A Lightweight Approach to Defect and Foreign Matter Detection in Children's Snacks Using YOLOv8 with Ghost and ECA Modules

Abstract

In this paper, we propose an efficient object detection approach to identify defects and foreign matters in children's snacks. To achieve both computational efficiency and high accuracy, the proposed model was enhanced with the Ghost module and Efficient Channel Attention (ECA) module. To reduce the parameter count and computational load, the Ghost module is applied to the backbone network. The ECA module compensates for potential accuracy losses by emphasizing important channel features. Our private dataset was used to evaluate the model. Experimental results demonstrate that our proposed model reduces computational complexity by 0.6 GFLOPs and decreases parameters by 279604 compared to the base YOLOv8 model. This leads to a minor accuracy improvement of 0.0117.