A Machine Learning Model Based on MobileNet V2 and IGOA-XGBoost for Structural Health Monitoring

Abstract

Monitoring the structural health of infrastructure such as buildings, bridges, and roads requires efficient and accurate crack detection methods. This research proposes the integration of MobileNet V2 for concrete crack image feature extraction with the Improved Grasshopper Optimization Algorithm (IGOA) and XGBoost for classification. MobileNet V2 was chosen for its efficiency in feature extraction, while IGOA was implemented to optimize the hyperparameters of XGBoost, improving classification performance. The dataset used includes 40,000 concrete crack images with a proportion of 70% training data and 30% test data. The experimental results show that the IGOA-XGBoost method provides the best performance with 99.75% accuracy, 99.80% precision, 99.70% recall, and 99.75% F1-score. The advantage of this method lies in IGOA's adaptive metaheuristic optimization mechanism, which balances exploration and exploitation in the search for optimal hyperparameters. This research proves the potential integration of MobileNet V2 and IGOA-XGBoost as an effective solution for an accurate and efficient machine learning-based crack detection system, supporting the development of real-world structural health monitoring systems.

Access to Document

Other files and links

• Link to publication in Scopus
• Open Access Version Available

Fingerprint