Authors: Zihao Huang; Hong Xiao; Tao Wang; Junhao Zhou
Addresses: Faculty of Computer, Guangdong University of Technology, Guangzhou, 510006, Guangdong, China ' Faculty of Computer, Guangdong University of Technology, Guangzhou, 510006, Guangdong, China ' Faculty of Automation, Guangdong University of Technology, Guangzhou, 510006, Guangdong, China ' Faculty of Information Technology, Macau University of Science and Technology, Macau, 999078, China
Abstract: There are many object detection algorithms that have performed well on public datasets, and they can be used in product defect detection. But there are still many details, which can affect the detection performance of actual product defect detection, need to be optimised. In this paper, we design a defect detector, called feature pair pyramid (FPP) detector, and optimise it for specific industrial application using three methods. Then we use FPP detector to detect defects of metal can products of an enterprise. The experimental results show that our FPP detector is more effective in detecting small size defects. The performance (AP@0.5) of our detector is better than current state-of-the-art detectors.
Keywords: defect detection; small object detection; feature pair pyramid; FPP; one-stage object detector; Resnet; k-means; anchor box.
International Journal of Modelling, Identification and Control, 2021 Vol.37 No.1, pp.10 - 21
Received: 07 May 2020
Accepted: 21 Dec 2020
Published online: 13 Nov 2021 *