search parameter is not reported.

free parameters (7)

theta_G (neuron generation threshold) = 0.001
Neuron generation threshold for Adaptive RBM, inherited from prior CIFAR experiments [13]; not tuned on road data.
theta_A (neuron annihilation threshold) = 0.100
Neuron annihilation threshold for Adaptive RBM, set as in [13].
theta_L1, theta_L2 (layer generation thresholds) = 0.05
Thresholds for Adaptive DBN layer generation, set as in [13].
theta_KL (KL divergence threshold for neuron transfer) = 0.0015
Threshold determining which student neurons are copied to the teacher; taken from preliminary experiments in self-cited [22].
Number of student models = 10
Set to 10 for CIFAR-100 and RoadTracer experiments based on the number of confusing classes; no sensitivity analysis.
Taboo search parameter n = not reported
Number of consecutive failed searches before a vertex is added to the taboo list; the rule is described but n is never given, so the search behavior is under-specified.
RoadTracer walk threshold T and angle resolution a = a=64, T=0.5
From original RoadTracer [12]; step distance D is not given.

assumptions (6)

domain assumption Adaptive DBN's neuron generation/annihilation and layer generation algorithms behave as described in [13]
The paper relies on the prior self-cited model's correctness and performance; no code or independent replication is provided.
domain assumption Train/test separation: the seven evaluation cities and the confusion cases used to train students are not part of the model's training set
Never explicitly stated; if violated, the reported accuracy gains would be inflated. This is load-bearing.
domain assumption Vertex detection ratio is a valid cross-method metric for comparing road extraction quality
Tables III-IV compare against segmentation-based methods (D-LinkNet, CoANet) without precision/recall or topology metrics like APLS.
domain assumption Google Map satellite images and OpenStreetMap graphs for the seven cities are accurate ground truth
No alignment or quality checks are described.
domain assumption Before/after disaster images from GSI are comparable (registration, illumination, resolution) so vertex-count differences indicate road damage
Used to infer available roads in Section V; no normalization or co-registration is described.
domain assumption The taboo search heuristic preserves the connectivity of the true road graph
Limiting search to areas not in the taboo list could in principle skip valid roads; the paper assumes it only avoids loops. Parameter n is unreported.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Automatic Extraction of Road Networks by using Teacher-Student Adaptive Structural Deep Belief Network and Its Application to Landslide Disaster." pith.science (2026). https://pith.science/paper/VWESSVPU

@misc{pith2026251105567,
  author       = {Pith},
  title        = {Pith review of: Automatic Extraction of Road Networks by using Teacher-Student Adaptive Structural Deep Belief Network and Its Application to Landslide Disaster},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VWESSVPU}},
  note         = {Machine review of arXiv:2511.05567}
}

read the original abstract

An adaptive structural learning method of Restricted Boltzmann Machine (RBM) and Deep Belief Network (DBN) has been developed as one of prominent deep learning models. The neuron generation-annihilation algorithm in RBM and layer generation algorithm in DBN make an optimal network structure for given input during the learning. In this paper, our model is applied to an automatic recognition method of road network system, called RoadTracer. RoadTracer can generate a road map on the ground surface from aerial photograph data. A novel method of RoadTracer using the Teacher-Student based ensemble learning model of Adaptive DBN is proposed, since the road maps contain many complicated features so that a model with high representation power to detect should be required. The experimental results showed the detection accuracy of the proposed model was improved from 40.0\% to 89.0\% on average in the seven major cities among the test dataset. In addition, we challenged to apply our method to the detection of available roads when landslide by natural disaster is occurred, in order to rapidly obtain a way of transportation. For fast inference, a small size of the trained model was implemented on a small embedded edge device as lightweight deep learning. We reported the detection results for the satellite image before and after the rainfall disaster in Japan. This version of the article was improved the search algorithm at the border around image.

Figures

Figures reproduced from arXiv: 2511.05567 by the authors.

