Façade defects classification from imbalanced dataset using meta learning‐based convolutional neural network

Abstract

Façade inspection is a regular but necessary maintenance task to ensure the safety, functioning, and aesthetics of a building. Traditional visual identification of façade defects is dangerous, time-consuming, and insufficient. Based on an image dataset and deep learning algorithms, an automatic façade defects classification technique is developed in this research. A layer-based categorization rule is proposed to categorize façade defects. To handle the problem of imbalanced data size among defect classes, a meta learning-based method is applied, which reassigns weights to the training data. Experiments demonstrated that the proposed method had a stronger capacity to deal with the imbalanced dataset problem comparing with previous methods by improving the classification accuracy from 71.43% of a basic convolutional neural network (CNN) model to 82.86% of a meta learning-based CNN model.

MultiDefectNet: Multi-Class Defect Detection of Building Façade Based on Deep Convolutional Neural Network

“…This optimizer has a feature that reflects the direction of the previous weight update when the weights are updated. The weight update formula of SGD with momentum is expressed in the form of Equation (7), where β, m, and ε refer to the momentum, momentum vector, and learning rate, respectively. In this study, the minibatch size for weight updates was set to eight, and β = 0.9 and ε=0.0001 based on experiments.…”

SCNet: A Generalized Attention-based Model for Crack Fault Segmentation

“…Crack detection has been the mainstay of research on visual surveillance of faults in infrastructures, over last many decades. Its applicability has been researched upon in variety of application scenarios: tunnel safety inspection [61], [38], dam safety inspection [9], rail corridor inspection [52], steel bridges [22], [14], concrete bridges [16], mine safety [11], pavement crack detection [18], [25], building safety inspection [8], to name a few. The techniques for detection belong to two phases.…”

Semi‐supervised learning based on convolutional neural network and uncertainty filter for façade defects classification

“…First, to create a clean and standardized defects dataset, some research categorized the collected images according to an official design manual (Hüthwohl, Lu, & Brilakis, 2019), specific inspection operation code such as the pipeline assessment and certification program (PACP) (Yin et al., 2020), or customized categorization rules based on materials (Guo, Wang, Li, & Liu, 2020). Then, considering the imbalanced dataset problem caused by the variant occurring rates of different types of defects, researchers have proposed various algorithms such as oversampling (Mundt, Majumder, Murali, Panetsos, & Ramesh, 2019), weighted loss function (Meijer, Scholten, Clemens, & Knobbe, 2019), hierarchical classification (D. Li, Cong, & Guo, 2019) and meta learning (Guo et al., 2020). Moreover, some studies have explored the solution to small dataset problem through data augmentation (Cheng & Wang, 2018) and transfer learning (Gao & Mosalam, 2018; Liang, 2019; Silva & Lucena, 2018).…”

“…After cropping into image patches, the image patches were resized to pixel sizes of 224 224. Next, under the instruction of the categorization rules proposed in a previous research (Guo et al., 2020), the image patches were categorized into seven classes including six defect classes (crack, blistering, biological‐growth, spalling, delamination, and peeling, as shown in Figure 5) and no‐defects class. The dataset in this paper was developed based on the dataset used in a previous study (Guo et al., 2020).…”