Rubber Tree Recognition Based on UAV RGB Multi-Angle Imagery and Deep Learning

• DOI: 10.3390/drones7090547
• Publication Date: 2023-08-24T14:23:40Z
• Authors: Yuying Liang; Yongke Sun; Weili Kou; Weiheng Xu; Juan Wang; Qiuhua Wang; Huan Wang; Ning Lu
• Keywords: RGB color model; Tree (set theory); Hevea

The rubber tree (Hevea brasiliensis) is an important species for the production of natural latex, which essential raw material varieties industrial and non-industrial products. Rapid accurate identification number trees not only plays role in predicting biomass yield but also beneficial to estimating carbon sinks promoting sustainable development plantations. However, existing recognition methods based on canopy characteristic segmentation are suitable detecting individual due their high coverage similar crown structure. Fortunately, have a defoliation period about 40 days, makes trunks clearly visible high-resolution RGB images. Therefore, this study employed unmanned aerial vehicle (UAV) equipped with camera acquire images plantations from three observation angles (−90°, −60°, 45°) two flight directions (SN: perpendicular planting row, WE: parallel rows) during deciduous period. Four convolutional neural networks (multi-scale attention network, MAnet; Unet++; Unet; pyramid scene parsing PSPnet) were utilized explore trunk counting. results indicate that Unet++ achieved best accuracy (precision = 0.979, recall 0.919, F-measure 94.7%) angle −60° mode SN among four deep learning algorithms. This research provides new idea by multi-angle forests specific phenological periods.