Object Augmentation for Automated Multi-damages Construction Detection using Lightweight Deep Learning

Abstract

Computer Vision (CV) -based construction damages has been widely developed and resulted in potential alternative over traditional visual inspection. Howerver, the current deep learning models require an enormous data size, and relatively computational-expensive. Therefore, collecting and create more data has been an crucial state in the process of excecution, especially in the context of multi-damages detection with surround different types of object images. Therefore, an object aumentation and lightweight neural network has been introduced with the aim to improve the computional perfomance and address limitation of shortage and imbalance data