Change Detection in Residential Areas Construction by Using Photogrammetry Products

 
Introduction: Change detection, one of the applications of remote sensing images and photogrammetric data, with a history of over four decades in various military and civilian domains, plays an important role in urban management, crisis management, monitoring natural resources, ensuring security, and governmental governance. Monitoring and controlling changes within the boundaries and urban areas, especially in addressing unauthorized land-use changes, is one of the most critical needs of urban management. For this purpose, using classical methods, despite their simplicity and accessibility, lacks the necessary efficiency due to limitations in accuracy, speed, and comprehensiveness. On the other hand, implementing new deep learning-based methods such as neural networks also faces challenges due to the difficulties in preparing training data, being time-consuming and costly, and requiring powerful computational and hardware resources. This paper aims to present a relatively fast, cost-effective, and high-accuracy process for detecting and identifying changes in residential areas.
Material and Methods: The proposed process, aimed at overcoming the limitations of previous methods, is based on the use of photogrammetric products, including Digital Surface Models (DSM) and orthophotomosaics, along with the application of various filters. The input data, with horizontal and vertical accuracies better than 30 cm, have been prepared and enable the identification of buildings that have undergone changes over time. The proposed process involves generating a Digital Difference Model (DDM) by subtracting two-time DSMs, which visualizes height changes in both positive and negative directions. Initial targets are then extracted by applying height and area threshold limits, combined with multiple filtering stages on the input data. To reduce recognition errors caused by factors such as shadows, vegetation, vehicles, and other existing features, orthophotomosaic classification using intelligent algorithms is performed and applied to the Digital Difference Model. This step reduces the impact of interfering features and leads to the extraction of the final targets.
Results and Discussion: To evaluate the performance of the proposed process, two study areas in Yazd Province were selected: one with a simple urban texture and the other with a complex urban texture. The data include orthophotomosaics with a pixel size of 10 cm for both study areas. Additionally, the Digital Surface Models (DSMs) have pixel sizes of 40 cm and 10 cm, respectively. It is worth noting that the time interval between data acquisitions was two months for the first study area and three years for the second. The results of implementing the proposed process achieved an overall accuracy of over 90% in the first study area and over 83% in the second. Optimal values for height and area thresholds and filter settings were determined through a trial-and-error process, by defining various events and precisely analyzing the counts of correct targets, missed targets, and false targets to achieve the highest accuracy. Analysis and evaluation of the proposed process show that applying appropriate filters in four stages increased the overall algorithm accuracy by more than 30%.
Conclusion: The proposed process is highly dependent on the study area and the threshold values corresponding to its urban texture. Despite this limitation, the presented method, due to its lower cost and higher speed compared to similar methods, has broad applicability in areas similar to those studied. Additionally, the results of this paper show that this process, due to its high accuracy and acceptable results, can serve as an effective tool in the field of urban management and monitoring authorized and unauthorized changes, contributing to improving decision-making processes in this domain and effectively addressing the need for urban boundary control. For other areas with textures different from those in this study, it is necessary to calculate optimal values for operational components and thresholds using a similar methodology