Document Type : Original Article

Authors

• Fatemeh Babaei ¹
• Gholamreza Fallahi ²

¹ Master's Student in Land Administration Systems, School of Civil, Water and Environmental Engineering, Shahid Beheshti University, Tehran, Iran

² Assistant Professor, School of Civil, Water and Environmental Engineering, Shahid Beheshti University, Tehran, Iran

Abstract

Accurate delineation of agricultural land parcels is a key requirement for implementing precision agriculture, natural resource management, and the development of Spatial Data Infrastructures (SDI). Considering the diversity of planting patterns, vegetation changes, and challenges such as occlusions, heterogeneous landscapes, and limited labeled data, traditional image processing and manual interpretation methods lack sufficient efficacy. In recent years, deep learning models have emerged as innovative solutions for extracting parcel boundaries from satellite imagery and spatial data due to their high capability in extracting complex features and processing large-scale data. This study aims to investigate the role of deep learning models in accurately delineating agricultural land parcel boundaries based on the analysis of satellite images and spatial data within the framework of Spatial Data Infrastructure (SDI). The research methodology is descriptive-analytical and based on a comprehensive literature review of reputable scientific sources. The study focuses on analyzing various deep learning architectures, including Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Generative Adversarial Networks (GANs). These models are evaluated from the perspectives of technical structure, input data types, performance metrics, and adaptability to operational challenges in precision agriculture. For comparative analysis, a set of recent studies and selected models are reviewed regarding accuracy, limitations, and compatibility with real-world conditions such as heterogeneous landscapes and scarce labeled data. The results indicate that CNN-based models perform well in processing satellite imagery but have limitations in capturing contextual dependencies, which can be improved by combining them with RNNs. Additionally, GAN models are effective in augmenting training data and generating synthetic images. The findings of this study can serve as a foundation for developing more intelligent and hybrid models in future SDI and smart agriculture systems.

Keywords

• Deep Learning
• Satellite Imagery
• Smart Agriculture
• Convolutional Neural Networks (CNN)
• Spatial Data Infrastructure (SDI)