Structural Identification and Quality Inspection of Underground Caverns in Hydroelectric Engineering via Computer Vision and Neural Network Computing

Structural Identification and Quality Inspection of Underground Caverns in Hydroelectric Engineering via Computer Vision and Neural Network Computing

연구 분야: Verification

학회: ICDIS '24: Proceedings of the 2024 International Symposium on Integrated Circuit Design and Integrated Systems

Ensuring the structural integrity and operational safety of underground caverns in hydropower projects is critical for the efficient and safe functioning of these facilities. Traditional inspection methods are time-consuming, prone to human error, and often inadequate for capturing complex structural details under challenging conditions. With the rapid advancement of artificial intelligence, neural networks have emerged as powerful tools for automating and improving the accuracy of structural inspections. In this study, we propose a novel approach leveraging the ConvNeXt-B model combined with a Weighted Channel Selection Module (WCSM) and Semantic Feature Enhancement to perform fine-grained image recognition of underground cavern structures. The WCSM enables the model to focus on high-dimensional and low-dimensional image features, filtering out redundant information and enhancing classification accuracy. Meanwhile, the Semantic Feature Enhancement Module improves the model's ability to capture detailed, discriminative features by grouping feature maps into semantic clusters. Experimental results on a specialized dataset of cavern images demonstrate that this approach significantly enhances recognition performance, addressing the limitations of manual inspections and advancing the use of neural networks in civil and industrial engineering.