Percorrer por autor "RIBEIRO, RAFAEL ALMEIDA"
A mostrar 1 - 1 de 1
Resultados por página
Opções de ordenação
- Development of a mass drug administration monitoring system for a Humanitarian NGOPublication . RIBEIRO, RAFAEL ALMEIDA; Sousa, Paulo Manuel Baltarejo deMass drug administration is the process of distributing medication to people who are in areas with risk of neglected tropical diseases. The monitoring of this process is supported by the application of forms in the communities, which are then sent to the headquarters, where they are manually entered into governmental health information systems. This dissertation aims at finding a solution for automating the process, making it more efficient and less error-prone. Optical Character Recognition (OCR) is a technology that converts different types of documents into searchable data. By reviewing the literature, it is possible to conclude that the application of OCR is feasible for accurate text extraction. The literature also suggests that AWS Textract is the tool with higher accuracy when extracting handwritten text. A system composed of a Backend, an Android Application and a Backoffice web application was designed and implemented. This solution was evaluated with two experiments. Text extraction tests were performed using 8 example test forms filled with fake data. Different effectiveness metrics where calculated, resulting in mean values of 0.8%, 2.5% and 3.7% for the Character Error Rate (CER), Word Error Rate (WER) and Field Error Rate (FER), while a mean value of 96.2% was achieved for the Precision, Recall and F1 score. It was possible to conclude that forms filled in English and forms filled with print handwriting style had better accuracy than forms filled in Portuguese and forms filled with cursive handwriting style, respectively. The other experiment performed was user testing, in which two testers used official forms filled with fake data to test the application. The feedback was positive, with some improvements being suggested, such as the order of the fields in the form response details screen and the size of the numeric fields. A mean value of 7.9% incorrectly extracted fields was achieved during the user testing.