PENERAPAN ALGORITMA CONVOLUTIONAL NEURAL NETWORK UNTUK PENGENALAN KARAKTER TULISAN TANGAN

SETO RAHARDYANTO, NIM. 17106050014 (2021) PENERAPAN ALGORITMA CONVOLUTIONAL NEURAL NETWORK UNTUK PENGENALAN KARAKTER TULISAN TANGAN. Skripsi thesis, UIN SUNAN KALIJAGA YOGYAKARTA.

Preview	Text (PENERAPAN ALGORITMA CONVOLUTIONAL NEURAL NETWORK UNTUK PENGENALAN KARAKTER TULISAN TANGAN) 17106050014_BAB-I_IV-atau-V_DAFTAR-PUSTAKA.pdf - Published Version Download (1MB) | Preview
	Text (PENERAPAN ALGORITMA CONVOLUTIONAL NEURAL NETWORK UNTUK PENGENALAN KARAKTER TULISAN TANGAN) 17106050014_BAB-II_sampai_SEBELUM-BAB-TERAKHIR.pdf - Published Version Restricted to Registered users only Download (4MB)

Abstract

This study aims to develop a Convolutional Neural Network (CNN) Model for the imposition of handwritten characters. Old documents that have important information are usually easily damaged because they are stored on paper media. The role of handwriting recognition is to convert the information contained in the document into a digital format that can be stored for a long time. This study uses 3 types of data, namely training data, validation data and test data. Training data and test data are obtained from the Kaggle public dataset. While the test data is made by the author himself. The training data and validation data are then used to develop the CNN model, and the test data is preprocessed to produce images that are ready to be recognized by the CNN model. The test results show that the best CNN Model Architecture for alphabetic handwritten character recognition is the Fourth Scenario CNN Architecture Model which obtained validation data accuracy of 96.47% and test data accuracy (alphabet) of 93.05%. The best CNN Model Architecture for recognizing alphabet and numeric handwritten characters is the Third Scenario CNN Architecture Model which obtained validation data accuracy of 94.70% and test data accuracy (alphabets and numbers) of 81.28%. A test data preprocessing algorithm was also developed which succeeded in producing 93.18% of the images recognized in the alphabetic test data image, and succeeded in producing 93.63% of the recognized images in the alphabetic and numeric test data images.

Share this knowledge with your friends :

Actions (login required)

View Item