Redundant object detection method for civil aircraft assembly based on machine vision and smart glasses

Abstract

Aiming at the problems of slow detection speed and low accuracy of redundant objects in the assembly process of civil aircraft, a redundant object detection method based on vision and augmented reality smart glasses is proposed. This method uses smart glasses as the hardware device, and takes the image collected by the camera as the input of the detection system, and proposes the FPN-CenterNet machine vision model based on the CenterNet detection head to detect redundant objects. The multi-scale feature fusion method is used to solve the problem of low detection accuracy of small-scale redundant targets, and a loss function with dynamic weights is designed to solve the problem of unbalanced proportion of large and small targets in training samples. The proposed network model is validated on the PASCAL VOC public dataset and the self-built redundant object dataset. This method can detect a large number of redundant objects within the visible range of smart glasses within 200ms. The research results provide a new reference for the quality process management of civil aircraft assembly.

A lightweight high-resolution algorithm based on deep learning for layer-wise defect detection in laser powder bed fusion

“…Deep learning-based object detection algorithms, capable of recognizing and locating multiple targets in images, have found widespread application across various industries [40][41][42]. Structurally, object detection can be divided into twostage algorithms, represented by Faster R-CNN [43], and onestage algorithms, represented by SSD [44] and you only look once (YOLO) [45].…”