Small sample target identification method based on transfer learning

Abstract

The invention belongs to the technical field of image processing, and particularly relates to a transfer learning-based small sample target identification method, which comprises the following steps of S1, dividing glass defect samples into three categories of scratches, edge breakage and smudginess for image sampling to obtain sample images; s2, making the sample images into a sample data set; s3, constructing a glass surface defect model, and pre-training the glass surface defect model by taking the hot-rolled strip steel surface defect data set as feature input to obtain an initial weight of the glass surface model; s4, acquiring a hot-rolled strip steel surface defect data set, and training the glass surface defect model through the sample data set by adopting a transfer learning method to obtain an optimized glass surface defect model; and S5, inputting the to-be-detected glass image into the glass surface defect model to obtain the defect category of the to-be-detected glass. The small sample target identification method provided by the invention can effectively solve the problem of small data sample size.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a small sample target recognition method based on transfer learning. Background technique [0002] In the complex deep processing process in the field of industrial application, such as the deep processing of the original flat glass generally needs to go through cutting, edging, cleaning, polishing and other processes, and some also need to go through drilling, printing and tempering, etc., due to technical conditions and production environment Or human factors, defects may occur. [0003] The target detection model in deep learning can be used to implement tasks such as defect detection, target positioning, and size measurement. It is a detection method that can be used in the field of defect detection. The neural network extracts the features of the defect, completes the training and outputs the target detection model, and completes the identification and ...

AI Technical Summary

Problems solved by technology

However, deep learning is a technology that requires a high amount of data. Its ability to achieve excellent results mainly depends on a large number of training samples with class labels, but there are not enough class labels in real industrial applications. samples, or the cost of obtaining a large number of labeled samples is very high

Images