Electronic Theses and Dissertation

Penelitian ini membahas tentang klasifikasi hasil budidaya udang putih (Litopenaeus vannamei) pada media kolam tanah dan kolam terpal menggunakan deep learning. Udang putih atau sering juga disebut para nelayan udang vaname merupakan salah satu komoditas perikanan yang memiliki nilai ekonomi tinggi. Hasil budidaya udang vaname pada media kolam tanah dan kolam terpal sekilas terlihat hampir sama, tapi jika diperhatikan berbeda dari berbagai segi baik itu dari warna, bentuk, ukuran, dan harga pasarannya. Hal ini menyebabkan para pedagang dapat mempermainkan harga bagi pembeli yang tidak begitu paham tentang perbedaan kedua udang. Alat otomatis untuk klasifikasi kedua jenis udang ini perlu ada, dan dalam pengembangan alat tersebut diperlukan dataset. Namun saat ini belum ada public dataset dan penelitian mengenai klasifikasi hasil budidaya udang vaname dari media kolam tanah dan kolam terpal. Tujuan penelitian ini membangun dataset primer hasil budidaya udang vaname dari media kolam tanah dan kolam terpal dan mengklasifikasi menggunakan arsitektur Visual Geometry Group (VGG)-16 dan Residual Network (ResNet)-50. Proses pembangunan dataset yaitu dengan cara akuisi data dan augmentasi, jumlah dataset setelah melakukan augmentasi yaitu 2.080 citra pada masing-masing kelas. Performasi dari pelatihan diukur berdasarkan waktu pelatihan, accuracy, Positive Predictive Value (PPV)/precision, Negative Predictive Value (NPV), recall (sensitivity), dan specificity. Hasil budidaya udang vaname dari media kolam tanah dan kolam terpal dapat diklasifikasikan menggunakan arsitektur VGG-16 dan ResNet-50, tercatat waktu pelatihan ResNet-50 177 menit, sedangkan VGG-16 185 menit, dengan nilai accuracy testing arsitektur VGG-16 98.8% dan ResNet-50 100%.
Kata Kunci: Udang Vaname, Deep Learning, Convolutional Neural Network (CNN), VGG-16, ResNet-50.

This study discusses the classification of the results of white shrimp (Litopenaeus vannamei) cultivation in earthen ponds and tarpaulin ponds using deep learning media. White shrimp or often also called vannamei shrimp fishermen is a fishery commodity that has high economic value. At first glance, the results of vannamei shrimp farming in soil ponds and tarpaulin ponds look almost the same, but if you pay attention, they are different from various aspects, both in terms of color, shape, size, and market price. This causes traders to play with prices for buyers who do not really understand the differences between the two shrimp. Automatic tools for the classification of these two types of shrimp need to exist, and the development of these tools requires datasets. However, currently there is no public dataset and research regarding the classification of the results of vannamei shrimp farming from soil ponds and tarpaulin ponds. The purpose of this study was to build a primary dataset of vannamei shrimp farming from soil ponds and tarpaulin media and classify it using the Visual Geometry Group (VGG)-16 and Residual Network (ResNet)-50 architectures. The dataset development process is by means of data acquisition and augmentation, the number of datasets after augmentation is 2,080 images in each class. Training performance is measured based on training time, accuracy, Positive Predictive Value (PPV)/precision, Negative Predictive Value (NPV), recall (sensitivity), and specificity. The results of vannamei shrimp cultivation from soil media and tarpaulin ponds can be classified using the VGG-16 and ResNet-50 architectures, the ResNet-50 training time is 177 minutes, while the VGG-16 is 185 minutes, with an accuracy value of the VGG-16 architecture testing of 98.8% and ResNet-50 100%. Keywords: Vaname Shrimp, Deep Learning, Convolutional Neural Network (CNN), VGG-16, ResNet-50.