Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Shin Kamada, Takumi Ichimura

TL;DR

This work tackles automatic road-network extraction from aerial imagery, with particular emphasis on rapid disaster response. It introduces a Teacher-Student ensemble built on an Adaptive Deep Belief Network to boost RoadTracer while enabling lightweight, CPU-compatible inference on embedded devices. The approach yields substantial gains, notably increasing road-detection accuracy from 40.0% to 89.0% on seven cities and achieving CIFAR-10/100 performance improvements; it also demonstrates real-time capable inference on Jetson hardware and applies the method to detect available roads after landslide events in Japan. Overall, the paper contributes a scalable, data-adaptive architecture and an effective search augmentation (taboo search) to deliver robust road-network extraction suitable for urgent disaster-response scenarios.

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.

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

TL;DR

This work tackles automatic road-network extraction from aerial imagery, with particular emphasis on rapid disaster response. It introduces a Teacher-Student ensemble built on an Adaptive Deep Belief Network to boost RoadTracer while enabling lightweight, CPU-compatible inference on embedded devices. The approach yields substantial gains, notably increasing road-detection accuracy from 40.0% to 89.0% on seven cities and achieving CIFAR-10/100 performance improvements; it also demonstrates real-time capable inference on Jetson hardware and applies the method to detect available roads after landslide events in Japan. Overall, the paper contributes a scalable, data-adaptive architecture and an effective search augmentation (taboo search) to deliver robust road-network extraction suitable for urgent disaster-response scenarios.

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.

Paper Structure

This paper contains 18 sections, 9 equations, 13 figures, 7 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Adaptive Structural Learning Method of DBN
  3. RBM and DBN
  4. Neuron Generation/Annihilation Algorithm of RBM
  5. Layer Generation Algorithm of DBN
  6. Ensemble Learning of Teacher-Student based Adaptive DBN
  7. Teacher-Student model
  8. Numeric Evaluation on CIFAR-10 / CIFAR-100
  9. Automatic Extraction of Road Networks using TS Adaptive Model
  10. Related works
  11. RoadTracer
  12. RoadTracer using Adaptive DBN
  13. Experiments results
  14. Detection of available roads in landslide disaster
  15. Experiment Environment
  16. ...and 3 more sections

Figures (13)

  • Figure 1: Network structure of RBM
  • Figure 2: Convergence situation of Walking Distance (WD)
  • Figure 3: Adaptive RBM
  • Figure 4: An overview of Adaptive DBN. The suitable number of hidden neurons is obtained for each pre-training of Adaptive RBM, and the suitable number of hidden layers is also generated by Adaptive DBN.
  • Figure 5: The ensemble learning model of Adaptive DBN. After training the original data by the teacher model, two or more student models are generated and trained for mis-classified samples. Finally, the trained student model's neurons are copied to the teacher model according to the KL divergence as knowledge distillation.
  • ...and 8 more figures