Rivansyah Suhendra, Ratih Sari Ayu, Rara Syifa Qaisa, Ilham Juliwardi, Nica Astrianda, Puput Arisna, Alfis Syahril, Uswatun Hasanah

Penerapan CNN Arsitektur VGG16 untuk Deteksi Kesegaran Ikan Berdasarkan Citra Digital

Introduction

Penerapan cnn arsitektur vgg16 untuk deteksi kesegaran ikan berdasarkan citra digital. Deteksi kesegaran ikan otomatis menggunakan CNN arsitektur VGG16 berdasarkan citra mata digital. Model mencapai akurasi 100% untuk kontrol mutu perikanan.

Abstract

Kesegaran ikan merupakan indikator utama dalam menentukan kualitas dan keamanan produk perikanan. Penilaian secara manual masih bersifat subjektif dan memerlukan keahlian khusus. Penelitian ini bertujuan untuk mengembangkan sistem klasifikasi tingkat kesegaran ikan secara otomatis menggunakan algoritma Convolutional Neural Network (CNN) dengan arsitektur VGG16. Data berupa 1.378 citra mata ikan dikumpulkan dari pasar ikan di Meulaboh dan Blangpidie, kemudian melalui proses preprocessing menggunakan teknik contrast stretching. Dataset dibagi menjadi data latih (80%) dan data validasi (20%). Proses pelatihan dilakukan dengan menerapkan augmentasi dan normalisasi data guna meningkatkan kemampuan generalisasi model. Hasil pengujian menunjukkan bahwa model mampu mengklasifikasikan citra dengan akurasi, precision, recall, dan F1-score sebesar 100%. Analisis confusion matrix menunjukkan tidak adanya kesalahan klasifikasi pada data validasi. Temuan ini menunjukkan bahwa citra mata ikan merupakan fitur visual yang efektif untuk mengidentifikasi tingkat kesegaran. Sistem yang dikembangkan memiliki potensi untuk diimplementasikan dalam proses sortir dan kontrol mutu hasil perikanan. Penelitian selanjutnya disarankan untuk memperluas cakupan jenis ikan dan pengujian dalam kondisi lingkungan nyata guna meningkatkan robustitas model.

Review

This paper presents an automated system for classifying fish freshness using a Convolutional Neural Network (CNN) with VGG16 architecture, focusing specifically on digital images of fish eyes. The study addresses a relevant problem, as manual assessment of fish freshness is subjective and requires expertise. The methodology, involving contrast stretching, data augmentation, and normalization, is clearly outlined. The reported results are exceptionally high, with 100% accuracy, precision, recall, and F1-score on the validation set, suggesting the model's strong capability to differentiate freshness levels based on the collected eye images from Meulaboh and Blangpidie. This work demonstrates a promising application of deep learning in quality control for fishery products. While the reported 100% classification accuracy is impressive, it also warrants closer scrutiny. Such perfect scores in machine learning applications, especially with real-world image data, can sometimes indicate a limited diversity within the validation set, potential data leakage, or that the classes are exceptionally well-separated and perhaps too simplistic for the robustness required in practical scenarios. The dataset size of 1,378 images, while not trivial, might benefit from expansion, particularly across a wider range of freshness stages and environmental conditions, to truly challenge the model's generalization capabilities. Furthermore, relying solely on eye images, although effective, might not capture all facets of fish freshness, which typically involves other indicators like gills, skin, and smell. To enhance the scientific contribution and practical applicability of this research, several directions are recommended. Future work should prioritize rigorous external validation of the 100% accuracy claim with entirely new, unseen datasets from different species and diverse geographical locations. Expanding the feature set to include images of other fish body parts (e.g., gills, skin) or even multi-modal data could lead to a more comprehensive and robust freshness assessment system. Moreover, testing the model's performance under varied real-world lighting and background conditions, as well as exploring lighter CNN architectures for potential deployment on edge devices, would strengthen its practical utility.

Full Text

Comments

You need to be logged in to post a comment.