Classification and Computing the Defected Area of Knots in Wooden Structures Using Image Processing and CNN

Abstract

Knots in wooden structures are common natural features in wood that result from where branches once joined the trunk of a tree. While they can add to the aesthetic appeal of wood, knots are often considered structural defects in construction because they can significantly affect the mechanical properties of wood. If knots are present in structural members, they cannot be ignored. Identifying the presence of knots and finding the corresponding defected area of a structural member is important to be able to reinforce the member, compensate for the reduced strength and ensure that it is safe and suitable for its intended use. In this study, the Inception-ResNet-V2 pre-trained Convolutional Neural Network (CNN) model is trained and validated with 2000 images for the classification of knots, and the defected area is calculated through Image Processing (IP) and other soft computing techniques. The images of knots are collected and equally classified into two categories: 1000 "Single knot" and 1000 "Multiple knots" images. 70% of the dataset is used for training, and 30% for model validation. Four statistical parameters, namely accuracy, precision, recall, and F1 score, are calculated to check the model performance for the classification task, as well as the corresponding confusion matrix. The model exhibited an overall accuracy of 84% in an independent evaluation with a new testing dataset of 200 images, while the defects could be properly quantified using IP techniques.The research work shows the potential of AI-based methodologies in structural health monitoring and damage identification. These methods can drastically improve our ability to assess the condition of structures and structural elements, offering enhanced precision and accuracy, real-time and cost-effective monitoring, predictive capabilities, and automation opportunities.