Perancangan Sistem Klasifikasi Hasil Tangkapan Nelayan Berbasis Citra Digital Dan Convolutional Neural Network

Yuana, Nadya Eka (2026) Perancangan Sistem Klasifikasi Hasil Tangkapan Nelayan Berbasis Citra Digital Dan Convolutional Neural Network. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5019221032_Nadya Eka Yuana.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Penelitian ini bertujuan menganalisis penerapan pengolahan citra berbasis ruang warna Hue, Saturation, Value (HSV) serta merancang sistem klasifikasi jenis ikan hasil tangkapan nelayan secara otomatis menggunakan Convolutional Neural Network (CNN). Pengolahan citra dilakukan melalui tahapan center crop untuk memusatkan objek ikan dan koreksi warna HSV untuk memperbaiki distribusi warna serta pencahayaan sebelum proses klasifikasi. Pengujian dilakukan pada dua skenario, yaitu menggunakan dataset ikan non-dilindungi dan dataset gabungan yang terdiri atas ikan non-dilindungi serta ikan dilindungi. Hasil pengujian pada dataset ikan non-dilindungi menunjukkan rata-rata akurasi sebesar 92,01%, yang menunjukkan bahwa CNN mampu mengenali karakteristik visual setiap spesies, seperti bentuk tubuh, tekstur, pola warna, dan ciri morfologi, dengan baik. Pada pengujian dataset gabungan, kelas ikan dilindungi Napoleon Wrasse dan Pari Manta memperoleh akurasi masing-masing sebesar 97,70% dan 91,57%, namun terjadi penurunan akurasi pada beberapa kelas ikan non-dilindungi. Penurunan tersebut dipengaruhi oleh keterbatasan jumlah data pelatihan akibat proses balancing dataset, variasi kualitas citra, serta kompleksitas latar belakang bawah laut yang menyebabkan proses ekstraksi fitur menjadi kurang optimal. Penerapan pengolahan citra berbasis HSV mampu meningkatkan akurasi pada beberapa kelas ikan, tetapi belum memberikan peningkatan performa secara konsisten karena perubahan representasi warna objek serta masih adanya informasi latar belakang yang ikut dipelajari oleh CNN. Secara keseluruhan, hasil penelitian menunjukkan bahwa kualitas dan keseragaman dataset merupakan faktor utama yang memengaruhi performa sistem klasifikasi berbasis CNN. Oleh karena itu, peningkatan kualitas dataset, penambahan jumlah data pelatihan, serta penerapan metode preprocessing yang lebih adaptif diperlukan untuk meningkatkan akurasi dan keandalan sistem.
===================================================================================================================================
This study aims to analyze the application of Image processing based on the Hue, Saturation, Value (HSV) color space and to develop an automatic fish catch classification system using a Convolutional Neural Network (CNN). The image preprocessing stage consists of center cropping to focus on the fish object and HSV color correction to improve color distribution and illumination before the classification process. The system was evaluated using two scenarios: a dataset containing non-protected fish species and a combined dataset consisting of both non-protected and protected fish species. The experimental results on the non-protected fish dataset achieved an average classification accuracy of 92.01%, indicating that the CNN was able to effectively recognize visual characteristics such as body shape, texture, color patterns, and morphological features. In the combined dataset, the protected fish species Napoleon Wrasse and Manta Ray achieved classification accuracies of 97.70% and 91.57%, respectively. However, the addition of protected fish classes reduced the classification accuracy of several non-protected fish species. This performance degradation was mainly caused by the limited number of training images after dataset balancing, variations in image quality, and the complexity of underwater backgrounds, which affected the feature extraction process. The HSV-based image preprocessing improved classification accuracy for several fish classes but did not consistently enhance the overall system performance due to changes in the original color representation of the fish and the remaining background information learned by the CNN. Overall, the results indicate that dataset quality and consistency are the primary factors influencing the performance of CNN-based fish classification systems. Therefore, improving dataset quality, increasing the number of training images, and implementing more adaptive preprocessing techniques are expected to enhance the accuracy and reliability of the proposed system.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Convolutional Neural Network, klasifikasi ikan, citra digital, Image processing, HSV Convolutional Neural Network, fish classification, digital image, Image processing, HSV.
Subjects:	T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Marine Technology (MARTECH) > Ocean Engineering > 38201-(S1) Undergraduate Thesis
Depositing User:	Nadya Eka Yuana
Date Deposited:	04 Aug 2026 04:25
Last Modified:	04 Aug 2026 04:25
URI:	http://repository.its.ac.id/id/eprint/143038

Actions (login required)

View Item