Hot-rolled strip steel surface defect detection method based on deep learning

Abstract

The invention discloses a hot-rolled strip steel surface defect detection method based on deep learning. The method comprises the following steps: 1, constructing a data set, marking all pictures, andrecording the position and category information of all defect targets; 2, clustering by using a weighted K-means algorithm to obtain a priori frame parameter required by detection; before clustering,calculating a sample weight according to a preset point corresponding to the feature map layer; 3, constructing an improved YOLOv3 algorithm network model; up-sampling the output of the previous detection layer in the FPN, fusing the up-sampled output with the shallow output added with the residual unit, and performing convolution to form a new feature layer; 4, setting the number of iterations,and optimizing network parameters by using an Adam optimizer; 5, training the training set, and storing the trained model and parameters; and 6, detecting the test set by using the stored model and parameters to obtain the detection precision and the detection speed of the model. The detection method can improve the detection precision and the detection speed of the surface defects of the hot-rolled strip steel.

Description

technical field [0001] The invention belongs to the technical field of hot-rolled strip surface defect detection, and relates to a method for detecting hot-rolled strip surface defects based on deep learning, in particular to a hot-rolled strip surface defect detection method based on YOLOv3 (You Only Look Once) algorithm method. Background technique [0002] As one of the important products of the steel industry, strip steel has played an important role in the fields of national defense equipment, automobile manufacturing, aerospace and other fields. Its surface quality seriously affects the performance of the final product, so improving the surface quality of the strip as much as possible is of great significance to improving the performance of the final product. At present, the detection task of hot-rolled strip surface defects is generally completed by traditional machine learning methods (deep learning algorithms). However, the surface defect detection method of hot-r...

AI Technical Summary

Problems solved by technology

[0005] However, the original YOLOv3 algorithm has great shortcomings in the detection of small targets

In the problem of strip surface defect detection, different types of defects vary greatly in shape and size. If the original YOLOv3 network is used for detection, its detection effect on small defects is poor

Taking a picture with a size of 256*256 as an example, the final grid sizes of the original YOLOv3 network are 32*32, 16*16, and 8*8 respectively, and the maximum number of preselection boxes that can be obtained is only 1344, which is difficult to meet the detection requirements. need

Images