<![CDATA[This study develops a web-based School Management Information System (SIMS) equipped with a deep learning Optical Character Recognition (OCR) module that extracts data from tuition payment receipts at SMK BIT Bina Aulia, Bogor. The system aims to accelerate transaction verification, reduce manual input errors, and improve administrative transparency. The Research and Development method was applied, with the Waterfall model used to construct the product. The OCR module was built on PaddleOCR PP-OCRv4 with DBNet text detection and SVTR_LCNet text recognition using a CTC decoder, fine tuned on 384 receipt images collected from 14 payment channels and augmented into 14,824 training crops. The best training checkpoint reached 79.04% exact match accuracy with a normalized edit distance of 0.9563 at epoch 90. Evaluated on 940 text crops, the deployed service achieved 94.79% character accuracy, a 5.21% Character Error Rate, a 25.67% Word Error Rate, and 76.60% exact match accuracy, rising to 86.49% when spacing differences are ignored. Fine tuning improved exact match accuracy by 4.05 percentage points, and the proposed model outperformed Tesseract OCR 5 and EasyOCR on every metric. Black box testing of 63 test items and white box basis path testing of 53 independent paths passed without failure.]]>
Copyrights © 2026
