INVESTIGATING MEGA-CUSP SHAPE EXTRACTION THROUGH AERIAL PHOTOS AND OBJECT DETECTION ALGORITHMS

Abstract

 Mega cusps are critical for risk management due to their connection with rip currents and erosion in the swash zone. This study aims to explore the effectiveness of aerial photos and object detection algorithms in capturing mega cusp shapes, focusing on both edge-based and object-based methods. The research area focused on the Hasaki Coast in Kamisu City, Ibaraki Prefecture, and data were gathered from aerial photos and topographic data sources on June 6th, 2023. Four algorithms—Random Forest, Otsu's, Canny Edge, and Laplacian detection—were utilized to detect mega cusp shapes, extending their use beyond identifying sharp edges of objects such as buildings and roads to extracting smooth features that represent mega cusp shape lines. The Random Forest algorithm closely aligns with topographic data, detecting the mega cusp shape extraction in the area between wet and dry sand after classifying sand pixel colors. Meanwhile, the Otsu’s, Canny Edge, and Laplacian algorithms focus on the sand berm object, which has edges. Edge-based algorithms, such as Canny Edge and Laplacian, concentrate on the top edge of the sand berm, whereas Otsu’s algorithm focuses on the base of the berm sand. Elevation data extracted through Digital Terrain Models (DTM) and GIS tools highlight the importance of detailed examination for understanding mega cusp shape elevations. The Random Forest algorithm provides an average elevation of 1.26 m, similar to the high tide level, with 0.01 standard error of elevation, while Otsu’s, Canny Edge, and Laplacian algorithms yield higher elevations ranging from 2.03 m, 2.25 m, and 2.20 m, respectively, with a larger standard error of position. In conclusion, the study recommends the Random Forest algorithm, a machine learning-based approach, emphasizing its ability to detect the mega cusp object using a training dataset during a neap tide.