Intellectualization and unmanned manufacturing have been an inevitable trend in industrial development. The landing of intelligent applications is one of the current challenges in the industry. Due to the hierarchical...
详细信息
Intellectualization and unmanned manufacturing have been an inevitable trend in industrial development. The landing of intelligent applications is one of the current challenges in the industry. Due to the hierarchical architecture of the industrial automation pyramid, traditional programmable logic controllers (PLCs) that are usually employed in the field cannot cooperate with artificial intelligence (AI) algorithms that require massive data and computing resources. Therefore, it is necessary to research the virtualization of traditional PLCs as dockers, which can be deployed in the cloud, edge, or field. Cloud PLCs can be easily integrated with AI, big data, and cloud computing to achieve intelligent decision-making and control and break down data islands. The visual sorting system has attracted increasing attention for its ability to accurately detect the position of objects. Many deep learning–based methods have achieved remarkable performance in computer vision. Additionally, the requirement of a network is fundamental for guaranteeing data transmission with low latency and high reliability. The combination of 5G and time-sensitive networking (TSN) can achieve the deterministic transmission of several industrial applications. According to the above challenges, joint control between cloud PLCs of low-level devices and visual sorting systems in a reliable network is critical and has industry potential. In this study, we propose a deep learning –based material recognition and location system with a cloud PLC, which is demonstrated in a 5G-TSN network. First, traditional PLC is virtualized to realize flexible PLC function deployment in the field and cloud. Second, we establish a cloud-based AI platform and design a You only look once v5 (YOLOv5)-based object detection algorithm to locate the position and recognize the types of materials to obtain pixel coordinates. Third, the camera calibration method is used to transform pixel and world coordinates, and the materia
With the increasing global demand for clean energy and sustainable development, hydrogen energy has gained significant attention as an environmentally friendly energy carrier. Hydrogen refueling stations, serving as c...
详细信息
目的可见光—红外跨模态行人再识别旨在匹配具有相同行人身份的可见光图像和红外图像。现有方法主要采用模态共享特征学习或模态转换来缩小模态间的差异,前者通常只关注全局或局部特征表示,后者则存在生成模态不可靠的问题。事实上,轮廓具有一定的跨模态不变性,同时也是一种相对可靠的行人识别线索。为了有效利用轮廓信息减少模态间差异,本文将轮廓作为辅助模态,提出了一种轮廓引导的双粒度特征融合网络,用于跨模态行人再识别。方法在全局粒度上,通过行人图像到轮廓图像的融合,用于增强轮廓的全局特征表达,得到轮廓增广特征。在局部粒度上,通过轮廓增广特征和基于部件的局部特征的融合,用于联合全局特征和局部特征,得到融合后的图像表达。结果在可见光—红外跨模态行人再识别的两个公开数据集对模型进行评估,结果优于一些代表性方法。在SYSU-MM01(Sun Yat-sen University multiple modality 01)数据集上,本文方法rank-1准确率和平均精度均值(mean average precision,mAP)分别为62.42%和58.14%。在RegDB(Dongguk body-based person recognition database)数据集上,本文方法rank-1和mAP分别为84.42%和77.82%。结论本文将轮廓信息引入跨模态行人再识别,提出一种轮廓引导的双粒度特征融合网络,在全局粒度和局部粒度上进行特征融合,从而学习具有判别性的特征,性能超过了近年来一些具有代表性的方法,验证了轮廓线索及其使用方法的有效性。
暂无评论