Research on Deep Learning Individual Tree Segmentation Method Coupling RetinaNet and Point Cloud Clustering

Increasing human activities have caused serious disturbance to global forest resources, so how accurately identify individual trees has become an important task of resources investigation. In order get the accurate number trees, this paper took coniferous and mixed broad-leaved as experimental samples, well Digital Orthophoto Map airborne LiDAR Point Cloud research data. We propose a deep Learning tree segmentation method based on RetinaNet model PCS algorithm by doing comparative analysis (classical Watershed Algorithm Layer Stacking Algorithm) at plots (with high, medium, low densities). The results show that proposed in can solve problem high density improve its degree automation. Compared with stacking algorithm, F-Measure is improved 6%-29% 7%-20%, respectively. other words, presented not only precision segmentation, but also maintain detection rate, which meet accuracy efficiency extraction, needs for modern forestry