The invention discloses a distributed learning method for edge end image recognition in intelligent traffic, and the method comprises the following steps: 1, a server selects a distributed learning task and a deep learning model in an intelligent traffic application scene; 2, the server carries out the model preprocessing; 3, the server starts a tth round of distributed model training process, step 5, the server selects a strategy based on the edge device, step 6, the edge end calculates a utility value delta k of the edge end according to information feedback in existing training, and sends the value delta k to the server, and step 7, the step 3 to the step 8 are repeated until model parameters converge. The method has the advantages that the joint training of the image recognition model with fine control granularity and high performance is realized; the number of participants can be reduced by adjusting the selection frequency of the edge end according to the model training effect, the energy cost is reduced, the selection frequency is dynamically adjusted, the participation of the edge end is reduced while the accuracy is kept, and the robustness of the system is effectively improved.
本发明公开了一面向智能交通中边缘端图像识别的分布式学习方法,包括如下步骤:步骤1:服务器选择智能交通应用场景下的分布式学习任务和深度学习模型:步骤2:服务器进行模型预处理:步骤3:服务器开始第t轮分布式模型训练过程,步骤5:服务器基于边缘设备选择策略,步骤6:根据已有训练中的信息反馈,边缘端计算自身效用值δk,发送至服务器:步骤7:重复步骤3至步骤8,直到模型参数收敛。本发明的优点是实现了细控制粒度、高性能的图像识别模型联合训练;根据模型训练效果调整边缘端选择频率能够减少参与者的数量,降低了能源成本,通过动态调整选择频率,在保持准确性的同时减少了边缘端的参与,有效提高了系统的鲁棒性。
Distributed learning method for edge end image recognition in intelligent traffic
一种面向智能交通中边缘端图像识别的分布式学习方法
2023-02-28
Patent
Electronic Resource
Chinese
IPC: | G06N COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS , Rechnersysteme, basierend auf spezifischen Rechenmodellen / G08G Anlagen zur Steuerung, Regelung oder Überwachung des Verkehrs , TRAFFIC CONTROL SYSTEMS / G06F ELECTRIC DIGITAL DATA PROCESSING , Elektrische digitale Datenverarbeitung |
INTELLIGENT EDGE-BASED ROAD TRAFFIC SITUATION RECOGNITION METHOD
European Patent Office | 2021
|Intelligent height locating traffic signal device with image recognition function
European Patent Office | 2020
|