Research on a Road Crack Detection Method Based on PCA Dimensionality Reduction and a Lightweight Neural Network

To address the challenges of high-dimensional feature redundancy and constrained edge deployment in road crack detection under small-sample conditions, this paper proposes a lightweight detection framework integrating principal component analysis (PCA) for dimensionality reduction with a multilayer perceptron (MLP) classifier. The proposed method first applies image preprocessing, followed by PCA-based dimensionality reduction to extract the most discriminative feature components, and subsequently employs a compact MLP to perform binary classification. A PCA combined with support vector machine (SVM) pipeline serves as the baseline for comparison. Model stability is evaluated through PCA dimension sensitivity analysis and five-fold stratified cross-validation. Furthermore, model size and inference latency are quantitatively analyzed to assess deployment feasibility. Experimental results demonstrate that the proposed method achieves a classification accuracy of 91.7% and an F1-score of 93.3% on the test set while maintaining low computational overhead, outperforming PCA+SVM in terms of overall performance and stability. Finally, a Python-based interactive visualization interface is implemented to verify the practical deployability of the system.

Keywords: Road crack detection, Principal Component Analysis (PCA), Lightweight neural network, Support Vector Machine (SVM), Image classification.