.png){kind=icon}
.png){kind=icon}
Return
Image enhancement of workpiece surface defects based on autapse coupled neural network
.png){kind=icon}
PRE
AI
X
W
K
Q
C
W
Z
DOI：10.1080/10589759.2025.2560601.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In industrial production, surface defects like pitting occur inworkpieces, seriously affecting the reliability of parts andequipment. However, obtaining clear defect images remains a majorchallenge. In this paper, an autapse-coupled neural network (ACNN)algorithm, inspired by the stochastic resonance phenomenon inbiology, is proposed to enhance images of workpiece surface defectsunder low illumination. The integration of the autapse model with theneural network increases the number of firing neurons, improves theenhancement of weak signals, and enriches the model's biologicalinterpretability. Selected 300 and 500 images from the Tianchi andAeBAD datasets, respectively, and conducted image enhancementexperiments using the ACNN algorithm alongside BEMD, SR, SSNN, andDnCNN algorithms. The peak signal-to-noise ratio (PSNR) andstructural similarity index measure (SSIM) were used as evaluationmetrics for the enhanced low-illumination workpiece surface defectimages. Furthermore, we compared the defect recognition rates ofimages enhanced by the different methods. The experimental resultsdemonstrate that the image enhancement capability of the proposedalgorithm is superior to the other four algorithms. This verifies thepracticality and superiority of the ACNN algorithm for enhancingsurface defect images of low-light workpieces, making it suitable forapplication in production practice.
Keywords:
Autapse
image enhancement
neural network
stochastic resonance
workpiece surface defects
Journal
N
IF:
4.2
Papers:
1.7K
Citations:
2.1K
Organization
Cited Papers