Penerapan Model Deep Learning pada Pembuatan Aplikasi CACA (Cari Cafe)

Pusriwijayanti, Salma (2023) Penerapan Model Deep Learning pada Pembuatan Aplikasi CACA (Cari Cafe). Undergraduate Thesis thesis, Institut Teknologi Telkom Purwokerto.

	Text Cover.pdf Download (426kB)
	Text Abstract.pdf Download (52kB)
	Text Abstrak.pdf Download (53kB)
	Text BAB I.pdf Download (70kB)
	Text BAB II.pdf Download (469kB)
	Text BAB III.pdf Download (144kB)
	Text BAB IV.pdf Restricted to Registered users only Download (2MB)
	Text BAB V.pdf Download (119kB)
	Text Daftar Pustaka.pdf Download (140kB)
	Text Lampiran.pdf Restricted to Registered users only Download (968kB)

Abstract

The increase in coffee production encourages the growth of cafes and creates confusion for consumers when choosing a cafe that suits their preferences. Therefore, cafe recommendations are important. However, prior to use, each application requires registration, which often takes time because it involves the stages of typing in identity information, uploading scanned KTPs, and waiting for data verification. This process is time-consuming. Previous research regarding the implementation of deep learning models for facial recognition systems made it possible to form a facial image recognition model on a computer. This research focuses on deep learning models for making café search applications. The deep learning model includes reading e-KTP images automatically, predicting the identity of the user's selfie photo, and verifying the e-KTP photo with the user's selfie photo. In achieving these three functions, artificial intelligence technologies such as optical character recognition (OCR), convolutional neural networks (CNN), and siamese neural networks (SNN) are used. This study succeeded in implementing OCR on e-KTP images with bounding boxes, where the box_loss value was 0.05211 and the cls_loss value was 0.01598. The use of the VGG16 transfer learning model with the sigmoid activation function to predict identity has also achieved optimal success. In addition, verification using SNN also gives good results, achieving an accuracy of 0.9285 with a loss value of 0.0170.

Actions (login required)

View Item