This paper focuses on the use of deep neural networks for crack detection in plate structures using seismic wave-based techniques. The authors extend previous work by conducting extensive experiments on different network components and proposing a new normalization strategy for input wave signals. The proposed methods are tested on an expanded crack detection dataset, and the results show that a robust backbone network, such as Densely Connected Convolutional Network (DenseNet), can effectively extract features characterizing cracks from wave signals. By using the reference wave field for normalization, the accuracy of detecting small cracks is further improved. The paper also discusses the challenges and considerations in selecting appropriate encoder and decoder networks, and evaluates the impact of crack size on model performance. The results highlight the effectiveness of advanced networks like DenseNet in predicting crack existence, especially for smaller cracks, which are more difficult to detect due to their insignificant wavefield changes.This paper focuses on the use of deep neural networks for crack detection in plate structures using seismic wave-based techniques. The authors extend previous work by conducting extensive experiments on different network components and proposing a new normalization strategy for input wave signals. The proposed methods are tested on an expanded crack detection dataset, and the results show that a robust backbone network, such as Densely Connected Convolutional Network (DenseNet), can effectively extract features characterizing cracks from wave signals. By using the reference wave field for normalization, the accuracy of detecting small cracks is further improved. The paper also discusses the challenges and considerations in selecting appropriate encoder and decoder networks, and evaluates the impact of crack size on model performance. The results highlight the effectiveness of advanced networks like DenseNet in predicting crack existence, especially for smaller cracks, which are more difficult to detect due to their insignificant wavefield changes.