Sistem Klasifikasi Korban Bencana Kategori Manusia Dengan Kondisi Tertutup Sebagian Menggunakan Metode Convolutional Neural Network (CNN)

Prayogi, Marceliananda (2024) Sistem Klasifikasi Korban Bencana Kategori Manusia Dengan Kondisi Tertutup Sebagian Menggunakan Metode Convolutional Neural Network (CNN). Diploma thesis, Institut Teknologi Sepuluh Nopember.

[thumbnail of 2040201021-Undergraduate_Thesis.pdf] Text
2040201021-Undergraduate_Thesis.pdf - Accepted Version
Restricted to Repository staff only until 1 October 2026.

Download (13MB) | Request a copy

Abstract

National Pingtung University of Science and Technology (NPUST) bekerja sama dengan Kementerian Pendidikan, Kebudayaan, Riset, dan Teknologi Republik Indonesia (Mendikbudristek) melalui program Indonesian International Student Mobility Award (IISMA) yang berfokus pada Artificial Intelligence and Mechatronics. Salah satu proyek dalam program ini adalah pembuatan UAV dengan teknologi camera vision, yang dikembangkan untuk mendeteksi korban bencana. Namun alat ini hanya mampu mendeteksi manusia secara utuh dan kesulitan untuk mendeteksi korban bencana yang tertutup sebagian. Oleh karena itu, diperlukan pengembangan pada sistem deteksi agar dapat mengklasifikasikan korban menjadi beberapa bagian tubuh seperti head, upper limbs, lower limbs, half body, dan full body. Terlepas dari potensi keterbatasan ini, beberapa upaya dan metode telah dilakukan hingga menemukan keberhasilan, salah satunya dengan menggunakan metode Convolutional Neural Network (CNN). Sistem ini memiliki fitur utama yang berfungsi mendeteksi objek dan mengklasifikasikan korban ke dalam jenis bagian tubuh untuk memudahkan user mengidentifikasi korban. Data gambar yang dihasilkan oleh kamera diproses menggunakan metode CNN untuk mengenali pola dan fitur kompleks dari gambar. CNN adalah vondasi You Only Look Once (YOLO), dimana YOLO memanfaatkan CNN untuk ekstraksi fitur, dengan membagi gambar menjadi grid, di mana setiap sel memprediksi bounding box dan kelas objek sekaligus, memungkinkan deteksi cepat dalam satu proses, ideal untuk aplikasi real-time. Dengan demikian, CNN mendasari kemampuan YOLO untuk mendeteksi dan mengklasifikasikan objek dengan akurasi dan kecepatan tinggi. Setelah data diproses, hasilnya ditampilkan melalui website yang dibangun menggunakan framework Streamlit. Secara keseluruhan, performa model menunjukkan akurasi 95,60%, presisi 90,40%, recall 82%, F1 score 85,90%, dan specificity 98,30%. Dengan performa ini, model menunjukkan potensi yang baik dalam aplikasi nyata, terutama dalam skenario yang membutuhkan deteksi bagian tubuh manusia secara akurat.
=================================================================================================================================
National Pingtung University of Science and Technology (NPUST) collaborated with the Ministry of Education, Culture, Research, and Technology of the Republic of Indonesia (Mendikbudristek) through the Indonesian International Student Mobility Award (IISMA) program, focusing on Artificial Intelligence and Mechatronics. One of the projects in this program was the development of UAVs with camera vision technology, designed to detect disaster victims. However, this tool was only able to detect fully visible humans and struggled to identify partially covered disaster victims. Therefore, further development of the detection system is required to classify victims into several body parts such as head, upper limbs, lower limbs, half body, and full body. Despite these potential limitations, several efforts and methods have been made, leading to success, one of which is using the Convolutional Neural Network (CNN) method. This system has a key feature that detects objects and classifies victims into body part types to assist users in identifying victims. The image data captured by the camera is processed using the CNN method to recognize patterns and complex features in the image. CNN is the foundation of You Only Look Once (YOLO), where YOLO leverages CNN for feature extraction, dividing the image into grids, with each cell predicting bounding boxes and object classes simultaneously, enabling fast detection in a single process, ideal for real-time applications. Thus, CNN underpins YOLO's ability to detect and classify objects with high accuracy and speed. After the data is processed, the results are displayed through a website built using the Streamlit framework. Overall, the model's performance demonstrated an accuracy of 95.60%, precision of 90.40%, recall of 82%, F1 score of 85.90%, and specificity of 98.30%. With this performance, the model shows great potential for real-world applications, especially in scenarios requiring accurate human body part detection.

Item Type: Thesis (Diploma)
Uncontrolled Keywords: Object Detection System, Natural Disaster Victims, Victim Search, Convolutional Neural Network (CNN), Deep Learning, Korban Bencana Alam, Pencarian Korban, Sistem Objek Deteksi, YOLOv5
Subjects: T Technology > T Technology (General)
T Technology > TJ Mechanical engineering and machinery
T Technology > TJ Mechanical engineering and machinery > TJ211 Robotics.
T Technology > TJ Mechanical engineering and machinery > TJ211.4 Robot motion
T Technology > TJ Mechanical engineering and machinery > TJ211.415 Mobile robots
T Technology > TJ Mechanical engineering and machinery > TJ230 Machine design
Divisions: Faculty of Vocational > 36304-Automation Electronic Engineering
Depositing User: Marceliananda Prayogi
Date Deposited: 03 Sep 2024 02:09
Last Modified: 03 Sep 2024 02:09
URI: http://repository.its.ac.id/id/eprint/115588

Actions (login required)

View Item View Item