University of Twente Student Theses


Comparison of low-cost methods for vegetation mapping using object based analysis of UAV imagery: a case study for the greater Côa Valley, Portugal

Sharma, Meenal (2022) Comparison of low-cost methods for vegetation mapping using object based analysis of UAV imagery: a case study for the greater Côa Valley, Portugal.

[img] PDF
Abstract:Conservation planning for a highly diverse vegetated area and its ecosystem relies on accurate vegetation maps. Unmanned Aerial Vehicles (UAVs) have been proven reliable, efficient, and cost-effective tools in improving the monitoring of natural habitats and facilitating their management. UAV allows monitoring of remote and inaccessible areas through ultra-high resolution imagery. The conventional method of using ground survey is expensive and requires manual efforts, while UAVs with various advantages like cost-effective, flexible, and high spatial resolution can be used efficiently for mapping vegetation in a heterogeneous landscape. Using UAV high resolution images and Structure from Motion (SfM) processing, the research aims to assess the low-cost methods and the effectiveness of UAVs in vegetation mapping on the aspect of image compression, sensors, and seasonality. For UAV data acquisition, this research used the DJI-Phantom 4 RGB camera and Parrot Sequoia multispectral camera together at two sites in Greater Côa Valley, Portugal. The study examines the feasibility and potential of using SfM photogrammetry and object-based image analysis (OBIA) to classify vegetation into seven dominant life form classes. The workflow of OBIA consisted of segmentation and classification using Random Forest in eCognition software. The workflow was used to assess the efficiency of the methods in improving the classification accuracy of vegetation maps. Further, each classification results were compared using accuracy assessment i.e., overall accuracy and kappa value. The study investigated the effect of image compression (raw DNG format and JPEG format) in vegetation classification, yielding a better classification accuracy in uncompressed RGB images (overall accuracy 80.65% and kappa 0.77) compared to compressed RGB images (overall accuracy 62.04% and kappa 0.54). Secondly, vegetation classification was compared between sensors, for which RGB camera and Parrot Sequoia multispectral camera were used to acquire RGB and multispectral images, respectively. The classification accuracy of different classes improved using multispectral images (overall accuracy 85.19% and kappa 0.83) compared to RGB images. The study also assesses the importance of Ground Control Points (GCPs) in vegetation mapping. Lastly, to determine the usefulness of seasonality in vegetation classification, single-season image classification was compared with tri-seasonal image classification. The outcome of vegetation classification for seasonal comparison showed that tri-seasonal images have higher accuracy in each life-form class than single-season images. The overall accuracy resulted in 92.99% and 0.91 kappa value for the orthomosaic of the combination of three seasons images. The results from the study show that SfM processing with OBIA is a useful method for vegetation classification, and OBIA is sensitive to the number of training samples used for image classification. It is also found that multispectral sensor is useful in identifying different life-form, but the combination of seasonal image classification (using RGB images) has resulted in the best overall classification accuracy. However, GCPs are significant in aligning multi-temporal images and can be helpful for future monitoring purposes. From the study, these low-cost methods have shown potential in classifying vegetation precisely, but processing requires high computational storage. The study elaborates that these accurate vegetation maps can be used to monitor vegetation changes for the factors such as the introduction and increase of herbivores and grazing in the area. The present research results offer valuable information for optimizing a UAV-based vegetation mapping protocol to be used in all Rewilding areas in Europe and other conservation areas.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page