Jakarta, INTI - The National Research and Innovation Agency (BRIN) has developed an artificial intelligence (AI)-based method to extract coastlines in the North Java Road (Jalur Pantura) using Sentinel-2 satellite imagery. This innovation aims to improve the accuracy of monitoring dynamic coastline changes caused by natural factors and human activity.
This research was conducted by a BRIN research team using the U-Net deep learning model to automatically separate land and sea areas from satellite imagery. This method was applied along the northern coast of Java, which has complex characteristics, ranging from sandy beaches and muddy beaches to areas with intensive infrastructure intervention.
U-Net Model is Used for Accuracy
Edwin Adi Wiguna, a researcher at BRIN's Hydrodynamic Technology Research Center, explained that the Pantura region is a densely populated strategic area. However, this area is vulnerable to various environmental pressures such as abrasion, tidal flooding, and land subsidence.
"Utilizing the U-Net model allows for more precise identification of land and sea boundaries. Test results show a high level of accuracy, with an Intersection over Union value of around 92 percent," explained Edwin.
During the process, Sentinel-2 satellite imagery was processed through several stages, starting with dataset preparation and model training, and then through image segmentation. The U-Net model generated pixel classifications that differentiated land and sea areas, which were then further processed using an edge detection algorithm to obtain detailed coastlines.
The results showed excellent model performance on sandy and gravel beaches. However, challenges remained in muddy coastal areas, particularly in fishpond areas, due to their similar spectral characteristics to shallow waters.
Quantitatively, the average deviation of the shoreline extraction results from the reference data reached approximately 55.73 meters, with a maximum error of up to 326.45 meters under certain conditions (in complex and muddy areas). Nevertheless, the model consistently handled the complexities of the Pantura coastal area.
The Technology Can Be Used for Further Observations
Edwin added that this technology has great potential to support routine and efficient coastline monitoring.
"This approach can be used for regular coastline observations at a lower cost, thereby assisting the government in data-driven policymaking, particularly regarding mitigation of abrasion, tidal flooding, and shoreline changes," he added.
Through this innovation, BRIN continues to encourage the use of artificial intelligence and remote sensing technology to support adaptive and sustainable coastal area management and strengthen resilience to the impacts of climate change.
Conclusion
The National Research and Innovation Agency (BRIN) has developed an AI-based method using the U-Net model and Sentinel-2 satellite imagery to extract coastlines in the Pantura coastline with a high accuracy of up to 92%. This technology can monitor coastal changes more precisely and efficiently, although it still faces challenges in muddy areas such as fish ponds. According to researcher Edwin Adi Wiguna, this innovation has the potential to support routine monitoring and assist the government in mitigating abrasion and tidal flooding.
Read more: Ericsson Said Cellular Network Consumers’ Satisfaction will Also Depend on AI Speed