The conversion of cropland to forest project as one of the nationwide forestry key programs was started in 1999, aiming at prevention of soil erosion, water resources conservation, combating land desertification in the key areas and mitigation of global climate change. Afforestation is the primary measure of the Project. Because of low land quality, ecological species planted and short-term growing, mid-coarse satellite data is serious limited to check planting effect in many areas. With the development of remote sensing, finer images provide better choice to exactly detect the distribution of afforestation parcels and forested area of the Project. In this study, an exact automatic recognition algorithm for the Project parcels is developed. It is based on finer RS data (less than 5m spatial resolution), and integrates object-oriented, BP artificial neural network, multi-segmentation and expert knowledge, considering the differential features between trees and other land vegetation in terms of image hue, grey value, textural features, RGB and NDVI. Kangping county is taken as an example to test the algorithm at the support of 5-times SPOT5 images since 2003. The study results show that: 1) most of planted parcels can be extracted more than two years old, and the accuracy can reach 85%, 2) closed arbor planted longer than forested year limit can be clear detected, the total accuracy for forest extraction of the Project parcels exceeds 85%.
[1]
Brijesh Verma,et al.
A Fusion of Neural Network Based Auto-associator and Classifier for the Classification of Microcalcification Patterns
,
2004,
ICONIP.
[2]
Ahmet B. Orun,et al.
Automated identification of man-made textural features on satellite imagery by Bayesian networks
,
2004
.
[3]
Wei Fengyu,et al.
Geographical distribution of plants of Celastraceae in China.
,
2011
.
[4]
Kuang Gang-yao,et al.
Overview of Image Textural Feature Extraction Methods
,
2009
.
[5]
Bao Shi-tai,et al.
Application of Object-Oriented Interpretation to Information Classification on Remote Sensing Image
,
2006
.
[6]
Yan Mei,et al.
Deriving Bamboos from IKONOS Image by Texture Information
,
2004
.
[7]
U. Benz,et al.
Multi-resolution, object-oriented fuzzy analysis of remote sensing data for GIS-ready information
,
2004
.
[8]
L. Halounová.
Textural classification of B & W aerial photos for the forest classification
,
2003
.