Title: Tree recognition based on HSV-corrosion method and SSD method combining colour complexity and adaptive switching idea
Authors: Meihui Song; Yufeng Wang; Qiaoling Han; Yandong Zhao; Yue Zhao
Addresses: School of Technology, Beijing Forestry University, Beijing, China; Beijing Laboratory of Urban and Rural Ecological Environment, Beijing Municipal Education Commission, Beijing 100083, China ' School of Technology, Beijing Forestry University, Beijing, China; Beijing Laboratory of Urban and Rural Ecological Environment, Beijing Municipal Education Commission, Beijing 100083, China ' School of Technology, Beijing Forestry University, Beijing, China; Beijing Laboratory of Urban and Rural Ecological Environment, Beijing Municipal Education Commission, Beijing 100083, China ' School of Technology, Beijing Forestry University, Beijing, China; Beijing Laboratory of Urban and Rural Ecological Environment, Beijing Municipal Education Commission, Beijing 100083, China ' School of Technology, Beijing Forestry University, Beijing, China; Beijing Laboratory of Urban and Rural Ecological Environment, Beijing Municipal Education Commission, Beijing 100083, China
Abstract: The recognition and quantification of trees in forest remote sensing images has become an important task. However, due to the complex background of forest remote sensing images, the accuracy of tree identification is low. To solve this problem, an adaptive switching tree recognition method based on HSV-corrosion method and single shot multibox detector (SSD) method is proposed. Firstly, the switching factor F was determined by the colour complexity. Then, based on the switching factor, forest remote sensing images are divided into two sub-datasets. Finally, HSV-corrosion and SSD methods were combined to adaptively identify trees in two types of forest remote sensing images. The HSV-corrosion method has better recognition performance for dataset 1. In contrast, the SSD method has better recognition performance for dataset 2. Compared with the HSV-corrosion method and SSD method, the proposed method had a higher accuracy (0.86), F1-score values (0.69) and efficiency for all forest remote sensing images.
Keywords: adaptive switching; deep learning; tree recognition; single shot multibox detector; SSD; colour feature; HSV colour space.
DOI: 10.1504/IJSCC.2022.122271
International Journal of Systems, Control and Communications, 2022 Vol.13 No.2, pp.174 - 192
Received: 05 Nov 2021
Accepted: 13 Jan 2022
Published online: 14 Apr 2022 *