Image-based prediction of concrete mechanical properties through CNN segmentation and ANN modeling

• Curated dataset of 13,138 labeled concrete images with gray aggregates. • DeepLabV3+ with EfficientNet-B5 achieved 92% mIoU in aggregate segmentation. • The integrated CNN–ANN model predicted concrete strength with R² = 0.968. • The proposed model outperformed traditional non-destructive testing methods. Accurate prediction of the mechanical properties of concrete is essential for assessing structural performance and ensuring safety. Traditional destructive and non-destructive testing methods are labor-intensive, costly, and often impractical for evaluating in-service structures due to specimen extraction requirements, calibration challenges, and limited field accuracy. This study presents an automated image-based framework that integrates Convolutional Neural Network (CNN) and Artificial Neural Network (ANN) models to predict the compressive and tensile strengths of concrete from image-derived features. High-resolution images of concrete cylinders with different mix ratios and water–cement ratios were used to train a DeepLabV3+ segmentation model with EfficientNet-B5 as the encoder. The model achieved a mean Intersection over Union of 0.92 and an F1-score of 0.95 for coarse aggregate segmentation. Image analysis algorithms were then applied to extract aggregate characteristics, including proportion, size, roundness, and nearest-neighbor distance. Meanwhile, statistical descriptors of mortar pixels were derived to represent the matrix properties. These features served as inputs to multiple ANN models. The hybrid AggMortarNet model, combining aggregate and mortar characteristics, achieved the best predictive performance with R 2 = 0.9680 and RMSE = 0.0363 for compressive strength, and R 2 = 0.9607 and RMSE = 0.0368 for tensile strength, outperforming non-destructive methods. The results demonstrate an automated, data-driven approach with strong potential for Structural Health Monitoring (SHM) and quality assessment of both newly cast and existing concrete structures.