Flat roof classification and leaks detections by Deep Learning

Abstract

This paper presents an efficient and accurate method for detecting flat roof leaks using a combination of unmanned aerial vehicles (UAVs) and deep learning. The proposed method utilizes a DJI M300 drone equipped with RGB and thermal cameras to capture high-resolution images of the roof. These images are then processed to create orthomosaics and digital elevation models (DEMs). A deep learning model based on the U-NET architecture is then used to segment the roof into different classes, such as PVC foil, windows, and sidewalks. Finally, the damaged insulation is identified by analyzing the temperature distribution within the PVC foil segments. The proposed method has several advantages over traditional inspection methods. It is much faster and more efficient. A UAV can collect images of a large roof in a matter of minutes, while traditional methods can take several days or weeks. The orthomosaics and temperature maps generated by the UAV are much more detailed than the images that can be collected by a human inspector. Third, the UAV-based system is safer. The UAV can collect images of the roof without the need for a human inspector to climb onto the roof, which can be dangerous. The results of this study show that the proposed method is an effective and accurate way to detect flat roof leaks. The deep learning model was able to achieve an overall accuracy of 95% in segmenting the roof into different classes. The method was also able to identify damaged insulation with a high degree of accuracy.