ABSTRACT Conventional methods for detecting defects in flexible packaging printing often struggle with suboptimal accuracy and slow processing speeds, a deep learning‐based flexible packaging printing defect detection method is proposed, and a model based on the improved YOLOv5 deep learning algorithm is constructed. The new algorithm adds a parameter‐free attention mechanism SimAM to the backbone network, which enhances the suppression of irrelevant background information in images and improves the extraction of small target features. Second, using an improved FasterNet network with a Partial Convolution (PConv) core structure as the backbone network of the model reduces the number of parameters in the algorithm model and reduces computational redundancy. Lastly, the model convergence speed and regression accuracy are improved by the introduction of the enhanced loss function ECIoU. The experimental results prove that the improved YOLO‐FasteretSimAM‐ECIoU (YOLO‐FSE) algorithm achieves a detection accuracy of 99. 0%, which is 2. 2% higher than that of the YOLOv5 algorithm, and the FPS is 119, which doubles the detection speed, and the size of the trained weights is only 12. 0 MB, which makes it more suitable for deploying in embedded systems, mobile devices and other computing resources are limited.
Zhu et al. (Fri,) studied this question.