A method for edge detection of steel coil end face based on deep learning

Abstract

The invention discloses a steel coil end face edge detection method based on deep learning, which includes improving the deep convolutional neural network structure, respectively improving the multi-scale feature extraction part and the multi-scale feature fusion part, and the multi-scale feature extraction part consists of It is composed of five feature extraction modules of different scales. The multi-scale feature fusion part is mainly composed of a convolutional layer and a deconvolutional layer for image acquisition. The acquired image is sent to a deep convolutional neural network for edge detection, and then the output edge Detection results; the present invention improves the deep convolutional neural network, the deep convolutional neural network automatically learns features from the data, can learn to ignore the edge of the number, and only express the edge of the steel coil, which has better robustness. Rod, good detection effect.

Description

technical field [0001] The invention relates to the technical field of industrial vision, in particular to a method for detecting the edge of a steel coil end face based on deep learning. Background technique [0002] The end face of the steel coil has strong regular texture edge features. In order to automatically detect the defect of the end face of the steel coil, the first step is to detect the edge of the end face of the steel coil to provide auxiliary information for subsequent defect detection. [0003] In the prior art, edge detection algorithms based on traditional image processing often start from the perspective of calculating image pixel gradients, and use various edge detection operators to perform sliding window calculations on images through convolution filtering. Among them, the most commonly used It is the canny edge detection algorithm. However, when applied to the edge detection field of steel coil end faces, the edge images obtained based on traditional ...

AI Technical Summary

Problems solved by technology

However, when applied to the edge detection field of steel coil end faces, the edge images obtained based on traditional image processing algorithms often have many noise points, such as highlights and shadows generated by light, and uneven parts caused by oxidation on the surface of steel coils, etc. The edge image detected by the traditional algorithm produces a lot of interference, and the loss function of the traditional edge detection uses a pixel-by-pixel binary cross-entropy loss function, and then sums the pixel-by-pixel loss values ​​to obtain the final loss value, which is Since the pixels occupied by the edge lines are often very few, and the background pixels are particularly large, the imbalance between the two categories will cause the network to easily fall into the local extremum during the training process, resulting in low accuracy of the calculation results.

Images