Article: Apple surface defect detection based on lightweight improved YOLOv5s Journal: International Journal of Information and Communication Technology (IJICT) 2024 Vol.24 No.7 pp.113 - 128 Abstract: Aiming at the current Apple surface defect detection algorithms with a large quantity of parameters, and poor real-time detection, a defect detection model to improve YOLOv5s is proposed. With the YOLOv5s model serving as a foundation, the EfficientNetv2 structure takes the role of the YOLOv5s model's backbone network. Second, by including the EMA attention mechanism in the neck component, the model's ability to extract important characteristics can be improved. Finally, using Alpha-IoU to optimise the IoU loss function can successfully raise the precision of the prediction box. The experimental findings demonstrate that the model size of the improved YOLOv5s model in this paper have been reduced by 20%, the recognition speed has been increased by 39.3%, and the mAP has been improved by 1.4%. In contrast to the initial model, the improved model has a smaller model size and a faster detection speed, while guaranteeing the detection accuracy. Inderscience Publishers - linking academia, business and industry through research

Title: Apple surface defect detection based on lightweight improved YOLOv5s

Authors: Lijun Lv; Yaermaimaiti Yilihamu; Yalin Ye

Addresses: School of Electrical Engineering, Xinjiang University, Urumqi, 830017, China ' School of Electrical Engineering, Xinjiang University, Urumqi, 830017, China ' School of Electrical Engineering, Xinjiang University, Urumqi, 830017, China

Abstract: Aiming at the current Apple surface defect detection algorithms with a large quantity of parameters, and poor real-time detection, a defect detection model to improve YOLOv5s is proposed. With the YOLOv5s model serving as a foundation, the EfficientNetv2 structure takes the role of the YOLOv5s model's backbone network. Second, by including the EMA attention mechanism in the neck component, the model's ability to extract important characteristics can be improved. Finally, using Alpha-IoU to optimise the IoU loss function can successfully raise the precision of the prediction box. The experimental findings demonstrate that the model size of the improved YOLOv5s model in this paper have been reduced by 20%, the recognition speed has been increased by 39.3%, and the mAP has been improved by 1.4%. In contrast to the initial model, the improved model has a smaller model size and a faster detection speed, while guaranteeing the detection accuracy.

Keywords: YOLOv5s; EfficientNetv2; EMA; defect detection.

DOI: 10.1504/IJICT.2024.139106

International Journal of Information and Communication Technology, 2024 Vol.24 No.7, pp.113 - 128

Received: 04 Mar 2024
Accepted: 07 May 2024
Published online: 13 Jun 2024 *

Title: Apple surface defect detection based on lightweight improved YOLOv5s

Keep up-to-date