Detecting urban tree canopy using convolutional neural networks with aerial images and LiDAR data

Abstract

The detection of urban tree canopy plays a crucial role in assessing the ecosystem of trees and reducing greenhouse gases in smart cities. This research proposes an intelligent model for detecting tree canopy in urban environments using aerial images and LiDAR data, leveraging convolutional neural networks (CNNs). The proposed models have been trained and evaluated in urban areas with vegetation in Qom city. To accomplish this, three datasets were utilized to train a single model. The first dataset derived from LiDAR data, achieved an accuracy of 88.05% with a loss of 0.341, indicating that the model made correct predictions with a high percentage but had some errors. Similarly, in the second dataset utilizing aerial image data, the algorithm achieved a higher accuracy of 90.04% with a lower loss of 0.298, suggesting improved performance with fewer mistakes. Lastly, in the third dataset, which incorporated data derived from both LiDAR and aerial images, the algorithm achieved an even higher accuracy of 91.05% with a lower loss of 0.276, indicating further enhancement in prediction accuracy and reduced errors. On the other hand, the third model demonstrates an average value of 94%, 83.1%, and 78.9% for completeness, correctness, and quality, respectively, in identifying tree canopies. Completeness pertains to the CNN's precision in detecting and extracting pertinent features from the input data, while correctness relates to the accuracy of the CNN's predictions. Furthermore, quality encompasses the overall performance and dependability of the model. This indicates that the integration of aerial images and digital surface model (DSM) data acquired from LiDAR, along with the utilization of convolutional Neural Networks (CNNs), enhances the outcomes compared to alternative models.