Deep Learning-Based Approach for Wall Surface Defect Detection

Wall surface defects are a common issue in the construction and maintenance sectors, typically resulting in compromised structural integrity, aesthetic issues and increased repair costs. Traditional manual inspection methods are time-consuming, labour-intensive and subject to human error; thus, there is a necessity for automation. This article introduces a deep learning-based system for the classification of six wall surface defects: cracks, paint flaking, holes, stains, chipping and defect-free surfaces. A well-prepared dataset was created, to which image pre-processing methods, such as Contrast Limited Adaptive Histogram Equalisation (CLAHE) and resizing, were applied to improve image quality. Data augmentation through rotation, brightness modification and adding noise was used to enhance resilience against natural variations. Three architectures of Convolutional Neural Network (CNN), namely VGG16, MobileNet and Inception V3, were trained and tested using validation accuracy and loss metrics. Results indicate that VGG16 had the highest accuracy, followed by MobileNet, which provided a good balance between performance and computational efficiency. Results prove that systems based on CNN can improve greatly in reliability, scalability and efficiency in wall defect detection, offering an alternative to hand inspection in industrial and construction applications.