BiSPD-YOLO: Surface Defect Detection Method for Small Features and Low-resolution Images

Yan, Shaohua; Chen, G. L.; Gao, Ao; Liu, Chao; Xiong, Zhenhua

doi:10.1109/aim46323.2023.10196194

Cited by 2 publications

(1 citation statement)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Compared to convolution with a stride of 1, YOLOv8's downsampling operation reduces the utilization of some features, which mainly exists in the computation when the convolution kernel wraps to the next line. To address this issue, we propose adding SPD convolution [10] as the downsampling module and changing the cross-row convolution in YOLOv8 to non-cross-row convolution. Figure 2 depicts the principle of SPD convolution.…”

Section: Spd Convolution Computationmentioning

confidence: 99%

Ship target detection based on CBAM-YOLOv8

Zhang

2024

International Conference on Mechatronic Engineering and Artificial Intelligence (MEAI 2023)

View full text Add to dashboard Cite

Ship target detection at sea has important strategic significance in military activities, maritime security, and other aspects. Traditional image processing algorithms struggle to capture the various scales of ship features. In this paper, we propose an algorithm for ship target detection based on CBAM-YOLOv8. Firstly, spatial-to-depth convolution is used in the model's downsampling section instead of cross-stride convolution to improve feature utilization. Secondly, the CBAM (Convolutional Block Attention Module) attention mechanism is added to the deep layers of the model to fuse spatial and channel feature information. Finally, MPDIOU is used to replace the CIOU loss function, enhancing the extraction accuracy of detection boxes. Experimental results on a maritime target dataset show that the detection algorithm achieves a mAP value of 93.16% and a detection speed of approximately 134 FPS, meeting the requirements of real-time ship detection at sea and providing an effective technical reference for various maritime activities and tasks.

show abstract

Section: Spd Convolution Computationmentioning

confidence: 99%