Steel Metallurgy Crack Detection and Material Pile Volume Measurement Based on 3D Point Cloud and Improved SSD Model

For crack detection and stockpile volume measurement in steel metallurgy production process, an improved single shot multibox detector model and Delaunay triangulation technique based on greedy projection triangulation are proposed, respectively. Among them, the study mainly adopts ShuffleNet V2 module and efficient channel attention network to optimize the detection efficiency and accuracy of the original detection model. Meanwhile, based on the 3D point cloud dissection results, the triangular projection with depth is used to realize the pile volume measurement. The results showed that compared with other advanced models, the improved single shot multibox detector model achieved a mean average precision of 98.68% in steel plate crack detection, which was 26.7% higher than the faster region based convolutional neural network. The area under the curve of the receiver operating characteristic curve of the improved model was as high as 0.8576, and the detection time was only 29.85 ms. Meanwhile, the mean relative error of the proposed stack metrology method in the metrology experiments was only 2.15%, while the mean relative error values of the Digital Surface Modeling Method and volume balance method were as high as 3.94% and 6.71%, respectively. This indicates that the proved crack detection and stockpile metrology methods for steel plates have significant performance advantages and provide effective technical support for quality control in the production process of iron and steel metallurgy.