Wang Jiahao
College of Art and Design, Xi'an Mingde Institute of Technology, Xi'an, China.
PeerJ Comput Sci. 2023 Jul 10;9:e1451. doi: 10.7717/peerj-cs.1451. eCollection 2023.
The integration of image segmentation technology into packaging style design significantly amplifies both the aesthetic allure and practical utility of product packaging design. However, the conventional image segmentation algorithm necessitates a substantial amount of time for image analysis, rendering it susceptible to the loss of vital image features and yielding unsatisfactory segmentation results. Therefore, this study introduces a novel segmentation network, G-Lite-DeepLabV3+, which is seamlessly incorporated into cyber-physical systems (CPS) to enhance the accuracy and efficiency of product packaging image segmentation. In this research, the feature extraction network of DeepLabV3 is replaced with Mobilenetv2, integrating group convolution and attention mechanisms to proficiently process intricate semantic features and improve the network's responsiveness to valuable characteristics. These adaptations are then deployed within CPS, allowing the G-Lite-DeepLabV3+ network to be seamlessly integrated into the image processing module within CPS. This integration facilitates remote and real-time segmentation of product packaging images in a virtual environment.Experimental findings demonstrate that the G-Lite-DeepLabV3+ network excels at segmenting diverse graphical elements within product packaging images. Compared to the original DeepLabV3+ network, the intersection over union (IoU) metric shows a remarkable increase of 3.1%, while the mean pixel accuracy (mPA) exhibits an impressive improvement of 6.2%. Additionally, the frames per second (FPS) metric experiences a significant boost of 22.1%. When deployed within CPS, the network successfully accomplishes product packaging image segmentation tasks with enhanced efficiency, while maintaining high levels of segmentation accuracy.
将图像分割技术融入包装风格设计，显著提升了产品包装设计的美学吸引力和实用功能。然而，传统的图像分割算法在图像分析时需要大量时间，容易导致重要图像特征丢失，分割结果不尽人意。因此，本研究引入了一种新型分割网络G-Lite-DeepLabV3+，它被无缝集成到信息物理系统（CPS）中，以提高产品包装图像分割的准确性和效率。在本研究中，用Mobilenetv2替换了DeepLabV3的特征提取网络，集成了分组卷积和注意力机制，以有效处理复杂的语义特征，并提高网络对重要特征的响应能力。这些改进随后部署在CPS中，使G-Lite-DeepLabV3+网络能够无缝集成到CPS中的图像处理模块中。这种集成便于在虚拟环境中对产品包装图像进行远程实时分割。实验结果表明，G-Lite-DeepLabV3+网络在分割产品包装图像中的各种图形元素方面表现出色。与原始的DeepLabV3+网络相比，交并比（IoU）指标显著提高了3.1%，平均像素精度（mPA）提高了6.2%，令人印象深刻。此外，每秒帧数（FPS）指标大幅提高了22.1%。当部署在CPS中时，该网络成功地以更高的效率完成了产品包装图像分割任务，同时保持了较高的分割精度。
