|
Application of convolutional neural networks single-shot multibox detector (SSD) in automatic detection and extraction of Qanat from Google Earth satellite images
|
Mostafa Kabolizadeh *   , Mohammad Abbasi |
| Shahid Chamran University of Ahvaz |
|
|
Abstract: (968 Views) |
Identifying qanat and mapping them is one of the vital tasks in the science of groundwater resources management. However, it is difficult to map qanats because conventional methods such as field mapping are very costly and time consuming, and in some cases, due to the structure of qanat, land extraction faces many challenges. Identifying objects from image data using computer image processing techniques and neural networks is one of the most promising techniques for identifying qanats. Due to the fact that satellite images of Google Earth system are the only satellite images with high spatial resolution, free of charge and available, in this research, Google Earth satellite images have been used. In this research, more than 600 educational samples have been prepared from the openings of qanats. In this research, the convolutional neural network single-shot multi-box detector based on ResNet network has been used to automatically detect and extract the geometric location of qanats due to higher processing speed. The proposed model is taught by training samples based on 85% of the training data and 15% of the validation data, with 50 repetition courses and an accuracy of more than 90%. The trained model is implemented on the image of the study area to discover the shafts of qanats. The results show that this model can work with the accuracy criterion equal to 0.91, the recovery criterion equal to 0.82, and the F1Score criterion equal to 0.86 in discovering the location of the rod of the aqueduct wells with a suitable shape. For the areas where the shape of the rod of the wells destroyed, the accuracy of detecting and extracting the position of the wells equal to 0.65. This research shows that it is possible to automatically identify and extract aqueduct well rods with appropriate accuracy from satellite images downloaded from the Google Earth system.
|
|
| Keywords: Convolution Neural Network, Single Shot Multi-box Detector, Qanat, ResNet, Google Earth. |
|
|
Full-Text [PDF 1510 kb]
(307 Downloads)
|
Type of Study: Research |
Subject:
Aerial Photogrammetry
Received: 2023/07/12 | Accepted: 2023/11/22 | Published: 2023/12/21
|
|
|
|
|
|
|
| Send email to the article author |
|
|
