Search

The paper used the unmanned aerial vehicle (UAV) which can quickly obtain a wide range of vegetation images to capture the low-altitude aerial photograph)' and associated with the multivariate HoG features to recognize vegetation types. Firstly, the texture infonnation of the image were extracted by Gabor filter and using HSV and Lab color space transform to extract the color infonnation. Next, the image were divided into N cells, then computing the gradient of histogram of each cell based on the texture and color infonnation to fonn the multivariate HoG features. Finally, taking the cell as the classification unit, combining the random forest machine learning algorithm, vegetation types recognition model were established. Taking the Anxi County in Fujian Province as the test area, the results showed that it was feasible to identify the vegetation types by using the optical image of low-altitude aerial photography; the highest recognition accuracy in recognizing vegetation and non-vegetation was 96.04%; the recognition accuracy rate was 82.44% in recognizing vegetation types at 20 m and decreased with the height increase. Moreover, the vegetation images of aerial photography were further collected, which was captured at the Changting County of Fujian Province, as another data to prove the robustness of the model of vegetation type recognition in different areas, the highest recognition accuracy was 73.31%, the accuracy was similar compared with the result of Anxi County. In this paper, the vegetation images by using the unmanned aerial vehicle with optical camera were conveniently, econmically obtained; the proposed vegetation type recognition model can be easily applied to the field vegetation type monitoring and management as the high precision and its robustness in various regions. In addition, according to the recognition result at various aerial height, the proposed aerial height should not be too high, 20 m as the reference height of aerial photography was suitable. [ABSTRACT FROM AUTHOR]