Abstract:

The UAV platform was used to acquire multispectral images and visible images of subtropical broad-leaved forest sample sites,and the multispectral images and visible images were band-synthesized to obtain a visible combination image with multispectral information.The dominant species were classified using four classification algorithms: support vector machine,maximum likelihood,neural network and Marxian distance.The classification results of different images were compared under different classification algorithms.The classification accuracies using only visible images in different algorithms were low,with the highest accuracy being the classification using support vector machines with 62.97% and a kappa coefficient of 0.225 6.The classification accuracies of the visible combined images using multispectral images and with the addition of multispectral images were improved.The combination of the combined image and the support vector machine achieved a classification accuracy of 90.64% with a Kappa coefficient of 0.786.The use of low-cost UAV visible images had a low classification accuracy,while the addition of multispectral images could improve the classification accuracy significantly.

Key words: multispectral, visible light, subtropical broadleaf forest species, tree species classification