Industrial image defect detection method and system based on reconstruction, storage medium and product

By improving the encoder and decoder structure and combining linear attention and adaptive feature perturbation, the problems of insufficient encoder extraction capability and decoder identity mapping in the prior art are solved, and more efficient industrial image defect detection is achieved.

CN121883482AActive Publication Date: 2026-04-17SICHUAN SHUJU INTELLIGENT MFG TECH CO LTD
8 Cites 0 Cited by

Patent Information

Application Number
CN202610323160.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-03-17
Publication Date
2026-04-17
Estimated Expiration
2046-03-17

AI Technical Summary

Technical Problem

In existing reconstruction-based industrial image defect detection algorithms, the encoder's extraction capability is insufficient, resulting in the inability to effectively extract abnormal features, and the decoder has an identity mapping problem, leading to missed detections and false alarms.

Method used

The DINOv3 network is used as the encoder, combined with an improved structure of bottleneck layer and decoder. The training process is optimized by introducing a linear attention mechanism and adaptive feature perturbation. The patch set features of the reference sample map are used for decoding. The loss function of cosine distance and structural similarity index is combined to suppress the identity mapping of the decoder and improve the reconstruction ability.

Benefits of technology

It improves the ability to extract and reconstruct abnormal features, reduces the rate of missed detections and false alarms, enhances the accuracy and stability of detection, and adapts to interference from different hardware environments.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention discloses an industrial image defect detection method and system based on reconstruction, a storage medium and a product, belongs to the technical field of industrial image processing, and solves the problem of missing detection caused by insufficient extraction capability of an encoder and incapability of effectively extracting abnormal features in the prior art. The method comprises the following steps: constructing a reference sample library, constructing a reconstruction network, and training to obtain a trained reconstruction network; the reconstruction network comprises an encoder for performing encoding feature extraction on the input image, a bottleneck layer for aggregating the encoding features extracted by the encoder, and a decoder for fusing the aggregated encoding features of the bottleneck layer and patch set features of the matched reference sample image and decoding to obtain reconstruction features; inputting a to-be-detected industrial image and a corresponding reference sample image into the trained reconstruction network to obtain reconstruction features; and calculating the distance between the coding feature and the reconstruction feature to generate an abnormal score graph, and locating a defect area through threshold segmentation and contour search. The method is used for defect detection of the reconstructed industrial image.
Need to check novelty before this filing date? Find Prior Art