Computer Recognition of Species Using Intelligent UAV Multispectral Imagery

Bamboo and eucalyptus are the main tree species that may become tree barriers along the transmission lines in Guangning County, Zhaoqing City, Guangdong Province, China, due to their fast growth and high canopy. Thus, accurate recognition of bamboo and eucalyptus is essential for early warning of tree barriers along the transmission lines in this area. In this study, multi-spectral imagery (including 450, 555, 660, 720, 750 and 840 nm) was obtained by using unmanned aerial vehicle (UAV) in Guangning County, and eight Vegetation Indices (VIs) based on spectral bands were calculated. Then, four classification schemes using different features combinations were designed: (1) 4 spectral bands without red edge band; (2) 6 spectral bands including two red edge bands; (3) Spectral bands combined with unoptimized VIs; (4) Spectral bands combined with optimized VIs. Random forest (RF) approach was applied to evaluate the importance of VIs features for feature optimization and recognize tree species. The results show that the overall accuracy of scheme 2 was 6.46% higher than scheme 1, scheme 4 had the highest classification accuracy of 90.11%, indicating that the red edge bands and VIs can improve the recognition accuracy of tree species, and compared to scheme 3, the classification accuracy of scheme 4 was increased by 0.72%, indicating that feature optimization can reduce the number of input features while maintaining the classification accuracy, thereby improving the efficiency of classification. This study can effectively recognize eucalyptus and bamboo and provide a sound reference for early warning in the transmission lines.