**Figure 1.** Network structure of RBM II. ADAPTIVE STRUCTURAL LEARNING METHOD OF DBN The Adaptive DBN has been developed as one of the original deep learning model which extents the traditional RBM and DBN. The overview of the method is described as the following sections. A. RBM and DBN An RBM [8] is a generative model with the visible neurons v ∈ {0, 1} I for an input and the hidden neurons h ∈ {0, 1} J to represent feature ve… view at source ↗

**Figure 2.** Convergence situation of Walking Distance (WD) [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Adaptive RBM and Adaptive DBN [13] have been developed. The proposed learning method enables the model to generate a new hidden neuron and a new hidden layer according to the training situation. The key idea of the neuron generation in Adaptive RBM is to monitor how the learning parameters is changed in the training epochs, where the amount of change is called Walking Distance (WD) as shown in [PITH_FULL_IMAGE:figu… view at source ↗

Figures from the paper (10 more)

**Figure 4.** Figure 4: An overview of Adaptive DBN. The suitable number of hi [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

**Figure 5.** Figure 5: The ensemble learning model of Adaptive DBN. After tr [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]

**Figure 6.** Figure 6: Activated neurons and paths Moreover, 10-fold cross validation test was conducted for only the Adaptive DBN. Although the original Adaptive DBN [13] had higher classification accuracy than the CNNs, some confusion cases for five classes ‘bird’, ‘cat’, ‘dog’, ‘automobi…

**Figure 7.** Figure 7: Overview of decision function Algorithm 2 Graph Search [12] Require: A satellite Image and a constant value D are given. 1: initialize a vertex v0 as starting location, an empty graph G, and a stack of vertexes S with v0. Stop is defined by the top position of S. 2: wh…

**Figure 8.** Figure 8: Example of Search Algorithm TABLE III DETECTION ACCURACY FOR SEVEN CITIES Searching time Detection Accuracy City Model Ave. Std. Max Min Ave. Std. Max Min London Adaptive DBN 8310.10 1193.68 9384 5168 16.7% 0.0264 18.8% 9.5% Adaptive DBN + TB 31663.40 107.05 31830 3153…

**Figure 9.** Figure 9: Detection results contain complicated features and then the searching process was finished early. By using the TB search, the searching time was increased with smaller variance and the explorable area was expanded because the search was progressed for the direction exc…

**Figure 10.** Figure 10: Rough map around areas A-F [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]

**Figure 11.** Figure 11: An example of system implementation obtain a way of transportation and rescue residents even when a natural disaster is occurred. In the experiments, the satellite images in Hiroshima, Japan in July 2018, were used, where a heavy rainfall disaster was actually occurre…

**Figure 12.** Figure 12: Detection result before and after the disaster [PITH_FULL_IMAGE:figures/full_fig_p012_12.png]

**Figure 13.** Figure 13: Available roads [18], the fine-tuning method was developed, which can remove unnecessary neurons from the trained DBN model as pruning. After the training, frequency of activating neurons at each layer for given input is calculated. If unnecessary or inactivated neur…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 2 linked inside Pith

[1]

Japan Meteorological Agency, Climate Change Monitoring Report 2018 , https://www.jma.go.jp/jma/en/NMHS/ccmr/ccmr2018.pdf (July 6, 2022 Access)

2018
[2]

Y .Bengio, Learning Deep Architectures for AI , Foundations and Trends in Machine Learning archive, V ol.2, No.1, pp.1-127 (2009)

2009
[3]

of Advances in Neural Information Processing Systems 25 (NIPS 2012) (2012)

A.Krizhevsky, I.Sutskever, G.E.Hinton, ImageNet Classiﬁcation with Deep Convolutional Neural Networks , Proc. of Advances in Neural Information Processing Systems 25 (NIPS 2012) (2012)

2012
[4]

of International Conference on Learning Representations (ICLR 2015) (2015)

K.Simonyan, A.Zisserman, V ery deep convolutional networks for large- scale image recognition , Proc. of International Conference on Learning Representations (ICLR 2015) (2015)

2015
[5]

C.Szegedy, W.Liu, et.al., Going Deeper with Convolutions , 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVP R), pp.1– 9 (2015)

2015
[6]

of 2016 IEEE Conference on Computer Vision and Patter n Recognition (CVPR), pp.770–778 (2016)

K.He, X.Zhang, S.R en, J.Sun, Deep residual learning for image recog- nition, Proc. of 2016 IEEE Conference on Computer Vision and Patter n Recognition (CVPR), pp.770–778 (2016)

2016
[7]

G.E.Hinton, S.Osindero and Y .Teh, A fast learning algorithm for deep belief nets , Neural Computation, V ol.18, No.7, pp.1527–1554 (2006)

