Construction Progress Estimation via Deep Learning

Deep learning-based construction progress estimation using a custom CNN architecture with quantitative performance evaluation.

This study develops a custom CNN-based deep learning pipeline for automated estimation of construction physical progress from site imagery. The dataset was balanced to 250 samples per class and divided into 80/10/10 train–validation–test partitions.

The architecture integrates convolutional layers, batch normalization, ReLU activations, and global average pooling, optimized via Adam with controlled learning rate scheduling. Performance evaluation was conducted using classification accuracy and confusion matrix analysis.

The model achieved stable convergence without significant overfitting, demonstrating its potential for integration into AI-assisted project monitoring and digital construction management systems.