A Novel Quality Inspection Method For Components With Small Diameter And Deep Aperture

This paper investigates a novel quality inspection method for the components with small diameter and deep aperture. A method for extracting the minimum circumscribed rectangle of the target is proposed, which would provide additional shape and angle information and avoid unexpected overlaps and background disturbance. In order to detect the low contrast targets in complex background, a novel method based on Faster R-CNN with several improvements is adopted. In the framework, a five-dimensional (5-D) vector is designed to represent the minimum circumscribed rectangle followed by an improved anchor selection mechanism in the region proposal network. Furthermore, an improved POI pooling process is employed to extract feature maps. Comparison experiments fully demonstrate the superiority of the proposed inspection method over existing methods. Meanwhile, successful inspection results on challenging real-world images prove that the system is of practical significance to industrial applications.