2006
[8]

G.E.Hinton, A Practical Guide to Training Restricted Boltzmann Ma- chines, Neural Networks, Tricks of the Trade, Lecture Notes in Comp uter Science (LNCS, V ol.7700), pp.599–619 (2012)

2012

Show all 49 references

[9]

C.H.Wang, K.Y .Huang, et.al., Lightweight Deep Learning: An Overview , in IEEE Consumer Electronics Magazine, pp.1–12 (2022)

2022
[10]

of the IEEE Conference on Computer Vision and Pattern Recognition (ICCV), pp.3438–3446 (2017)

G.M´ attyus, et.al., DeepRoadMapper: Extracting road topology from aerial images , Proc. of the IEEE Conference on Computer Vision and Pattern Recognition (ICCV), pp.3438–3446 (2017)

2017
[11]

R.Lian and L.Huang, DeepWindow: Sliding Window Based on Deep Learning for Road Extraction From Remote Sensing Images, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, V ol.13, pp.1905–1916 (2020)

1905
[12]

F.Bastani, et.al., RoadTracer: Automatic Extraction of Road Networks from Aerial Images , arXiv:1802.03680 [cs.CV] (2018)

2018 arXiv
[13]

S.Kamada, T.Ichimura, A.Hara, and K.J.Mackin, Adaptive Structure Learning Method of Deep Belief Network using Neuron Generat ion- Annihilation and Layer Generation , Neural Computing and Applications, pp.1–15 (2018)

2018
[14]

of 2016 IEEE International Conference on Systems, Man , and Cybernetics (IEEE SMC 2016), pp.1273–1278 (2016)

S.Kamada and T.Ichimura, An Adaptive Learning Method of Restricted Boltzmann Machine by Neuron Generation and Annihilation Al gorithm, Proc. of 2016 IEEE International Conference on Systems, Man , and Cybernetics (IEEE SMC 2016), pp.1273–1278 (2016)

2016
[15]

S.Kamada and T.Ichimura, A Structural Learning Method of Restricted Boltzmann Machine by Neuron Generation and Annihilation Al gorithm, Neural Information Processing, V ol.9950 of the series Lect ure Notes in Computer Science, pp.372–380 (2016)

2016
[16]

of IEEE TENCON2016, pp.2971–2974 (2016)

S.Kamada and T.Ichimura, An Adaptive Learning Method of Deep Belief Network by Layer Generation Algorithm , Proc. of IEEE TENCON2016, pp.2971–2974 (2016)

2016
[17]

of IEEE TENCON 2020, pp.1203–1208 (2020)

S.Kamada, T.Ichimura, An Embedded System for Image-based Crack Detection by using Fine-Tuning model of Adaptive Structura l Learning of Deep Belief Network , Proc. of IEEE TENCON 2020, pp.1203–1208 (2020)

2020
[18]

S.Kamada, T.Ichimura, Adaptive Structural Learning of Deep Belief Network and its Application to Real Time Crack Detection of C oncrete Structure using Drone , Handbook on Artiﬁcial Intelligence-Empowered Applied Software Engineering, Artiﬁcial Intelligence-En hanced Software ...

2022
[19]

A.Krizhevsky, Learning Multiple Layers of Features from Tiny Images , Master of thesis, University of Toronto (2009)

2009
[20]

Neural Computation, vol.14, pp.1771–1800 (2002)

G.E.Hinton, Training products of experts by minimizing contrastive divergence. Neural Computation, vol.14, pp.1771–1800 (2002)

2002
[21]

16, 2023 15

A.Mollahosseini, B.Hasani, M.H.Mahoor: AffectNet: A Database for Facial Expression, V alence, and Arousal Computing in the Wi ld, IEEE Transactions on Affective Computing, V ol.10, No.1 pp.18–3 1 (2017) IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERV A TIO NS AND REMOT...

