EARLY DETECTION OF FAÇADE DAMAGE IN DUTCH COLONIAL BUILDINGS IN SEMARANG USING A YOLOV8-BASED DEEP LEARNING ALGORITHM

Abstract

Dutch colonial buildings in Semarang have high historical and architectural value; however, many of their façades have experienced material degradation in the form of hairline cracks, structural cracks, and plaster peeling due to the humid tropical climate and the age of the buildings. Conventional visual inspection methods tend to be subjective, time-consuming, and pose safety risks to surveyors.

This study aims to develop a vision-based assessment approach to automatically detect and identify façade cracks using the YOLO algorithm. The research method involved collecting 2,400 images of colonial building façades in the Kota Lama area of Semarang using a 24 MP resolution camera and a drone at a height of 5–15 meters. The dataset was manually annotated using a bounding box format and divided into training data (70%), validation data (20%), and testing data (10%). The model was trained using YOLOv8 with an input resolution of 640×640 pixels, 150 epochs, a batch size of 16, and AdamW optimization. Performance evaluation used the metrics Precision, Recall, and mAP@0.5.

The results show that the model achieved an mAP@0.5 score of 0.91, with a Precision of 0.88 and Recall of 0.86, and was capable of detecting hairline cracks larger than 2 mm in real time. This approach has proven effective as a digital-based conservation support system for historic buildings.