Cross‐Border Land Tenure Detection With Satellite Imagery Data Using Siamese Neural Network: A Deep Learning Approach

Abstract

ABSTRACT Land disputes, particularly cross‐border issues, are a significant source of conflict in Ethiopia, where about 80% of the population depends on land‐based livelihoods and agriculture. These disputes significantly affect the rural population, whose primary source of income depends on land. To address this challenge, this study has investigated the application of artificial intelligence methods for automated detection and analysis of disputed land parcels using satellite imagery data. In particular, a Siamese Neural Network (SNN) model was developed and evaluated on a benchmark dataset comprising 612 larger reference parcel images and 2170 cropped template satellite images (1731 disputed parcel‐training images and 439 corresponding reference‐testing images) extracted from high‐resolution satellite imagery in the Amhara region of Ethiopia. The proposed model achieved 95.31% training accuracy and 84.56% testing accuracy, with an F1‐score of 0.88 at the optimal threshold. Experimental results showed that the SNN's performance is sensitive to the selection of the similarity threshold, revealing a precision–recall trade‐off of up to 10% across tested thresholds from 0.85 to 0.99. In addition, the paper has introduced a benchmark high‐resolution satellite imagery (HRSI) dataset to support reproducibility and future comparative research. The findings highlighted the potential of Artificial Intelligence‐driven geospatial analysis to enhance transparency, reduce manual verification time, and support equitable land administration systems. However, the paper was limited to exploring more advanced deep learning methods, such as recent transformer‐based deep learning models, which could further improve the performance of the cross‐border land tenure detection model. Using only one of the four collected layers due to the lack of advanced GPU computational requirements is another limitation of this paper.