| CNN-BASED FEATURE LEVEL FUSION OF HGH RESOLUTION AERIAL IMAGE AND LiDAR DATA |
| Paper ID : 1361-SMPR |
| Authors: |
|
Somaye Daneshtalab1, Heidar Rastiveis *2, Benyamin Hosseiny3 1Department of Photogrammetry and Remote Sensing, School of Surveying and Geo-Spatial Engineering College of Engineering, University of Tehran, 2Department of Photogrammetry and Remote Sensing,
School of Surveying and Geo-Spatial Engineering
College of Engineering, University of Tehran, 3Department of Geospatial engineering, Faculty of Engineering, University of Tehran, Tehran, Iran |
| Abstract: |
| Land-cover classification of Remote Sensing (RS) data in urban area has always been a challenging task due to the complicated relations between different objects. Recently, the fusion of aerial imagery and light detection and ranging (LiDAR) data has obtained a great attention in RS communities. Traditional methods can hardly use the deep features of the data, while CNN solves this difficulty and extracts its well. In this paper, we propose a new feature fusion framework using convolutional neural network (CNN) for object classification in urban area. The first step involves pre-processing and normalizing the input data. after pre-processing the proposed framework employs a novel convolutional neural network to extract the spectral-spatial features of aerial imagery and LiDAR data, and then a fully connected multilayer perceptron network (MLP) to classify the area based on these features. The experimental results revealed that the proposed deep fusion model provides more than 10% improvement of overall accuracy in comparison with other conventional feature-level fusion techniques. |
| Keywords: |
| Convolutional neural network (CNN), feature fusion, deep learning, feature extraction, aerial imagery, LiDAR |
| Status : Conditional Accept (Oral Presentation) |
