Vegetation extraction from digital orthophoto maps using object-based segmentation and decision tree classifier
AbstractThe main objective of this study is to automatically extract tea gardens from large geographic areas using high-resolution digital orthophoto maps. To achieve this objective, object-based image analysis and decision tree (DT) classifier were integrated. For segmentation, multi-resolution image segmentation algorithm was used which is implemented in Definiens Developer commercial software. Both scale and compactness parameters were empirically calculated that produced optimal results. The segmented objects were selected manually for training the DT classifier from all used images. Spectral and textural features were extracted from each segment and to make the features robust against local variations, they were extracted at two image scales and final feature vector was formed by averaging the two feature vectors. The selected optimal features were used to train the DT classifier and then applied it on the test data to generate thematic maps for tea gardens. The performance of the proposed method was evaluated by comparing results with the reference data that produced promising results for mapping tea gardens (overall accuracy 88%) on our dataset.