Electronic Theses and Dissertation

Penilaian kualitas bahan pangan, khususnya daging sapi, merupakan aspek krusial dalam sektor pertanian dan industri pangan. Metode tradisional yang mengandalkan untuk menentukan kualitas daging sapi cenderung bersifat subjektif dan rentan terhadap ketidakakuratan dalam membedakan antara daging segar dan busuk. Studi ini mengusulkan sistem klasifikasi otomatis berbasis deep learning untuk mengatasi keterbatasan metode konvensional tersebut. Dataset penelitian terdiri dari 3.696 citra dengan berbagai resolusi gambar daging sapi yang diambil menggunakan kamera smartphone Realme C3 pada saat akuisisi citra. Pengambilan citra dilakukan pada suhu ruang 28-30°C dengan interval satu jam selama 16 jam untuk menangkap degradasi kualitas daging dari segar hingga busuk. Pelabelan citra dilakukan berdasarkan kriteria visual perubahan warna dan tekstur serta perubahan bau; daging diklasifikasikan sebagai segar jika diambil sebelum 6 jam dengan ciri warna merah cerah dan tekstur normal, sedangkan daging diklasifikasikan sebagai busuk diambil setelah 10 jam dengan ciri warna coklat gelap, tekstur lengket, dan bau busuk. Selanjutnya dataset dibagi menjadi 1.120 citra untuk masing-masing kategori segar dan busuk setelah dialakuakan pra-pemrosesan data gambar. Tiga arsitektur Convolutional Neural Network (CNN) dengan pendekatan transfer learning VGG16, ResNet50-V2, dan Inception-V3 digunakan untuk klasifikasi model. Hasil eksperimen menunjukkan bahwa VGG16 memberikan performa terbaik, mencapai akurasi 99,46%, presisi 99,29%, recall 99,64%, dan F1-score 99,47. Kemudian model diimplementasikan kedalam aplikasi berbasis web, sehingga dapat dirasakan langsung oleh masyarakat.

The assessment of food material quality, particularly beef, is a crucial aspect in the agricultural and food industry sectors. Traditional methods that rely on subjective judgments to determine the quality of beef tend to be prone to inaccuracy in distinguishing between fresh and rotten meat. This study proposes an automatic classification system based on deep learning to overcome the limitations of conventional methods. The research dataset consists of 3,696 images with various resolutions of beef, captured using a Realme C3 smartphone camera during image acquisition. Image capture was conducted at room temperature of 28-30°C at one-hour intervals over 16 hours to capture the degradation in meat quality from fresh to spoiled. Image labeling was performed based on visual criteria of color and texture changes as well as odor; meat was classified as fresh if taken before 6 hours with characteristics of bright red color and normal texture, while meat was classified as spoiled if taken after 10 hours with characteristics of dark brown color, sticky texture, and foul odor. The dataset was then divided into 1,120 images for each category, fresh and rotten, after undergoing pre-processing of the image data. Three Convolutional Neural Network (CNN) architectures using transfer learning approaches VGG16, ResNet50-V2, and Inception-V3 were utilized for model classification. Experimental results demonstrated that VGG16 provided the best performance, achieving an accuracy of 99.46%, precision of 99.29%, recall of 99.64%, and an F1-score of 99.47%.. The model was subsequently implemented into a web-based application to enable direct utilization by the public.