A backbone-edge feature extraction method for varied industrial parts. (11th March 2023)

• Record Type: Journal Article
• Authors: Wan, Ping; Guo, Ling; Li, Ming; Gao, Min; Zhang, Hongping; Li, Jie
• Is Part Of: International journal of system control and information processing. Volume 4:Number 1(2022)
• Page Start: 28
• Page End: 42
• Publication Date: 2023-03-11
• Subjects: edge detection -- backbone feature -- neighbour information -- CNN -- convolutional neural networks -- industrial parts -- pixel neighbour -- feature combination -- adhesion processing -- multiple scale -- edge details -- edge enhancement; System design -- Data processing -- Periodicals; Information technology -- Periodicals; 003.5
• Languages: English
• ISSNs: 1759-9334

Abstract

Empty-loading ratio (ELR) in manufacturing factories is challenged due to plenty of industrial parts with diverse sizes and shapes, where edge detection is a primary operation to obtain ELR. Unlike traditional edge detection tasks, the purpose of ELR is to gain backbones of industrial parts and filter the details. Therefore, we present a backbone-edge feature extraction approach to deal with ELR computation problem. There into, a multi-scale block CNN model is structured to learn primary information of industrial parts through hybrid operations (i.e., horizontal and vertical combinations) with both deep and shallow features. In the model, a neighbour-information-based loss function is designed to enhance backbone information. Furthermore, a ELR value is obtained through discovering minimised closed-regions based on backbone edge information from our model. Simulations on industrial parts in conveyor boxes from real world indicate that the proposed approach outperforms other state-of-the-art methods.