Share Email Print

Proceedings Paper

Improved neural network algorithm for classification of UAV imagery related to Wenchuan earthquake
Author(s): Na Lin; Wunian Yang; Bin Wang
Format Member Price Non-Member Price
PDF $14.40 $18.00
cover GOOD NEWS! Your organization subscribes to the SPIE Digital Library. You may be able to download this paper for free. Check Access

Paper Abstract

When Wenchuan earthquake struck, the terrain of the region changed violently. Unmanned aerial vehicles (UAV) remote sensing is effective in extracting first hand information. The high resolution images are of great importance in disaster management and relief operations. Back propagation (BP) neural network is an artificial neural network which combines multi-layer feed-forward network and error back-propagation algorithm. It has a strong input-output mapping capability, and does not require the object to be identified obeying certain distribution law. It has strong non-linear features and error-tolerant capabilities. Remotely-sensed image classification can achieve high accuracy and satisfactory error-tolerant capabilities. But it also has drawbacks such as slow convergence speed and can probably be trapped by local minimum points. In order to solve these problems, we have improved this algorithm through setting up self-adaptive training rate and adding momentum factor. UAV high-resolution aerial image in Taoguan District of Wenchuan County is used as data source. First, we preprocess UAV aerial images and rectify geometric distortion in images. Training samples were selected and purified. The image is then classified using the improved BP neural network algorithm. Finally, we compare such classification result with the maximum likelihood classification (MLC) result. Numerical comparison shows that the overall accuracy of maximum likelihood classification is 83.8%, while the improved BP neural network classification is 89.7%. The testing results indicate that the latter is better.

Paper Details

Date Published: 9 October 2009
PDF: 9 pages
Proc. SPIE 7471, Second International Conference on Earth Observation for Global Changes, 74711E (9 October 2009); doi: 10.1117/12.836316
Show Author Affiliations
Na Lin, Chengdu Univ. of Technology (China)
Chongqing Technology and Business Univ. (China)
Wunian Yang, Chengdu Univ. of Technology (China)
Bin Wang, Chongqing Geomatics Ctr. (China)

Published in SPIE Proceedings Vol. 7471:
Second International Conference on Earth Observation for Global Changes
Xianfeng Zhang; Jonathan Li; Guoxiang Liu; Xiaojun Yang, Editor(s)

© SPIE. Terms of Use
Back to Top