|
MMP-MLDANet: Classification of LiDAR-DSM Image Based on Multishape Morphological Profiles and Multiscale LDA-based Deep Random Patches Networks
|
Behnam Asghari Beirami *   , Mohammadreza Seif |
| Imam Hossein university |
|
|
Abstract: (28 Views) |
Similar elevation information in different land classes makes it challenging to classify features and land classes in Digital Surface Model (DSM) images obtained from LiDAR sensors. In addition to elevation data, DSM images contain valuable spatial information that can greatly improve classification accuracy. In recent years, deep learning techniques have emerged as powerful tools for feature extraction and image classification. Despite achieving appropriate results, many proposed deep learning approaches have complex structures and require advanced hardware and large training datasets. As a solution, this paper proposes a new strategy for DSM classification to produce accurate land-cover maps. The proposed method, named MMP-MLDANet, uses multishape morphological profiles (MMP) and multiscale LDA-based deep random patches networks (MLDANet) for feature extraction. Finally, a multiple-classifier system based on support vector machines (SVMs) is developed to classify the resulting deep features. Based on experiments conducted on single-band DSM images of the Houston and Trento areas, the proposed method achieves the desired overall accuracies of 90.85% and 98.49% with a limited number of training samples, outperforming some existing methods in this field.
|
|
| Keywords: Digital Surface Model, Lidar, Deep Learning, Classification, Morphological Profile |
|
|
|
Type of Study: Research |
Subject:
RS
Received: 2025/01/7 | Accepted: 2026/06/13 | ePublished ahead of print: 2026/06/17
|
|
|
|
|
|
|
| Send email to the article author |
|
|
