Módulo de reconhecimento de imagens anômalas baseado em microsserviços

Abstract

This work presents the development of an anomalous image recognition module in the context of civil construction based on microservices architecture. The KNN and OCSVM classifiers are used together with the local descriptors HOG, SIFT, SURF and ORB in different PCA dimensions. The dataset was built using Web Scrapping which includes images that belongs to construction and also from other anomalous classes. For microservices, the Spring Framework was used to build services in the frontend, backend and management levels. The best combination of descriptor, classifier and PCA originated a classification model to be deployed in a web service. The HOG local descriptor no use of PCA with the KNN classifier obtained the best performance, achieving 96.2 % metric precision rate when performing the anomalous image detection task, in which the resulting classification model was later embedded and made available for use in a web application.