Deep Learning Activation Layer-Based Wall Quality Recognition Using Conv2D ResNet Exponential Transfer Learning Model

Crack detection is essential for observing structural health and guaranteeing safety. The manual crack other damage process time-consuming subject to surveyors’ biased judgments. proposed Conv2D ResNet Exponential model wall quality was trained with 5000 images, including various imperfections such as cracks, holes, efflorescence, damp patches, spalls. initial weights form the layers of base integrated Xception, VGG19, DenseNet, convolutional neural network (CNN) models retrieve general high-level features. A transfer deep-learning-based approach implemented create a custom layer CNN models. combined estimate quality. were fitted different activation softplus, softsign, tanh, selu, elu, exponential, along learning. performance evaluated using loss, precision, accuracy, recall, F-score measures. validated by comparing performances ResNet, Exponential. experimental results show that an exponential outperforms it value 0.9978 can potentially be viable substitute classifying defects.