Cloth defect detection method based on deep learning multi-layer feature fusion

Abstract

The invention discloses a cloth defect detection method based on deep learning multilayer feature fusion, and the method comprises the following steps: S01, building cloth detection image libraries and label libraries of different defect types, and importing defect type pictures; S02, expanding data in the cloth detection image library by utilizing the data augmentation technology; S03, dividing the expanded cloth detection image library data into a training set and a test set, and processing the data; S04, designing a target detection network model with multi-layer feature fusion; S05, training the designed network model by adopting the training set; S06, deploying the trained network model to an upper computer, and applying the network model to an automatic cloth detection production line to detect the cloth. According to the invention, the defect detection method integrates a network structure with multi-layer features, can accurately detect the position and type of the defect, doesnot need to preset a preset frame, also achieves the end-to-end microscopicity of the model, does not need to be observed by naked eyes, and unifies the detection standard.

Description

technical field [0001] The invention relates to the technical field of cloth defect detection, in particular to a cloth defect detection method based on deep learning multi-layer feature fusion. Background technique [0002] In cloth production, in order to ensure that the produced cloth meets the production standards, it is necessary to carry out quality defect detection on the cloth, and most of the cloth defect detection in the production line relies on manual detection, but the detection speed of manual detection of cloth defects is slow, and at the same time The test results are affected by the experience, proficiency and some subjective factors of the inspectors, and lack consistency and reliability; the traditional cloth defect detection techniques mainly include statistical methods, spectrum methods, model methods, and learning methods. These methods involve complex The statistics and analysis of the characteristics of the system require a large amount of calculation...

AI Technical Summary

Problems solved by technology

[0002] In cloth production, in order to ensure that the produced cloth meets the production standards, it is necessary to carry out quality defect detection on the cloth, and most of the cloth defect detection in the production line relies on manual detection, but the detection speed of manual detection of cloth defects is slow, and at the same time The test results are affected by the experience, proficiency and some subjective factors of the inspectors, and lack consistency and reliability; the traditional cloth defect detection techniques mainly include statistical methods, spectrum methods, model methods, and learning methods. These methods involve complex Statistical and analysis of the characteristics of the system, the amount of calculation is large, it is difficult to meet the industrial requirements

[0003] At present, the method of using deep learning method for cloth defect detection mainly uses a two-stage target detection algorithm. This method first needs to generate a large number of candidate pictures, and then uses convolutional neural network to classify and return the pictures, thereby reducing the cost of cloth defect detection. The rate increases the time of cloth inspection and cannot guarantee the quality consistency of cloth products

Images