2017
[22]

T.Ichimura, S.Kamada, An Ensemble Learning Method of Adaptive Structural Deep Belief Network for AffectNet , IIAI International Journal Series, International Journal of Smart Computing and Artiﬁ cial Intelli- gence, V ol.6, No.1, pp.1–17(2022)

2022
[23]

I.Goodfellow, D.Warde-Farley, M.Mirza, A.Courville , and Y .Bengio, Maxout Networks. Proc. of Machine Learning Research (PMLR), V ol.28, No.3, pp.1319–1327 (2013)

2013
[24]

D.A.Clevert, T.Unterthiner, and S.Hochreiter, Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs) . Proc. of ICRL 2016 (2016)

2016
[25]

arXiv:1412.6071 (2015)

G.Benjamin, Fractional Max-Pooling. arXiv:1412.6071 (2015)

2015 arXiv
[26]

of the British Machine Vision Conference (BMVC), pp.87.1–87.12 ( 2016)

S.Zagoruyko and N.Komodakis, Wide Residual Networks , Proc. of the British Machine Vision Conference (BMVC), pp.87.1–87.12 ( 2016)

2016
[27]

X.X.Zhu, et.al., Deep Learning in Remote Sensing: A Comprehensive Review and List of Resources , IEEE Geoscience and Remote Sensing Magazine, V ol.5, No.4, pp.8–36 (2017)

2017
[28]

M.Lei, et.al., Deep learning in remote sensing applications: A meta- analysis and review , ISPRS Journal of Photogrammetry and Remote Sensing, V ol.152, pp.166–177 (2019)

2019
[29]

J.D.Wegner, et.tl., Cataloging Public Objects Using Aerial and Street- Level Images — Urban Trees, 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp.6014–6023 (2016)

2016
[30]

X.Wu, et.al., UIU-Net: U-Net in U-Net for Infrared Small Object Detection, in IEEE Transactions on Image Processing, vol.32, pp.364– 376 (2023)

2023
[31]

2013 IEEE GRSS Data Fusion Contest – Fusion of Hyperspec tral and LiDAR Data, https://hyperspectral.ee.uh.edu/?page id=459, [online, 2023/5/5]

2013
[32]

D.Hong, et.al., Graph Convolutional Networks for Hyperspectral Image Classiﬁcation, in IEEE Transactions on Geoscience and Remote Sensing, vol.59, no.7, pp.5966–5978 (2021)

2021
[33]

Hong et.al., More Diverse Means Better: Multimodal Deep Learning Meets Remote-Sensing Imagery Classiﬁcation , in IEEE Transactions on Geoscience and Remote Sensing, vol

D. Hong et.al., More Diverse Means Better: Multimodal Deep Learning Meets Remote-Sensing Imagery Classiﬁcation , in IEEE Transactions on Geoscience and Remote Sensing, vol. 59, no. 5, pp.4340–4354 (2021)

2021
[34]

M.Zhang, et.al., Hyperspectral and LiDAR Data Classiﬁcation Based on Structural Optimization Transmission , in IEEE Transactions on Cy- bernetics, vol.53, no.5, pp.3153–3164 (2023)

2023
[35]

34, no.6, pp.2817–2830 (2023)

Y .Zhang, et.al., Topological Structure and Semantic Information Transfer Network for Cross-Scene Hyperspectral Image Classiﬁcatio n, in IEEE Transactions on Neural Networks and Learning Systems, vol. 34, no.6, pp.2817–2830 (2023)

2023
[36]

Y .Zhang, et.al., Graph Information Aggregation Cross-Domain Few-Shot Learning for Hyperspectral Image Classiﬁcation , in IEEE Transactions on Neural Networks and Learning Systems (2022)

2022
[37]

Y .Zhang, et.al., Single-Source Domain Expansion Network for Cross- Scene Hyperspectral Image Classiﬁcation, in IEEE Transactions on Image Processing, vol.32, pp.1498-1512 (2023)

2023
[38]

Y .Zhang, et.al., Language-Aware Domain Generalization Network for Cross-Scene Hyperspectral Image Classiﬁcation , in IEEE Transactions on Geoscience and Remote Sensing, vol.61, pp.1–12 (2023)

2023
[39]

of the AAAI Conference on Artiﬁcial Intelligence, V ol.34, No.07, pp.10965–10972 (2020)

S.He, et.al., RoadTagger: Robust Road Attribute Inference with Graph Neural Networks, Proc. of the AAAI Conference on Artiﬁcial Intelligence, V ol.34, No.07, pp.10965–10972 (2020)

2020
[40]

of 2020 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pp.8907–8915 (2020)

