基于YOLOv9-DS的预制混凝土叠合板质量检测模型研究OA

To address issues such as missing embedded parts and positional deviations during the production and installation of prefabricated concrete composite slabs,an improved YOLOv9-DS model for quality inspection of prefabricated concrete composite slabs was proposed.Data augmentation was further employed to expand the samples for model training and evaluation.First,using the YOLOv9 detection algorithm as the baseline framework,dynamic snake convolution(DSConv)and spatial and channel reconstruction convolution(SCConv)modules were incorporated.Second,Taking a prefabricated concrete factory as the engineering background,a detection dataset comprising three types of embedded parts,namely junction boxes,water pipe openings,and truss steel bars,was established.The results indicate that the improved YOLOv9-DS model achieves an overall mAP50 of 87.8％on the test set,with an mAP50 value of 89.9％for truss steel bars,89.0％for junction boxes,and 84.5％for water pipe openings,representing improvements of 3.2 percentage points,1.2 percentage points,and 3.3 percentage points over the original YOLOv9 model,respectively.The model contains 56.89M parameters,and the inference time was reduced from 13.4 ms to 9.7 ms,corresponding to a decrease of 27.6％.The proposed YOLOv9-DS model effectively improves both the accuracy and efficiency of multi-scale embedded parts recognition in composite slabs,demonstrating its capability to satisfy the requirements of robust and real-time detection in complex industrial scenarios.