摘要
在非法占用公交车道违规车辆等领域的边缘计算与识别中,针对基于深度卷积神经网络的图像物体分类算法模型算力需求大与边缘设备部署后有限资源的突出矛盾,如何设计边缘计算设备的加速单元以保证分类算法的精度与实时性具有重要意义。针对上述问题,提出一种基于深度卷积神经网络的公交分类算法,该方法在现场可编程逻辑门阵列上实现了公交车图像分类算法的加速。通过基于迁移学习方法对ResNet50预训练模型进行微调,采用嵌入式端的推理加速实现对模型的推理,并对FPGA加速方案进行推理部署实现。结果表明,该算法具有硬件配置灵活、信息处理加速快的优点,这为实现神经网络在嵌入式平台的高效、高速应用提供了有效解决方案。
In view of the prominent contradiction between the large demand for computing power of the image object classification algorithm model based on deep convolutional neural network(DCNN)and the limited resources after the deployment of edge devices,it is of great significance to design the acceleration unit of edge computing equipment to ensure the accuracy and real-time performance of the classification algorithm in the field of edge computing and recognition of illegal occupation of bus lanes and illegal vehicles.Focused on the above,a bus classification algorithm based on DCNN,which realizes the acceleration of bus image classification algorithm on FPGA,is proposed.The ResNet50 pre-trained model is fine-tuned based on the transfer learning method.The inference acceleration on the embedded side is used to realize the inference of the model,and the inference deployment implementation of the FPGA acceleration scheme is implemented.The results show that the proposed algorithm has flexible hardware configuration and fast information processing acceleration,which provides an effective solution for the efficient and high-speed application of neural networks in embedded platforms.
作者
黄佳美
张伟彬
熊官送
HUANG Jiamei;ZHANG Weibin;XIONG Guansong(Beijing Institute of Automation Control Equipment,Beijing 100074,China)
出处
《现代电子技术》
北大核心
2024年第7期140-144,共5页
Modern Electronics Technique