Concrete Roughness Grade Identification Based on the Improved AlexNet

Because the roughness of the concrete binding surface affects its shear resistance and bond strength, so designing a deep learning model can identify the roughness of the concrete binding surface, which can detect the roughness of the binding surface more intuitively and quickly. First, based on the AlexNet model, we identify the roughness level of the collected depth map of the concrete binding surface, and obtain the deep learning model that identifies the roughness of the main network. Later, the recognition accuracy of the model is improved by embedding the attention mechanism behind the convolutional layer of AlexNet in three ways. In Method 1, the CBAM channel attention mechanism is embedded behind each convolution layer of AlexNet to obtain the AlexNet_CC model. In Method 2, the CBAM spatial attention mechanism is embedded behind each convolution convolution layer of AlexNet to obtain the AlexNet_CS model. In Method 3, the CBAM attention mechanism is embedded behind each convolution layer of AlexNet to obtain the AlexNet_C model. Through the simulation, the accuracy of the discovery method 3 is better than that of method 1, method 2 and the original network, and the accuracy can reach 95.36%, so the AlexNet_C model can better identify the roughness level of the concrete binding surface.