Leather Defect Classification in Footwear Manufacturing Industries

Abstract

Leather is a natural and durable material derived from tanning animal skins and hides. The value of leather is contingent upon its quality and surface condition, making it imperative to detect defects accurately. Traditionally, detecting leather defects has been labor-intensive and time-consuming, prone to human error and eye strain. This paper proposes a fully automatic defect detection system that employs the EfficientNetB0 neural network on the leather to provide defect-free leather. The system can classify leather patches into different classes, such as folding marks, grain off, growth marks, loose grain, non-defective, and pinhole. This proposed work suggests this defect identification software system can be employed with a few mechanical components such as fixing the camera in a conveyor system to identify the defects of moving leather pieces. This research contributes a practical and efficient solution for enhancing quality control processes in footwear manufacturing industries, ultimately leading to streamlined production and improved product quality.