Ship weld defect detection method based on deep convolutional neural network model

Abstract

The invention discloses a ship weld defect detection method based on a deep convolutional neural network model. Firstly, an M-SMOTE algorithm based on the distance extreme value is adopted to carry out weld sample unbalanced data set processing, and fine generation of minority class new samples is achieved; then, a deep convolutional neural network model which comprises 13 layers and faces ship weld defect detection is constructed, the input of an input layer takes a result obtained by carrying out improved affinity propagation clustering on feature data of a ship X-ray weld image set, and input data of each middle layer of the deep convolutional neural network is subjected to adaptive PCA dimensionality reduction; then training the deep convolutional neural network model; and finally, performing weld defect detection by using the trained deep convolutional neural network model. According to the ship welding seam defect detection method, the technical difficulties of welding seam data imbalance, defect feature selection, data high dimension and the like are effectively solved, and efficient and effective detection of ship welding seam defects is achieved.

Description

technical field [0001] The invention belongs to the technical field of pattern recognition, and relates to a ship weld defect detection technology based on an X-ray weld image set, in particular to a ship weld defect detection method based on a deep convolutional neural network model. Background technique [0002] Welding is the main process of shipbuilding, but because the welding process is a complex phase transition process coupled with multiple factors, it is easily disturbed by the external environment and human factors, and slag inclusions, cracks, pores, Unpredictable welding defects such as incomplete penetration will seriously reduce the mechanical properties of welded components, thus often having a great adverse effect on the safety performance of key ship components. In order to find the defects in the weld in time, non-destructive testing of welding defects has become the main technical means, and X-ray non-destructive testing technology has been widely used in ...

AI Technical Summary

Problems solved by technology

[0005] (1) In the actual welding process of the ship, the probability of occurrence of each defect (such as slag inclusion, crack, air hole, incomplete penetration) is not the same, so the pictures obtained by X-ray detection usually may appear serious unbalanced type Problem, unbalanced defect data will affect model learning, thereby reducing model generalization ability and prone to underfitting;

[0006] (2) The feature extraction of existing ship welding defects is mainly based on the geometry and strength characteristics of the weld, but the gray value distribution of different types of welding defects is often different from its background, and the background contrast is ignored in the feature extraction , which has a great impact on the training of the deep learning network model, and directly affects the recognition accuracy;

[0007] (3) Since the deep convolutional neural network model has a large number of trainable variable architectures, the model often performs well in recognition accuracy, but because the structure is relatively complex, the model requires a large amount of storage space and the operation speed is slow. The common method is the main Component dimensionality reduction method, but the traditional PCA algorithm requires a certain prior knowledge of the dimensionality reduction object, and cannot intervene in the dimensionality reduction process through parameterization and other methods. It may not achieve the expected effect and the efficiency is not high.

Images