Intelligent detection of bridge surface cracks based on an improved convolutional neural network and metric representation learning
With the rapid development of artificial intelligence, the Internet of Things, and computer vision technologies, bridge structural health monitoring is increasingly evolving toward intelligent and automated inspection. As one of the most common structural defects, bridge surface cracks require timely and accurate identification, which is of great significance for bridge safety assessment, maintenance decision-making, and long-term service performance evaluation. However, conventional crack detection approaches mainly rely on manual inspection and traditional image processing algorithms, which are often limited by low efficiency, insufficient detection accuracy, strong subjectivity, and poor adaptability to complex environmental conditions. To address these limitations, this paper proposes an intelligent bridge surface crack detection method based on an improved convolutional neural network and metric representation learning. Specifically, a crack detection model is constructed using the TensorFlow framework and the Keras deep learning library. The proposed model employs convolutional neural networks to automatically extract edge, texture, and morphological features from bridge crack images. Meanwhile, the loss function and model parameters are further optimized, and metric representation learning is introduced to transform crack detection into an anomaly detection problem in the pixel embedding space, which improves the ability to distinguish subtle cracks from complex backgrounds by constructing a discriminative feature manifold. Experimental results demonstrate that the proposed method can effectively identify cracks on bridge surfaces, achieves a detection accuracy of 98.23% on the self-built dataset, and provides a feasible technical paradigm for the application of deep learning in bridge structural health monitoring.