Extracting Desert Vegetation Feature Information based on Spectral Angle Mapping Integrated Random Forest Algorithm

Desert vegetation in semi-arid and arid locations prevents wind and sand erosion, conserves water and soil, and maintains a balanced ecosystem. To identify features that differentiate vegetation texture from barren surfaces like sand is difficult with spectral-only methods. To overcome these, the re...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:2024 International Conference on Distributed Systems, Computer Networks and Cybersecurity (ICDSCNC) s. 1 - 5
Hlavní autor: Zhang, Tao
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 20.09.2024
Témata:
Data mining
desert vegetation characteristics
Deserts
Feature extraction
gray-level co-occurrence matrix
Indexes
Nearest neighbor methods
Noise
Radio frequency
random forest
spectral angle mapping
Support vector machines
Surface texture
Vegetation mapping
visible difference vegetation index
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Desert vegetation in semi-arid and arid locations prevents wind and sand erosion, conserves water and soil, and maintains a balanced ecosystem. To identify features that differentiate vegetation texture from barren surfaces like sand is difficult with spectral-only methods. To overcome these, the research proposed a hybrid approach that integrates Spectral Angle Mapping (SAM) with Random Forest (RF). The SAM-RF along with Visible Difference Vegetation Index (VDVI) and Gray-Level Co-occurrence Matrix (GLCM) is also used. This hybrid approach helps to extract spectral features and also describe the spatial relationship i.e., spectrally based data and increases the discriminative ability of the method. The Unmanned Aerial Vehicles - Red, Green, Blue (UAV-RGB) images are considered as input data and pre-processed to reduce the noise and enhance the image contrast for better quality. The extracted features with SAM-RF along with VDVI and GLCM are classified using RF and achieved better classification accuracy of 99.49%, precision of 97.32%, and recall of 95.48% when compared to the traditional methods such as K-Nearest Neighbor (KNN) and Support Vector Machines (SVM).
DOI:10.1109/ICDSCNC62492.2024.10939337