B. G. Vijay Kumar

Babloo Kumar, Sayantari Ghosh

Due to cyclic loading and fatigue stress cracks are generated, which affect the safety of any civil infrastructure. Nowadays machine vision is being used to assist us for appropriate maintenance, monitoring and inspection of concrete structures by partial replacement of human-conducted onsite inspections. The current study proposes a crack detection method based on deep convolutional neural network (CNN) for detection of concrete cracks without explicitly calculating the defect features. In the course of the study, a database of 3200 labelled images with concrete cracks has been created, where the contrast, lighting conditions, orientations and severity of the cracks were extremely variable. In this paper, starting from a deep CNN trained with these images of 256 x 256 pixel-resolution, we have gradually optimized the model by identifying the difficulties. Using an augmented dataset, which takes into account the variations and degradations compatible to drone videos, like, random zooming, rotation and intensity scaling and exhaustive ablation studies, we have designed a dual-channel deep CNN which shows high accuracy (~ 92.25%) as well as robustness in finding concrete cracks in realis-tic situations. The model has been tested on the basis of performance and analyzed with the help of feature maps, which establishes the importance of the dual-channel structure.

Rafael Felix, B. G. Vijay Kumar, Ian Reid et al.

In generalized zero shot learning (GZSL), the set of classes are split into seen and unseen classes, where training relies on the semantic features of the seen and unseen classes and the visual representations of only the seen classes, while testing uses the visual representations of the seen and unseen classes. Current methods address GZSL by learning a transformation from the visual to the semantic space, exploring the assumption that the distribution of classes in the semantic and visual spaces is relatively similar. Such methods tend to transform unseen testing visual representations into one of the seen classes' semantic features instead of the semantic features of the correct unseen class, resulting in low accuracy GZSL classification. Recently, generative adversarial networks (GAN) have been explored to synthesize visual representations of the unseen classes from their semantic features - the synthesized representations of the seen and unseen classes are then used to train the GZSL classifier. This approach has been shown to boost GZSL classification accuracy, however, there is no guarantee that synthetic visual representations can generate back their semantic feature in a multi-modal cycle-consistent manner. This constraint can result in synthetic visual representations that do not represent well their semantic features. In this paper, we propose the use of such constraint based on a new regularization for the GAN training that forces the generated visual features to reconstruct their original semantic features. Once our model is trained with this multi-modal cycle-consistent semantic compatibility, we can then synthesize more representative visual representations for the seen and, more importantly, for the unseen classes. Our proposed approach shows the best GZSL classification results in the field in several publicly available datasets.