This paper focuses on automatic identification of defective roofs and window casings using anomaly detection. A dataset of drone images from the Netherlands is utilized. The PatchCore anomaly detection model as one of the state of the art model was used in this study. We utilized different backbone networks, retrained them with same dataset for classification and segmentation task and different feature extraction layers. WideResNet50 with layers 2 and 3 performs best on the Casings dataset, while the model’s performance on the Roof dataset is unsatisfactory. The tiled roof dataset shows improved results and Retrained ResNet50 backbone was the best configuration. The study approved influence of using different setting for the model.
