Feasibility analysis and opposition white shark optimizer for optimizing modified EfficientNetV2 model for road crack classification

Pavement cracks start as obvious cracks in the surface layer and progress to deeper layers, affecting the overall structure. Therefore, detecting asphalt pavement cracks is a critical and time-consuming task. This study establishes an intelligent approach for automatic asphalt pavement crack classification, based on the integration of Opposition-based Learning (OBL), White Shark Optimizer (WSO), and Convolution Neural Networks (CNNs). This involves creating a Modified variant of the pre-trained CNN EfficientNetV2 (MEfficientNetV2) model. The proposed method was combined with Principal Component Analysis (PCA) to ensure efficient feature selection and dimensionality reduction. A set of pre-processing steps was applied to further improve the quality of images and remove any noise or artifacts. Extensive experiments were managed to train and validate the proposed asphalt pavement crack classification method on low-contrast natural images with several class labels. These images were gathered from many publicly available asphalt crack datasets to assess the adaptability and specificity of the proposed method. The proposed pavement crack classification method surpasses existing methods with a precision of 96.9317%, a recall of 98.6358%, a specificity of 98.9377%, an accuracy of 98.8368%, and an F1-score, representing the harmonic average of a classification model’s recall and precision, of 98.6358%. Thus, the proposed method is a viable substitute to assist transportation agencies in routine pavement inspection tasks.