Aluminum material image defect detection method based on self-adaptive anchor frame

Abstract

The invention provides an aluminum material image defect detection method based on a self-adaptive anchor frame to solve the problems that a current defect detection method is not flexible enough, lowin detection precision and the like. Firstly, ResNeXt-101 using packet convolution and deformable convolution ideas is adopted as a backbone network, a feature enhancement module containing an attention mechanism is integrated into the backbone network, and then the feature enhancement module is sent into a feature pyramid network for multi-scale feature fusion, so that the defect detection precision is improved; secondly, a self-adaptive anchor frame neural network is used, corresponding anchor frame parameters are learned automatically according to defect features, and the precision of anchor frame positioning detection is improved; then, a cascade network structure is adopted in the frame prediction stage, and the problem that the precision in the training stage is not matched with that in the prediction stage is solved; and finally, the detection precision of the defects with large shape difference and small target defects is greatly improved, the overall precision of aluminum material image defect detection is relatively high, and the method has relatively high application value in the field of defect detection.

Description

technical field [0001] The invention belongs to the field of computer vision and defect detection, mainly adopts the idea of ​​deep learning, and specifically relates to an aluminum material image defect detection method based on an adaptive anchor frame. Background technique [0002] Aluminum profile is the pillar industry of all industrialized countries, with the characteristics of high strength, light weight, wear resistance, good decoration, environmental protection and economy, more and more architects start to use aluminum profile as building material. Therefore, quality has become a key evaluation criterion for aluminum profiles. However, in actual production, due to the impact of the production process, the aluminum surface may have defects such as non-conductivity, scratches, coarse grains, exposed boards, blisters, pits, protruding particles, exposed board corners, paint marks and variegated colors. . Due to the harsh environment and high cost, non-contact inspec...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to provide an aluminum image defect detection method based on an adaptive anchor frame to solve the problems that the current detection method is not effective and the detection means are not flexible enough

Images