BiFPN-KPointNet-CBAM: Application of 3D Point Cloud Technology Based on Deep Learning in Measuring Vegetation

Abstract

The results of traditional vegetation-measuring methods are mostly two-dimensional data, which can only convey limited information. The greening situation of many cities or regions in the world cannot be fully assessed by these results. In this regard, this paper proposes the use of the air–ground integrated point cloud data acquisition mode for measuring vegetation. This mode combines a backpack-mounted laser scanning system, a vehicle-mounted laser scanning system, and UAV tilt photography technology to collect greening data in a comprehensive park and along a municipal road in Guangzhou, China. To classify the collected greening data, we propose the BiFPN-KPointNet-CBAM model, which was derived from PointNet. The model was introduced to analyze the distribution of green plants in study areas. The experimental findings indicate that our model achieved a notable enhancement in the overall accuracy by approximately 8% compared with other state-of-the-art models. Compared with the traditional greening survey method, this method obtained three-dimensional and more accurate greening data, and thus, provides higher quality greening data for urban managers.