Author(s): Salah Ataallah Awad

Abstract: The study aims to employ deep learning methods to interpret geometric drawings in spatial data to increase the accuracy and quality of the surveying task allocation process. Vector data is a fundamental component for surveying and geographic information systems (GIS) specialists. Due to the significant development and complexity in this field, there is a need to develop new methods that are more accurate and effective compared to traditional methods based on utilizing prior information, which is prone to errors and missing important information. A comparison was made between the effectiveness and role of deep learning models and methods that directly address sequences of geometric coordinates, and simpler traditional models based on surface learning models, which rely on traditional methods such as elliptic Fourier descriptors. The study consisted of three types of classification tasks. The first relied on neighborhood shapes to classify residential areas, the second relied on building plans to classify and identify building types, and the third relied on the shapes of archaeological sites to identify and classify their archaeological characteristics. A set of data was obtained from open sources and divided into several sections: one for training, one for validation, and one for testing. Two deep learning models were developed: a convolutional neural network and a bidirectional LSTM network, in addition to four traditional models (k-NN, logistic regression, SVM, and decision tree). The study concluded that deep learning models were more efficient in three of the tasks. For the fourth task, surface models achieved higher performance and efficiency using low-order Fourier descriptors, while higher-order models were inefficient. Deep learning models faced challenges, manifested in the need for extensive and expensive equipment and massive data. Therefore, the study proposed integrating diverse data sources and developing LSTM models. The study also recommended using deep learning, with its different models, in the analysis and classification of vector spatial information, which would make significant progress in the study engineer.

DOI: 10.22271/2707840X.2025.v6.i1a.32

Pages: 01-08 | Views: 82 | Downloads: 33

Download Full Article: Click Here