Training method of object defect detection model and object defect detection method and device

Abstract

The invention discloses an object defect detection model training method and device and an object defect detection method and device, and the method comprises the steps: obtaining a to-be-trained image of an object, carrying out the feature extraction of the to-be-trained image through a feature extraction sub-network of the object defect detection model, and obtaining a feature map of the to-be-trained image; classifying the feature map of the to-be-trained image through a classification sub-network of the object defect detection model to obtain a classification result and a classification loss value of the image; performing defect detection on the defective image to be trained through a defect detection sub-network of the object defect detection model to obtain a defect detection resultand a defect detection loss value of the defective image; and optimizing parameters of the object defect detection model by using a gradient descent algorithm according to the classification loss value and the defect detection loss value to obtain a trained object defect detection model. According to the training method, the overall consumed time of model training is shortened, and the detection precision of the model is improved.

Description

technical field [0001] The present application relates to the technical field of object detection, in particular to a training method for an object defect detection model, a method and a device for object defect detection. Background technique [0002] In the technical field of object detection such as workpieces, collecting ultra-high resolution workpiece images is an important guarantee for improving the accuracy of workpiece detection results. Ultra-high-resolution images usually refer to images with more than 50 million pixels. However, when using ultra-high-resolution images for training and testing of defect detection models, the following problems will be encountered: 1) The image resolution is too large, resulting in existing The defect detection model of our company takes up too much video memory during training, and it cannot be trained normally on a GPU (Graphics Processing Unit, Image Processor) with only 16G video memory; 3) The overall running time of the mode...

AI Technical Summary

Problems solved by technology

Ultra-high-resolution images usually refer to images with more than 50 million pixels. However, when using ultra-high-resolution images for training and testing of defect detection models, the following problems will be encountered: 1) The image resolution is too large, resulting in existing The defect detection model of our company takes up too much video memory during training, and it cannot be trained normally on a GPU (Graphics Processing Unit, Image Processor) with only 16G video memory; Large defects and small defects that may exist on the model; 3) The overall running time of the model is longer

[0004] However, the inventors found that the above-mentioned training method for the defect detection model for ultra-high resolution images still has the problem that the training efficiency and detection accuracy of the model cannot be guaranteed at the same time.

Images