Automobile sheet metal part surface defect detection method based on deep transfer learning

Abstract

The invention discloses an automobile sheet metal part surface defect detection method based on deep transfer learning. The method comprises the following steps: 1, image acquisition; 2, image preprocessing and feature extraction; 3, performing first-order network training; 4, performing first-order network output and performance analysis; 5, adjusting samples, and performing second-order network training; 6, inputting an image to be detected; 7, network decision making; 8, reconstructing a slice image, and positioning defects; and 9, outputting a detection result. According to the method, through the steps, the automobile metal plate surface defects are effectively detected based on image processing, machine vision and deep learning.

Description

technical field [0001] The invention relates to the field of defect detection, and specifically relates to a method for detecting surface defects of automobile sheet metal based on deep transfer learning. Background technique [0002] With the development of information technology, visual inspection systems based on pattern recognition algorithms are increasingly used in various industrial production activities instead of manual visual inspection. The automobile industry is the benchmark of a country's high-end manufacturing industry. It has high requirements for product quality and stability. Any tiny defect in the body sheet metal parts, such as scratches, abrasions, bumps, cracks, etc., may cause have a huge impact on the final product quality. Therefore, automatic identification and detection of defects on the surface of body sheet metal parts is one of the most important processes in the automobile manufacturing process. [0003] Traditional machine vision detection m...

AI Technical Summary

Problems solved by technology

However, the defect detection method based on the convolutional network at this stage mainly has the following problems: First, the object of the existing algorithm is usually an independent image reflecting an independent scene, and whether it is using the sliding window detection strategy of AlexNet or Using the target detection strategy of Faster R-CNN, once the network makes misjudgments or missed judgments, it will directly affect the detection results

That is, in one calculation, the single-order network has a low error tolerance rate for a single image

Second, the existing algorithms preprocess the input image, mainly random cropping, flipping, mirroring, contrast and histogram adjustment, etc., which cannot adaptively weigh the relationship between the detection subject and the background area, causing the network to sometimes Take up too many resources to learn irrelevant content, affecting detection efficiency

Third, the filter denoising method of the existing algorithm will act on the detection subject and the noise at the same time, which will cause part of the detailed information of the detection subject to be filtered out, and this information will affect whether the network can learn effective features

In addition, the automotive industry has higher requirements for the detection accuracy of sheet metal surface defects than the general industry, which is difficult to achieve with existing algorithms.

At the same time, deep learning is a big data-driven technology, and the data samples that can be provided in industrial scenarios are few, and using small samples to directly conduct network training can easily cause overfitting

Images