Efficient Convolutional Neural Architecture Search for LiDAR DSM Classification

Wang, Aili; Deng, Xue; Wu, Haibin; Gu, Yanfeng

doi:10.1109/tgrs.2022.3171520

Cited by 5 publications

(3 citation statements)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this section, we evaluate the classification performance of MS 3 ANAS by comparing it with several advanced methods. These methods include Extended Morphological Profile combined with Support Vector Machine (EMP-SVM) [44], 2D-CNN [11], 3D-CNN [14], Spectral-Spatial Residual Network (SSRN) [16], Residual Network (ResNet) [45], Multi-Layer Perceptron Mixer (MLP Mixer) [46], CNN model designed using NAS (CNAS) [34], and Efficient Convolutional Neural Architecture Search for LiDAR DSM Classification (AN-AS-CPA-LS) [47]. To ensure the rigor of the experiment, we randomly selected three The WHU-Hi-Hanchuan dataset was collected in Hanchuan, Hubei Province, using the 17 mm focus Headwall Nano Hyperspec sensor installed on the Leica Airbot X6 UAV V1 platform.…”

Section: Classification Comparison With State-of-the-art Methodsmentioning

confidence: 99%

Multi-Scale Spatial–Spectral Attention-Based Neural Architecture Search for Hyperspectral Image Classification

Song,

Wang,

Zhao

et al. 2023

Electronics

View full text Add to dashboard Cite

Convolutional neural networks (CNNs) are indeed commonly employed for hyperspectral image classification. However, the architecture of cellular neural networks typically requires manual design and fine-tuning, which can be quite laborious. Fortunately, there have been recent advancements in the field of Neural Architecture Search (NAS) that enable the automatic design of networks. These NAS techniques have significantly improved the accuracy of HSI classification, pushing it to new levels. This article proposes a Multi-Scale Spatial–Spectral Attention-based NAS, MS3ANAS) framework for HSI classification to automatically design a neural network structure for HSI classifiers. First, this paper constructs a multi-scale attention mechanism extended search space, which considers multi-scale filters to reduce parameters while maintaining large-scale receptive field and enhanced multi-scale spectral–spatial feature extraction to increase network sensitivity towards hyperspectral information. Then, we combined the slow–fast learning architecture update paradigm to optimize and iteratively update the architecture vector and effectively improve the model’s generalization ability. Finally, we introduced the Lion optimizer to track only momentum and use symbol operations to calculate updates, thereby reducing memory overhead and effectively reducing training time. The proposed NAS method demonstrates impressive classification performance and effectively improves accuracy across three HSI datasets (University of Pavia, Xuzhou, and WHU-Hi-Hanchuan).

show abstract

Section: Classification Comparison With State-of-the-art Methodsmentioning

confidence: 99%

Multi-Scale Spatial–Spectral Attention-Based Neural Architecture Search for Hyperspectral Image Classification

Song,

Wang,

Zhao

et al. 2023

Electronics

View full text Add to dashboard Cite

show abstract

“…Typically, there are two main ways to obtain highresolution DSM datasets. The first way is to improve the resolution of surveying equipment (e.g., light detection and ranging laser scanner and stereo image pairs from aerial camera), which includes the utilization of more sophisticated sensors, instruments or techniques capable of capturing topographic data with higher resolution [2,5,29]. However, this way often requires a significant investment in labor, materials, and technical expertise.…”

Section: Introductionmentioning

confidence: 99%

Digital Surface Model Super-Resolution by Integrating High-Resolution Remote Sensing Imagery Using Generative Adversarial Networks

Sun,

Chen,

Huang

et al. 2024

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Digital Surface Model (DSM) is the fundamental data in various geoscience applications, such as city 3D modeling and urban environment analysis. The freely available DSM often suffers from limited spatial resolution. Super-resolution (SR) is a promising technique to increase the spatial resolution of DSM. However, most existing SR models struggle to reconstruct spatial details such as buildings, valleys, and ridges. This paper proposes a novel DSM super-resolution (DSMSR) model that integrates high-resolution remote sensing imagery using generative adversarial networks. The generator in DSMSR contains three modules. The first DSM feature extraction module uses the residual-in-residual dense block (RRDB) to extract features from low-resolution DSM. The second multiscale attention feature extraction module employs the pyramid convolutional residual dense (PCRD) blocks to capture spatial details of ground objects at multiple scales from remote sensing imagery. The third DSM reconstruction module uses a squeeze-and-excitation (SE) block to fuse the extracted features from low-resolution DSM and highresolution remote sensing imagery for generating SR DSM. The discriminator of DSMSR uses the relativistic average discriminator (RaD) for adversarial learning. The slope loss is further introduced to ensure the accurate representation of topographic features. We evaluate DSMSR on four different terrain regions in the UK to downscale the 30-m AW3D30 DSM to 5-m DSM. The experimental results indicate that DSMSR outperforms traditional interpolation algorithms and four existing deep learning-based SR models. The DSMSR restores more spatial detail of topographic features and generates more accurate image quality, elevation and terrain metrics.

show abstract

“…LiDAR sensors generate a digital surface model (DSM) as a secondary outcome, which is produced by implementing various techniques like denoising and rasterization on a point cloud. DSM contains elevational data of ground objects, which can be used to distinguish objects of different heights [5].…”

Section: Introductionmentioning

confidence: 99%

Land Covers Classification from LiDAR-DSM Data Based on Local Kernel Matrix Features of Morphological Profiles

Beirami

Mokhtarzade

2023

IJE

View full text Add to dashboard Cite

Accurate land cover classification from the digital surface model (DSM) obtained from LiDAR sensors is a challenging topic that researchers have considered in recent years. In general, the classification accuracy of land covers leads to low accuracy using a single-band DSM image. Hence, it seems necessary to develop efficient methods to extract relevant spatial information, which improves classification accuracy. In this regard, using spatial features based on morphological profiles (MPs) has significantly increased classification accuracy. Despite MPs' efficiency in increasing the DSM's classification accuracy, the classification accuracy results under the situation of limited training samples are not still at satisfactory levels. The main novelty of this paper is to propose a new feature space based on local kernel descriptors obtained from MP for addressing the mentioned challenge of MP-based DSM classification. These innovative feature vectors consider local nonlinear dependencies and higher-order statistics between the morphological features. The experiments of this study are conducted on two wellknown DSM datasets of Houston and Trento. Our results show that support vector machine (SVM)based DSM classification with the new local kernel features achieved an average accuracy of 93.75%, which is much better than conventional SVM classification with single-band DSM and MP features (by about 57% and 11.5% on average, respectively). Additionally, our proposed method outperformed two other DSM classification methods by an average of 4.7%.

show abstract

Efficient Convolutional Neural Architecture Search for LiDAR DSM Classification

Cited by 5 publications

References 40 publications

Multi-Scale Spatial–Spectral Attention-Based Neural Architecture Search for Hyperspectral Image Classification

Multi-Scale Spatial–Spectral Attention-Based Neural Architecture Search for Hyperspectral Image Classification

Digital Surface Model Super-Resolution by Integrating High-Resolution Remote Sensing Imagery Using Generative Adversarial Networks

Land Covers Classification from LiDAR-DSM Data Based on Local Kernel Matrix Features of Morphological Profiles

Contact Info

Product

Resources

About