The ageing infrastructure in ports requires regular inspection. This inspection is currently carried out manually by sensing the entire infrastructure by hand. Such a process is costly as it requires a lot of time and manpower. To overcome these difficulties, we propose to map the harbour structure above and below water with a multi-sensor system and try to automate the classification process in terms of common damage types using deep learning approaches. In the images taken \rev{above} water, damaged and undamaged zones are localised using a semantic segmentation approach. We make use of a real data set captured at JadeWeserPort Wilhemlshaven to test our approach. The images are divided into smaller sections of 512x512 pixels and these are propagated through the DeepLabv3+ architecture, a modern convolutional neural network for semantic segmentation tasks, which is trained in particular to detect corrosion or rust anomalies. We achieve with a pre-trained ResNet-50 backbone and fully supervised data set IoU scores of 96.0 % and 55.9 % for undamaged and damaged zones as well as F1-scores of 98.0 % and 71.7 %. We show that our approach can achieve a fully automated and reproducible image segmentation and damage detection which can analyse the whole structure instead of the sample-wise manual method.
|