ANNOUNCEMENTS
The aim for the research is to classify different horticulture plantations using OBIA against pixel-based classification, due to high resolution data which makes it challenging to classification with traditional methods. Secondly, tree height estimation is attempted using elevation data, local maxima is used to extract each tree height. Tree height is essential for the calculation of biomass and further it is useful for carbon stock estimation.
The classification is attempted using Mean Shift Segmentation for Object-Based Image Analysis (OBIA) and the Support Vector Machine (SVM) algorithm for classification to assess the classification accuracy of horticultural plantations using UAV imagery. Mean Shift Segmentation aids in grouping pixels with similar spectral properties, crucial for identifying distinct regions corresponding to different crop types. The SVM model, trained using labeled samples, achieved an impressive overall accuracy of 86.8%, demonstrating its effectiveness in classifying Arecanut, Bamboo, Date Palm, and Jackfruit. To enhance classification accuracy, vegetation indices derived from UAV imagery were analyzed. NDRE, sensitive to chlorophyll content, proved particularly effective for classifying crops.
For tree height estimation, Canopy Height Models (CHMs) derived from UAV imagery was compared to ground-truth measurements. Box and whisker plots demonstrated high consistency between ground truth and UAV-derived estimates. Linear regression analysis, With an R-squared value (R²) of 0.83, the model explains 83% of the variance in ground-truth tree heights using the estimated heights. Additionally, the Root Mean Square Error (RMSE) of 0.48 meters reflects a high level of precision, indicating that the average deviation of the estimated heights from the actual measurements is less than half a meter.
Overall, this study contributes to the advancement of UAV-based methodologies for agricultural management and tree height estimation.
