Estimasi Kuat Tekan Beton Normal Menggunakan Metode Image Processing Dengan Convolutional Neural Network (CNN)

Conventional compressive strength testing of concrete is destructive, timeconsuming, and damages samples, necessitating more efficient alternatives. This study aims to develop a predictive model for the compressive strength of normal concrete using image processing based on Convolutional Neural Network (CNN) as a non-destructive approach. The study involved 60 concrete samples with a
target strength of 17.5 MPa, divided into 70% training data and 30% testing data. The process included image
preprocessing, normalization, data augmentation, and
CNN model training using a modified MobileNetV2 architecture, optimized with the Adam algorithm and Mean Squared Error (MSE) loss function. Results showed
actual compressive strengths ranging from 19.1 MPa to 39.0 MPa (average 29.8 MPa), while CNN model predictions ranged from 17.97 MPa to 40.64 MPa (average 30.29 MPa). Although the average MSE of 3.21 MPa indicates moderate
performance, the coefficient of determination (R²) of 0.0023 suggests that model accuracy needs improvement. This study provides an initial contribution to non destructive visual-based concrete testing using CNN. Recommendations for further development include expanding the dataset and fine-tuning the CNN architecture to enhance prediction accuracy.