Loom Shed Image Detection Using Pseudo-Anomalous Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing loom abnormality detection systems face accuracy issues in specifying the range of the shed portion due to components obstructing the view of warp yarn end points and cloth fell, leading to inaccurate pseudo-anomalous image data and decreased classification model performance.
Innovation Solution
A loom abnormality detector with a pseudo-anomalous image creation unit that provisionally specifies virtual warp yarn end points and cloth fell points, using reference lines to align with actual warp yarns, allowing for accurate definition of the shed portion range, and creates pseudo-anomalous images to enhance the classification model's training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If pseudo-anomalous image data are created by modifying shed portion images to increase training data, then the quantity of training data increases, but the accuracy of specifying the shed portion range decreases when components obstruct the view
Solution Approach 1:
The system performs preliminary actions by automatically specifying the shed portion range using image processing and recognition techniques before creating pseudo-anomalous images. This ensures accurate baseline definition of the shed portion boundaries (upper warp yarn end point, lower warp yarn end point, and cloth fell) even when components may obstruct the view, thereby maintaining measurement precision while enabling subsequent data augmentation.
2Reliability
If the shed portion range is specified using captured images, then the detection system can identify weaving abnormalities, but components near the shed portion may hide key points reducing specification accuracy
Solution Approach 1:
The system introduces an intermediary image processing approach that processes the captured shed portion images to automatically identify and specify the three critical points (upper warp yarn end point, lower warp yarn end point, and cloth fell). This intermediary processing step enhances the reliability of abnormality detection by ensuring accurate specification of the shed portion range even when components obstruct the direct view of key points.
Data Source
Figure 1
Figure 2~3
Figure 4~5
AI summary
A loom abnormality detector (30) includes a classification model that has learned a normal image that shows a shed portion (R) in a normal state and an anomalous image that shows the shed portion (R) in an abnormal state as a shed portion image that shows the shed portion (R) defined by a reed (16), upper warp yarns (Tu) above a weft yarn (Y), and lower warp yarns (Td) below the weft yarn (Y), and a cloth fell (Wa). The loom abnormality detector (30) detects a weaving abnormality from the shed portion image using the classification model. The loom abnormality detector (30) further includes a pseudo-anomalous image creation unit (50) configured to artificially create the abnormal image from the shed portion image. the pseudo-anomalous image creation unit (50) includes a provisionally specifying unit (51), a reference line illustrating unit (52), a decision unit (53), and an illustrating unit (54).