In:
Mathematical Biosciences and Engineering, American Institute of Mathematical Sciences (AIMS), Vol. 20, No. 7 ( 2023), p. 12341-12359
Abstract:
〈abstract〉〈p〉This paper addresses the issue of artificial visual inspection being overly reliant on subjective experience and the difficulty for the human eye to accurately identify dense and non-significant defects. To solve this problem, we have implemented an automatic object detection algorithm based on an improved version of YOLOv5.First, we use the K-means++ clustering algorithm to automatically calculate the Anchor of the model to reduce the effect of the close location of the initial clustering centers on the clustering of the sample data.Second, we add the Coordinate Attention (CA) attention mechanism to the model to allow the model to better capture and understand important features in the images. Then, we add a new detection layer with a downsampling multiplier of 4 to the Neck network to improve the precision of the model. Finally, we use the lightweight network MobileNetV3 instead of YOLOv5's backbone network to reduce the model detection time overhead.Our model achieves 85.87% mAP, which is 6.44% better than the YOLOv5 network, and the detection time for a single image is only 54ms, which is 50% faster than the YOLOv5 network. After testing, we have proven that our proposed algorithm can quickly and accurately detect the condition of bearing appearance defects, improving detection efficiency and reducing costs.〈/p〉〈/abstract〉
Type of Medium:
Online Resource
ISSN:
1551-0018
Language:
Unknown
Publisher:
American Institute of Mathematical Sciences (AIMS)
Publication Date:
2023
detail.hit.zdb_id:
2265126-3
