.png){kind=icon}
.png){kind=icon}
Return
A lightweight algorithm for bottle cap sealing defect detection
.png){kind=icon}
PRE
AI
DOI：10.1016/j.engappai.2026.115318.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
To address the challenges of difficult deployment and slow detection speeds in existing defect detection models for bottle cap sealing scenarios, we propose a lightweight detection model, Context Guided-Cascaded Group Attention-Feature Pyramid Shared Convolution-Group Normalization Lightweight Detection-You Only Look Once (CCFG-YOLO), based on improved You Only Look Once version 11 nano (YOLO11n). Firstly, Context-Guided Cross Stage Partial with Kernel Size 2 (C3k2_CG) module, which takes the place of original Cross Stage Partial with kernel size 2 (C3k2) module in original backbone, is used to enhance the learning capability of joint features by incorporating global contextual features. Secondly, Cross Stage Partial with Cascaded Group Attention (C2CGA) module is incorporated into the backbone network to decrease the parameter count and floating-point operations. To reduce Floating Point Operations (FLOPs), we incorporate a module based on Spatial Pyramid Pooling–Fast (SPPF), Feature Pyramid Shared Convolution (FPSConv), into the architecture. Additionally, the detection head with a large receptive field is removed, and a new lightweight detection head, Group Normalization Lightweight Detection Head (GNLD), is devised to realize the lightweight design. Test results indicate that CCFG-YOLO reaches a mean Average Precision (mAP) of 84.3% at an Intersection over Union (IoU) threshold of 0.5 and 55.0% across the IoU range of 0.5:0.95. Additionally, it features low computational complexity, with 4.3 giga floating point operations (GFLOPs) per image (640 × 640), allowing for fast detection. The CCFG-YOLO model can detect surface defects on bottle caps rapidly and accurately, which provides algorithmic support for related fields in cap defect detection.
Journal
IF:
8
Papers:
5.3K
Citations:
3.5W