Y .Q.Tan, et.al., V ecRoad: Point-Based Iterative Graph Exploration for Road Graphs Extraction , Proc. of 2020 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pp.8907–8915 (2020)

2020
[41]

Y .Liu, et.al., RoadNet: Learning to Comprehensively Analyze Road Networks in Complex Urban Scenes From High-Resolution Remo tely Sensed Images , IEEE Transactions on Geoscience and Remote Sensing, V ol.57, No.4, pp.2043–2056 (2018)
[42]

Bahri and F

G.Bahl, M. Bahri and F. Lafarge, Single-Shot End-to-end Road Graph Extraction, 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pp. 1402–1411, (2022)

2022
[43]

L.Zhou, C.Zhang and M.Wu, D-LinkNet: LinkNet with Pretrained En- coder and Dilated Convolution for High Resolution Satellit e Imagery Road Extraction , 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pp.192–1924 (2018 )

2018
[44]

J.Mei, et.al., CoANet: Connectivity Attention Network for Road Extrac- tion From Satellite Imagery , in IEEE Transactions on Image Processing, vol.30, pp.8540–8552 (2021)

2021
[45]

J.Zhang, X. Hu, Y . Wei and L. Zhang, Road Topology Extraction From Satellite Imagery by Joint Learning of Nodes and Their Conne ctivity, in IEEE Transactions on Geoscience and Remote Sensing, vol.61 , pp.1–13 (2023)

2023
[46]

of 10th International Conference on Informatics, Electron ics & Vision (ICIEV 2021), No.37 (2021)

S.Kamada, T.Ichimura, Automatic Extraction of Road Networks from Satellite Images by using Adaptive Structural Deep Belief N etwork, Proc. of 10th International Conference on Informatics, Electron ics & Vision (ICIEV 2021), No.37 (2021)

2021
[47]

S.Coast, OpenStreetMap, https://www.openstreetmap .org/ [online, 2021/3/15]

2021
[48]

Geospatial Information Authority of Japan, https://w ww.gsi.go.jp/ [on- line, 2022/5/20]

2022
[49]

PASCO, https://www.pasco.co.jp/products/aerial survey/ [online, 2023/3/27] Shin Kamada is an Associate Professor at the Graduate School of Information Sciences, Hiroshima City University. He received his ME and DE in Management and Information System from Prefec- tural Univers...

2023

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.

[1] [1]

Japan Meteorological Agency, Climate Change Monitoring Report 2018 , https://www.jma.go.jp/jma/en/NMHS/ccmr/ccmr2018.pdf (July 6, 2022 Access)

2018

[2] [2]

Y .Bengio, Learning Deep Architectures for AI , Foundations and Trends in Machine Learning archive, V ol.2, No.1, pp.1-127 (2009)

2009

[3] [3]

of Advances in Neural Information Processing Systems 25 (NIPS 2012) (2012)

A.Krizhevsky, I.Sutskever, G.E.Hinton, ImageNet Classiﬁcation with Deep Convolutional Neural Networks , Proc. of Advances in Neural Information Processing Systems 25 (NIPS 2012) (2012)

2012

[4] [4]

of International Conference on Learning Representations (ICLR 2015) (2015)

K.Simonyan, A.Zisserman, V ery deep convolutional networks for large- scale image recognition , Proc. of International Conference on Learning Representations (ICLR 2015) (2015)

2015

[5] [5]

C.Szegedy, W.Liu, et.al., Going Deeper with Convolutions , 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVP R), pp.1– 9 (2015)

2015

[6] [6]

of 2016 IEEE Conference on Computer Vision and Patter n Recognition (CVPR), pp.770–778 (2016)

K.He, X.Zhang, S.R en, J.Sun, Deep residual learning for image recog- nition, Proc. of 2016 IEEE Conference on Computer Vision and Patter n Recognition (CVPR), pp.770–778 (2016)

2016

[7] [7]

G.E.Hinton, S.Osindero and Y .Teh, A fast learning algorithm for deep belief nets , Neural Computation, V ol.18, No.7, pp.1527–1554 (2006)

2006

[8] [8]

G.E.Hinton, A Practical Guide to Training Restricted Boltzmann Ma- chines, Neural Networks, Tricks of the Trade, Lecture Notes in Comp uter Science (LNCS, V ol.7700), pp.599–619 (2012)

2012