IDENTIFICATION OF MARINE DEBRIS ON THE INDRAMAYU COAST USING AN UNMANNED AERIAL VEHICLE (UAV) WITH A SUPERVISED LEARNING METHOD

• Main Author: Irandah, Annisa
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/81056
• Institution: Institut Teknologi Bandung

Marine debris has become a global issue related to its large impact on ecosystems, humans, the economy, coastal aesthetics, and others. Given the high levels of marine debris pollution over time, proper target handling is required to minimize the impact. Advances in remote sensing technology provide effective solutions for environmental monitoring, including monitoring marine debris in coastal areas. In this research, dominant marine debris was identified using an Unmanned Aerial Vehicle (UAV) and direct transects on the Moi-Moi Sea Coast and Junti Beach. Air photography was taken five times, each at different times, to look at the factors that influenced the marine debris. Pickups on the Moi-Moi coast were at 09:46, 09:56, 11:24, 11:48, and 14:28, while on the Junti coast, pickups were at 9:08, 10:46, 11:10, 14:18, and 14:59. The orthophoto was taken using DJI Phantom 4 Multispectral and DJI Mavic Air 2S UAVs flying at a height of 50 m at Moi-Moi Coast and 35 m at Junti Coast. The research uses supervised learning methods such as Object-Based Image Analysis (OBIA) and U-Net Architecture. Training data used for OBIA is 1.249 for sand, 780 for plastic, and 46 for wood, so the total training data is as much as 2.075 data. OBIA uses scale parameters 15, shape 0,1, and compactness 0,9. U-Net uses 114 images of 512x512 pixels, using epoch 50, batch size 8, and Adam Optimizers. From the results of the direct transactions, it was found that the waste that dominates on both shores is plastic and wood, with a percentage of plastic of 33,58% and wood of 23,33% on the Moi-Moi Sea Coast, while on the Junti Coast, the percentage of plastic is 39,80% and wood is 46,59%. Thus, in the process of classification, three classes are defined: wood, sand, and plastic. The OBIA method uses the results of the classification to identify the type of marine debris and yield the highest accuracy value of 0.97. Whereas U-Net obtains the final value for loss of 0,48, validation loss of 0,50, accurate value of 0,97, and validation accuracy of 0,96. Based on 5 aerial photos taken at the Moi-Moi Sea Coast and Junti Beach, it can be concluded that marine debris in the intertidal zone is affected by swash and backwash. This is based on the appearance of marine debris increasing, decreasing or moving in each orthophoto produced.