Intelligent identification method, system and device for quality defects of hardened concrete

Abstract

This invention discloses an intelligent identification method, system, and device for quality defects in hardened concrete vibration: A spatiotemporal-process joint database is constructed through data acquisition and annotation; enhanced visual feature vectors are obtained through image preprocessing and feature extraction; knowledge graph vectors are obtained through knowledge graph construction and semantic encoding, generating a unified semantic vector; cross-modal fusion and reasoning are performed to fuse the visual feature vectors and knowledge graph vectors, and the causes and levels of defects are inferred in conjunction with engineering specifications, resulting in fused multimodal features; structured defect description text is generated and optimized through text generation and optimization; based on the optimized defect description text and the original acquired image data, a structured report is generated, and the defect area and reasoning path are visualized on the image. This invention combines computer vision and natural language processing to automatically identify concrete vibration defects and generate structured reports, replacing manual inspection and improving quality control efficiency.

Images