ABSTRAK
Ikan merupakan bahan pangan bergizi tinggi yang bersifat mudah rusak (highly perishable) akibat aktivitas enzimatik, pertumbuhan mikroorganisme, dan oksidasi lemak setelah kematian (post-mortem). Penentuan tingkat kesegaran ikan di pasar tradisional masih dilakukan secara manual melalui pengamatan visual pada bagian mata dan insang, sehingga hasil penilaian cenderung subjektif dan kurang konsisten. Penelitian ini bertujuan membangun model klasifikasi tingkat kesegaran ikan nila, ikan lele, dan ikan patin berdasarkan citra RGB pada area mata dan insang menggunakan metode Convolutional Neural Network (CNN). Dataset terdiri dari 1.800 citra .jpg dari tiga jenis ikan yang dikelompokkan menjadi tiga kategori berdasarkan waktu simpan pada suhu ruang, yaitu segar (1–5 jam), kurang segar (6–10 jam), dan busuk (11–15 jam). Tahapan penelitian meliputi Preprocessing (cropping, Resize 256×256 piksel, normalisasi RGB), pelabelan data, pembagian dataset dengan rasio 80:10:10, pelatihan model CNN selama 50 epoch, serta evaluasi menggunakan Confusion Matrix, Accuracy, Precision, Recall, dan F1-Score. Hasil pengujian menunjukkan model mencapai Test Accuracy sebesar 78,89% dengan rata-rata F1-Score sebesar 0,79, dengan performa terbaik pada kelas busuk dan performa terendah pada kelas kurang segar akibat kemiripan visual dengan kelas segar. Penelitian ini diharapkan menghasilkan sistem klasifikasi kesegaran ikan yang akurat, objektif, cepat, dan konsisten untuk mendukung penerapan kecerdasan buatan pada bidang keamanan pangan, khususnya sektor perikanan.
Kata Kunci: Kesegaran Ikan, Citra RGB, Convolutional Neural Network, Mata dan Insang, Waktu Simpan

ABSTRACT
Fish is a highly nutritious food that is highly perishable due to enzymatic activity, microbial growth, and lipid oxidation after death (post-mortem). Freshness determination in traditional markets is still carried out manually through visual observation of the eyes and gills, making the assessment subjective and inconsistent. This study aims to build a classification model for the freshness level of tilapia, catfish, and pangasius based on RGB images of the eye and gill areas using the Convolutional Neural Network (CNN) method. The dataset consists of 1,800 .jpg images from three fish species, grouped into three categories based on room-temperature storage time: fresh (1–5 hours), less fresh (6–10 hours), and rotten (11–15 hours). The research stages include preprocessing (cropping, resizing to 256×256 pixels, RGB normalization), data labeling, dataset splitting at an 80:10:10 ratio, CNN model Training for 50 epochs, and evaluation using a Confusion Matrix along with Accuracy, Precision, Recall, and F1-Score metrics. Testing results show that the model achieved a Test Accuracy of 78.89% with an average F1-Score of 0.79, with the best performance in the rotten class and the lowest performance in the less-fresh class due to visual similarity with the fresh class. This study is expected to produce an accurate, objective, fast, and consistent fish freshness classification system to support the application of artificial intelligence in food safety, particularly in the fisheries sector.
Keywords: Fish Freshness, RGB Image, Convolutional Neural Network, Eye and Gill, Storage Time