Estimasi Umur Ikan Berdasarkan Citra Digital Otolith Greenland Halibut (Reinhaardtius Hippoglossoides) Menggunakan Kombinasi Convolutional Neural Network dan Support Vector Regression

Bumi Damaika, Thival (2026) Estimasi Umur Ikan Berdasarkan Citra Digital Otolith Greenland Halibut (Reinhaardtius Hippoglossoides) Menggunakan Kombinasi Convolutional Neural Network dan Support Vector Regression. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003221195-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Penentuan umur ikan melalui interpretasi visual otolith memerlukan keahlian khusus dan berpotensi menghasilkan perbedaan penilaian antarpembaca. Penelitian ini bertujuan membangun dan mengevaluasi model estimasi umur Greenland Halibut (Reinhardtius hippoglossoides) berdasarkan citra digital otolith menggunakan kombinasi Convolutional Neural Network dan Support Vector Regression (SVR). Data penelitian terdiri atas 4.109 pasangan otolith atau 8.218 citra individual serta 657 citra otolith tunggal, sehingga diperoleh total 8.875 citra dengan rentang umur referensi 1–26 tahun. Citra diproses dalam format RGB berukuran 400 × 400 × 3 menggunakan EfficientNetB0 berbobot awal ImageNet dengan seluruh lapisan jaringan dasar dibekukan. Global Average Pooling digunakan untuk menghasilkan 1.280 fitur numerik dari setiap citra. Fitur dan target selanjutnya distandardisasi dan digunakan sebagai masukan SVR dengan kernel Radial Basis Function. Hiperparameter C, ε, dan γ dioptimalkan menggunakan 50 kandidat RandomizedSearchCV dengan GroupKFold tiga lipatan. Model EfficientNetB0 dengan kepala regresi digunakan sebagai pembanding. Evaluasi akhir dilakukan pada tingkat pasangan dengan merata-ratakan hasil prediksi otolith kiri dan kanan menggunakan Mean Squared Error (MSE) dan mean coefficient of variation (mean CV). Pada data uji yang terdiri atas 164 pasangan otolith, EfficientNetB0-SVR menghasilkan MSE sebesar 2,830674 tahun² dan mean CV sebesar 8,692491%. Sementara itu, model EfficientNetB0 menghasilkan MSE sebesar 3,285867 tahun² dan mean CV sebesar 10,265758%. Dengan demikian, EfficientNetB0-SVR menghasilkan MSE yang lebih rendah sebesar 0,455193 tahun² dan mean CV yang lebih rendah sebesar 1,573267 poin persentase. Hasil tersebut menunjukkan bahwa kombinasi fitur EfficientNetB0 dan SVR memberikan kinerja yang lebih baik daripada model EfficientNetB0 pada data uji penelitian ini serta mampu menghasilkan estimasi umur Greenland Halibut yang kompetitif berdasarkan citra digital otolith.
===================================================================================================================================
Determining fish age through visual interpretation of otoliths requires specialized expertise and may result in variations among readers. This study aims to develop and evaluate an age estimation model for Greenland Halibut (Reinhardtius hippoglossoides) based on digital otolith images using a combination of Convolutional Neural Network and Support Vector Regression (SVR). The dataset comprised 4,109 paired otolith samples, equivalent to 8,218 individual images, and 657 unpaired otolith images, resulting in a total of 8,875 images with reference ages ranging from 1 to 26 years. The images were processed in RGB format at a resolution of 400 × 400 × 3 using EfficientNetB0 with ImageNet-pretrained weights while all backbone layers were frozen. Global Average Pooling was applied to produce 1,280 numerical features from each image. The features and target values were subsequently standardized and used as inputs to an SVR model with a Radial Basis Function kernel. The C, ε, and γ hyperparameters were optimized using 50 RandomizedSearchCV candidates with three-fold GroupKFold cross-validation. An EfficientNetB0 model with a regression head was used as the benchmark. The final evaluation was performed at the paired-sample level by averaging the predictions from the left and right otolith images using Mean Squared Error (MSE) and mean coefficient of variation (mean CV). On the test dataset comprising 164 otolith pairs, EfficientNetB0-SVR achieved an MSE of 2.830674 years² and a mean CV of 8.692491%. In comparison, the EfficientNetB0 model produced an MSE of 3.285867 years² and a mean CV of 10.265758%. Thus, EfficientNetB0-SVR reduced the MSE by 0.455193 years² and the mean CV by 1.573267 percentage points. These results demonstrate that the combination of EfficientNetB0 features and SVR outperformed the EfficientNetB0 model on the test dataset and provided competitive estimates of Greenland Halibut age based on digital otolith images.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	EfficientNetB0, Greenland Halibut, otolith, RandomizedSearchCV, Support Vector Regression, EfficientNetB0, Greenland Halibut, otolith, RandomizedSearchCV, Support Vector Regression.
Subjects:	Q Science > QA Mathematics > QA278.2 Regression Analysis. Logistic regression Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) Q Science > QL Zoology > QL639.1 Fishes--Growth.
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Thival Bumi Damaika
Date Deposited:	05 Aug 2026 04:29
Last Modified:	05 Aug 2026 04:29
URI:	http://repository.its.ac.id/id/eprint/143980

Actions (login required)

View Item