Liu Huaipeng, PhD

Extracting key features from remote sensing imagery for the purpose of identifying tree species holds significant importance. To assess the importance of the newly proposed polygon area index (PAI) and other features in classifying tree species, this research employed WorldView-3/2 (WV-3/2) as the experimental dataset and constructed four distinct feature sets: spectral bands, HSVs (hue, saturation, and value), textures and PAIs. Then, these feature sets and their combinations were utilized, and the random forest method was applied to classify tree species and ascertain the importance of the features involved in the classification process. The experimental results demonstrate that all texture features play a crucial role in the identification of tree species, with eight features being selected. Additionally, numerous HSV features (WV-3: 29, WV-2: 21) and PAI features (WV-3: 7, WV-2: 4) also make significant contributions. However, the spectral bands did not show a notable positive impact, as none were selected. When compared with the use of single-type features, the integration of multiple features from WorldView-3/2 resulted in more effective tree species identification, achieving an overall accuracy of 90.45% for WV-3 and 84.60% for WV-2, surpassing the highest overall accuracy achieved by single-type features, which was 76.64% for WV-3 and 78.27% for WV-2. The experimental results indicate that multi-type features should be used for tree species classification. Additionally, the newly proposed PAI feature demonstrates a positive contribution and is recommended for active use in tree species classification.

Extracting key features from remote sensing imagery for the purpose of identifying tree species holds significant importance. To assess the importance of the newly proposed polygon area index (PAI) and other features in classifying tree species, this research employed WorldView-3/2 (WV-3/2) as the experimental dataset and constructed four distinct feature sets: spectral bands, HSVs (hue, saturation, and value), textures and PAIs. Then, these feature sets and their combinations were utilized, and the random forest method was applied to classify tree species and ascertain the importance of the features involved in the classification process. The experimental results demonstrate that all texture features play a crucial role in the identification of tree species, with eight features being selected. Additionally, numerous HSV features (WV-3: 29, WV-2: 21) and PAI features (WV-3: 7, WV-2: 4) also make significant contributions. However, the spectral bands did not show a notable positive impact, as none were selected. When compared with the use of single-type features, the integration of multiple features from WorldView-3/2 resulted in more effective tree species identification, achieving an overall accuracy of 90.45% for WV-3 and 84.60% for WV-2, surpassing the highest overall accuracy achieved by single-type features, which was 76.64% for WV-3 and 78.27% for WV-2. The experimental results indicate that multi-type features should be used for tree species classification. Additionally, the newly proposed PAI feature demonstrates a positive contribution and is recommended for active use in tree species classification.