Kurniawan, Arya Fajar (2023) Automated Approaches Deep Learning Untuk Deteksi Reruntuhan Trestle Di Bawah Laut Menggunakan Data Multibeam Echosounder. Other thesis, Institut Teknologi Sepuluh Nopember.
Text
03311940000062-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2025. Download (4MB) | Request a copy |
Abstract
Pemanfaatan gelombang akustik contohnya penggunaan multibeam echosounder (MBES) yang ditujukan untuk memperoleh visualisasi bentuk topografi laut dan kedalaman. MBES juga dimanfaatkan salah satunya untuk mendeteksi keberadaan target/objek yang berada di dasar laut salah satunya adalah reruntuhan trestle. Biasanya dalam mendeteksi reruntuhan trestle umumnya melibatkan pengamatan visual manusia. Namun seiring dengan perkembangan kecerdasan buatan dan computer vision, dapat dilakukan dengan pendekatan deep learning dengan menggunakan metode yang diusul mengambil inspirasi dari Convolutional Neural Network (CNN). Hal tersebut menghasilkan suatu objek yang tersegmentasi pada masing-masing objek yang terdeteksi. Maka penentuan model deteksi reruntuhan trestle salah satunya dengan menggunakan arsitektur Mask R-CNN. Penggunaan epoch dan batch size per image yang berbeda juga membantu dalam mencari parameter terbaik yang digunakan. Hasil penelitian bisa dipilih opsi pada epoch ke 500 dalam batch size per image 64 karena selisih akurasi < 2% dan waktu training yang lebih rendah daripada pada epoch ke 1100. Dari variasi tersebut model berhasil mendeteksi delapan objek pada reruntuhan trestle dengan memperoleh tingkat score predict dengan nilai di atas 94%. Perhitungan selisih secara manual dan otomatis yang diperoleh RMSE terkecil sebesar 18,337m pada epoch ke 500 dalam batch size per image 64.
==================================================================================================================================
The application of acoustic waves is, for instance, the use of a multibeam echosounder (MBES) to visualise sea and depth topographic forms. MBES is also utilised to detect the existence of objects on the ocean floor, including the ruins of a trestle. Typically, trestle debris is detected through human visual observation. With the development of artificial intelligence and computer vision, however, this can be accomplished through a deep learning approach employing methodologies inspired by Convolutional Neural Networks (CNN). This results in a segmented object for each of the detected objects. Using the Mask-R-CNN architecture, the determination of the detection model for the collapsed trestle is then one of them. Using various epoch and sample sizes per image also aids in locating the optimal parameters. Due to the 2% difference in accuracy and the shorter training time compared to the 1100 epoch, the study results could be chosen as an option at the 500 epoch with batch size per image 64. The model was able to detect eight objects on the Trestle ruins, achieving a predicted score rate of over 94%. Manual and automatic differential calculations yielded the lowest RMSE of 18.337m at 500 epoch in batch size per image 64.
Item Type: | Thesis (Other) |
---|---|
Uncontrolled Keywords: | Deep Learning, Mask R-CNN, Multibeam Echosounder, Reruntuhan trestle, Collapsed Trestle |
Subjects: | G Geography. Anthropology. Recreation > GC Oceanography Q Science > QA Mathematics > QA336 Artificial Intelligence R Medicine > R Medicine (General) > R858 Deep Learning |
Divisions: | Faculty of Civil, Planning, and Geo Engineering (CIVPLAN) > Geomatics Engineering > 29202-(S1) Undergraduate Thesis |
Depositing User: | Arya Fajar Kurniawan |
Date Deposited: | 24 Aug 2023 05:07 |
Last Modified: | 24 Aug 2023 05:07 |
URI: | http://repository.its.ac.id/id/eprint/103409 |
Actions (login required)
View Item |