Please use this identifier to cite or link to this item: http://repository.ipb.ac.id/handle/123456789/160315
Title: Perbandingan Model Instance Segmentation, Oriented Bounding Boxes, dan Object Detection dalam Pengukuran Panjang Ikan Kerapu Hidup dalam Keramba Jaring Apung
Other Titles: Comparison of Instance Segmentation, Oriented Bounding Boxes, and Object Detection Models in Length Measurement of Live Grouper at Floating Net Cages
Authors: Jaya, Indra
Iqbal, Muhammad
Yan, Cepy Septy
Issue Date: 2024
Publisher: IPB University
Abstract: Pengembangan teknologi tepat guna untuk membantu kemajuan komoditas perikanan Indonesia bernilai ekonomis sangat dibutuhkan khususnya pada sektor budidaya komoditas ikan kerapu. Penelitian ini mencoba untuk melakukan pengukuran panjang ikan kerapu hidup, dengan mengaplikasikan algoritma kecerdasan buatan (deep learning) dari YOLOv8 (Instance Segmentation, Oriented Bounding Boxes, dan Object Detection) di dalam kolam KJA yang diperoleh dari hasil perekaman kamera stereo UTS (Underwater Televisual System). Data gambar ikan kerapu hidup yang dikumpulkan berjumlah 400 gambar kemudian dilakukan de-haze, (preprocessing), dan diberi label. Pengukuran panjang ikan kerapu hidup menggunakan algoritma YOLOv8 dan diukur menggunakan Photoshop. Hasil pengukuran menunjukkan bahwa model Instance Segmentation yang memiliki nilai rata-rata sebesar 218,3 px lebih mendekati hasil nilai rata-rata panjang total (PT) ikan kerapu hidup dari penglihatan sebenarnya yang dilakukan oleh manusia dengan nilai rata-rata sebesar 212,2 px. Nilai koefisien determinasi (R Square) model Instance Segmentation 0,9744 (97,44%), Oriented Bounding Boxes 0,9481 (94,81%), dan Object Detection 0,9512 (95,12%). Dengan demikian dapat disimpulkan, model YOLOv8 Instance Segmentation memiliki nilai keakuratan yang lebih baik dalam melakukan pengukuran panjang yang mencakup ukuran panjang total (PT) ikan kerapu hidup.
The development of appropriate technology to help advance Indonesia's fisheries commodities with economic value is needed, especially in the grouper aquaculture sector. This research attempts to take length measurements of live grouper fish, by applying the artificial intelligence (deep learning) algorithm of YOLOv8 (Instance Segmentation, Oriented Bounding Boxes, and Object Detection) in KJA ponds obtained from UTS (Underwater Televisual System) stereo camera recordings. The collected live grouper image data amounted to 400 images and were then subjected to de-haze, (preprocessing), and labelled. Length measurement of live grouper using YOLOv8 algorithm and measured using Photoshop. The measurement results show that the Instance Segmentation model, which has an average value of 218.3 px, is closer to the results of the average value of the total length of live grouper fish than the actual vision performed by humans with an average value of 212.2 px. The coefficient of determination (R Square) of the Instance Segmentation model is 0.9744 (97.44%), Oriented Bounding Boxes 0.9481 (94.81%), and Object Detection 0.9512 (95.12%). Thus it can be concluded, the YOLOv8 Instance Segmentation model has a better accuracy value in performing length measurement which includes the total length of live grouper fish.
URI: http://repository.ipb.ac.id/handle/123456789/160315
Appears in Collections:UT - Marine Science And Technology

Files in This Item:
File Description SizeFormat 
cover_C54180029_e7ffd3a02c154595b4c5d3b48d572f51.pdfCover320.48 kBAdobe PDFView/Open
fulltext_C54180029_11864317e04b435ebd07ebbfda4cde5e.pdf
  Restricted Access
Fulltext1.74 MBAdobe PDFView/Open
lampiran_C54180029_527fbdae681547a9986c8ce1ee836646.pdf
  Restricted Access
Lampiran3.94 MBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.