Research On Industrial Steel Defect Detection Based On Deep Learning

Inspection of industrial steel defects is an important part of industrial production and quality management.Cracks,spots and other defects on the surface of steel will seriously affect the quality of products and bring uncontrollable harm.Intelligent detection of steel defects is a technical bottleneck that has troubled the industry for many years.The efficiency of manual detection is difficult to grasp and the detection standard lacks consistency.The target detection algorithm using deep learning technology has a remarkable effect in the field of steel defect detection,but most of them are not applicable to mobile or embedded devices with low power consumption and limited computing resources.Combined with deep learning technology,this thesis improves the Tiny-YOLOv3 algorithm and uses embedded devices to build a model of steel defect detection system to ensure the accuracy and realtime performance of defect detection.The main research results are as follows:1?In order to balance the accuracy and speed of steel target defect detection,an improved target detection algorithm is proposed,R-Tiny-YOLOv3.The improved algorithm is based on the Tiny-YOLOv3 network structure.First,the residual network is added to improve the depth of the network and the accuracy of detection.The improved Spatial Pyramid Pooling layer module is added to enhance the feature extraction capability of the network.Then,combining the characteristic information of different network layers,the detection scale is increased from two to three scales.Finally,CIOU is selected as the loss function to make the regression of target detection more stable.The improved R-TinyYOLOv3 algorithm and Tiny-YOLOv3 algorithm were used for comparative experimental analysis.The detection accuracy of R-Tiny-YOLOv3 algorithm for steel defects reached71.5%,which was improved by 10.8% compared with the Tiny-YOLOv3 algorithm,and the detection speed reached 39.8 frames per second.The balance between detection accuracy and speed is realized,and the real-time detection requirements of embedded equipment are also met.2?The model of industrial steel defect detection system based on edge intelligence is designed.The system model consists of hardware detection platform and software management platform.The hardware detection platform adopts Cambrian edge intelligent1H8 development board,and the improved defect detection model is transplanted to the 1H8 edge intelligent embedded platform.The computing tasks of the whole detection algorithm are all deployed at the edge end and real-time detection is completed.The detection platform can operate independently from the data center,which improves the inspection efficiency of steel defects.Then,based on the hardware platform,a defect detection software system is designed,and the user interface is provided.It is specifically divided into defect detection module,alarm module,data query module,and output detection report module.Edge quality inspection personnel to remote control by the software system testing equipment,to be able to live on the production line of steel production,processing and maintenance of batches of data query and report output of monitoring results of steel production,to provide data support for the subsequent product quality traceability,improve the efficiency of quality inspection personnel steel flaw detection task.Figure [51] table [9] reference [70]...