|
Provide a Deep Convolutional Neural Network Optimized with Morphological Filters to Map Trees in Urban Environments Using Aerial Imagery
|
Moslem Darvishi   , Reza Shah-Hosseini * |
| School of Surveying & Geospatial Engineering, College of Engineering, University of Tehran, Tehran, Iran. |
|
|
Abstract: (1547 Views) |
| Today, we cannot ignore the role of trees in the quality of human life, so that the earth is inconceivable for humans without the presence of trees. In addition to their natural role, urban trees are also very important in terms of visual beauty. Aerial imagery using unmanned platforms with very high spatial resolution is available today. Convolutional neural networks based deep learning methods are very efficient in processing large amounts of data from this type of imaging and facing its special challenges. The current paper proposed an encoder-decoder networks based convolutional neural network architecture, the results of which are based on reference data provided by the International Society of Photogrammetry and Remote Sensing from Potsdam, Germany for urban tree detection, provide 96.10% overall accuracy and F1 score equals 80.72%. However, very little change in accuracy can be attributed to the time of imaging, at which the trees in the study region generally did not have any greenery. Another discussion in this research is to create a differential image of the terrestrial reality map and the output estimation map of the proposed algorithm. According to this differential map, the application of a morphological filter seems to be useful, which in practice has increased the accuracy of the final classification. The final issue is the use of training network parameters by red, green and blue band combination as primary parameters for the network with near, red and green infrared band input, in which case the time to achieve the best network performance is reduced by about 83% and this pre-training of neural network parameters has caused the rapid convergence of the target network. |
|
| Keywords: Deep Learning, Remote Sensing, Convolutional Neural Network, High Spatial Resolution Aerial images |
|
|
Full-Text [PDF 1267 kb]
(359 Downloads)
|
Type of Study: Research |
Subject:
RS
Received: 2021/09/22 | Accepted: 2022/09/21 | Published: 2022/11/1
|
|
|
|
|
|
