skip to main content
Idioma:
Tipo de recurso Mostra resultados com: Mostra resultados com: Índice

Space to depth convolution bundled with coordinate attention for detecting surface defects

Wan, Wenqian ; Wang, Lei ; Wang, Bingbing ; Yu, Haoyang ; Shi, Kuijie ; Liu, Gang

Signal, image and video processing, 2024-07, Vol.18 (5), p.4861-4874 [Periódico revisado por pares]

London: Springer London

Texto completo disponível

Citações Citado por
  • Título:
    Space to depth convolution bundled with coordinate attention for detecting surface defects
  • Autor: Wan, Wenqian ; Wang, Lei ; Wang, Bingbing ; Yu, Haoyang ; Shi, Kuijie ; Liu, Gang
  • Assuntos: Algorithms ; Channels ; Computer Imaging ; Computer Science ; Convolution ; Image Processing and Computer Vision ; Metal plates ; Modules ; Multimedia Information Systems ; Object recognition ; Original Paper ; Pattern Recognition and Graphics ; Plastic plates ; Signal,Image and Speech Processing ; Steel plates ; Surface defects ; Vision
  • É parte de: Signal, image and video processing, 2024-07, Vol.18 (5), p.4861-4874
  • Descrição: Surface defects of steel plates unavoidably exist during the industrial production proceeding due to the complex productive technologies and always exhibit some typical characteristics, such as irregular shape, random position, and various size. Therefore, detecting these surface defects with high performance is crucial for producing high-quality products in practice. In this paper, an improved network with high performance based on You Only Look Once version 5 (YOLOv5) is proposed for detecting surface defects of steel plates. Firstly, the Space to Depth Convolution (SPD-Conv) is utilized to make the feature information transforming from space to depth, helpful for preserving the entirety of discriminative feature information to the greatest extent under the proceeding of down-sampling. Subsequently, the coordinate attention mechanism is introduced and embedded into the bottleneck of C3 modules to effectively enhance the weights of some important feature channels, in favor of capturing more important feature information from different channels after SPD-Conv operations. Finally, the Spatial Pyramid Pooling Faster module is replaced by the Spatial Pyramid Pooling Fully Connected Spatial Pyramid Convolution module to further enhance the feature expression capability and efficiently realize the multi-scale feature fusion. The experimental results on NEU-DET dataset show that, compared with YOLOv5, the mAP and mAP50 dramatically increase from 51.7, 87.0 to 61.4, 92.6%, respectively. Meanwhile, the frame rate of 250 FPS implies that it still preserves a well real-time performance. Undoubtedly, the improved algorithm proposed in this paper exhibits outstanding performance, which may be also used to recognize the surface defects of aluminum plates, as well as plastic plates, armor plates and so on in the future.
  • Editor: London: Springer London
  • Idioma: Inglês
Voltar para lista de resultados
Ir para página anteriorAnterior Resultado  6 AvançarIr para próxima página

  • Realização: ABCD-USP USP   Logos de Redes Sociais: Freepik

Buscando em bases de dados remotas. Favor aguardar.