Method of identifying woven fabric defect

Abstract

The invention discloses a method to identify flaw on surface of woven. Input the image signal which gained from image acquisition system to computer and proceed filtering and enhancement. Extract energy, entropy, contrast, deficit moment and correlation from vertical and horizon high frequency detail of image signal on woven flaw as five texture distinction parameter of woven. Use optimal edge detection algorithm on image signal of woven flow. Abstract apsidal ratio from image signals as a form characteristic parameter of woven. Input these six parameters to neural network system to proceed identifying and assorting on flows. The invention put texture and morphological character as evidence of differentiating flows of woven. Using neural network system to proceed identifying and assorting, the accuracy of flows can be increased obviously. Because of simple operation, fast detection and high efficient, the method has good prospect for applying.

Description

technical field [0001] The invention relates to a method for identifying surface defects of woven fabrics. Background technique [0002] In modern textile enterprises, fabric defect inspection is a necessary process. However, most of the inspections are still done manually, and the inspectors first receive training according to the inspection standards. When the fabric is actually inspected, the fabric is first unwound to the inspection device, and the inspector uses the method of visual inspection to find the defects on the surface of the fabric, compares it with the defects in the inspection standard, and then classifies them. When a defect is found, if it is not allowed by the inspection standard, mark the defect and correct the defect. If the defect cannot be corrected, the quality of the fabric will be degraded. This inspection method is completed offline by artificial vision, and the inspection standard is manually controlled, which is time...

AI Technical Summary

Problems solved by technology

From WO98 / 08080, a method and device for identifying defects in fabrics can be known, and the brightness value of a piece of fabric is collected and input to a neural network for linear filtering operation, thereby judging whether there are defects. Obviously, using a simple linear neural network The network discrimination has insufficient accuracy, low efficiency and other deficiencies; it can be learned from the invention patent with the publication number CN1203229C "Method and Device for Evaluating Fabric Defects". Defects with different lengths and contrasts are stored in the computer. During detection, the defects are evaluated by comparing a known defect with a manual illustration

However, as far as the surface texture of a fabric is concerned, the actual changes are infinite, and it is difficult to accurately simulate defects, which affects the accuracy of the test results to a certain extent.

Images