[1]章佳琪,肖 建 *.DID-YOLO:一种适用于嵌入式设备的移动机器人目标检测算法[J].计算机技术与发展,2023,33(10):8-14.[doi:10. 3969 / j. issn. 1673-629X. 2023. 10. 002]
　ZHANG Jia-qi,XIAO Jian *.DID-YOLO:A Mobile Robot Target Detection Algorithm for Embedded Devices[J].,2023,33(10):8-14.[doi:10. 3969 / j. issn. 1673-629X. 2023. 10. 002]

点击复制

DID-YOLO:一种适用于嵌入式设备的移动机器人目标检测算法()

分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

33

期数:

2023年10期

页码:

8-14

栏目:

嵌入式计算

出版日期:

2023-10-10

文章信息/Info

Title:

DID-YOLO:A Mobile Robot Target Detection Algorithm for Embedded Devices

文章编号:

1673-629X(2023)10-0008-07

作者:

章佳琪¹ ; 肖 建^{2 *}

1. 南京邮电大学 电子与光学工程学院、柔性电子(未来技术)学院,江苏 南京 210046;

2. 南京邮电大学 集成电路科学与工程学院,江苏 南京 210046

Author(s):

ZHANG Jia-qi1 ; XIAO Jian2 *

1. School of Electronic and Optical Engineering & School of Flexible Electronics ( Future Technology) ,Nanjing University of Posts and Telecommunications,Nanjing 210046,China;

2. School of Integrated Circuit Science and Engineering,Nanjing University of Posts and Telecommunications,Nanjing 210046,China

关键词:

移动机器人; 目标检测; 嵌入式设备; 轻量化; 知识蒸馏

Keywords:

mobile robot; object detection; embedded device; lightweight; knowledge distillation

分类号:

TP249

DOI:

10. 3969 / j. issn. 1673-629X. 2023. 10. 002

摘要:

近些年来,目标检测算法在移动机器人环境感知领域表现出了突出的性能。 但是目标检测算法存在模型庞大和计算复杂的问题,制约了目标检测算法在移动嵌入式设备上的部署和发展。 YOLO 是一种单阶段的目标检测算法,具有较高的准确度和较快的运行速度。 该文提出了一种基于 YOLOv5s 改进后适用于嵌入式设备的移动机器人目

标检测算法DID-YOLO。 首先,使用深度可分离卷积和倒置残差模块对 YOLOv5s 的 backbone 网络进行重构,降低模型复杂度和计算量,达到轻量化的目的;其次,利用特征层和输出层结合的知识蒸馏训练提高重构后目标检测网络的精度。 在目标检测通用数据集 PASCAL VOC 上实验表明:DID-YOLO 模型尺寸为 3. 63 MB,相较原网络模型尺寸减小了 48. 65% ;经过特征层和输出层蒸馏后,DID-YOLO 的 mAP@ 0. 5 提升至 73. 83% ;DID-YOLO 在 Jetson AGX Xavier 上实现了每秒 31. 2 帧的实时图像处理速度。 提出的 DID-YOLO 性能显著,满足了移动机器人嵌入式平台的实时高精度检测需求。

Abstract:

In recent years, object detection algorithm has shown outstanding performance in the field of mobile robot environmentperception. However,the problem of huge?

model and complex calculation of target detection algorithm restricts the deployment and development of such algorithm on mobile embedded devices. YOLO is?

a single-stage target detection algorithm with high accuracy and fastrunning speed. We propose an improved DID-YOLO mobile robot target detection algorithm?

based on YOLOv5s,which is suitable forembedded devices. Firstly,deep separable convolution and inverted residual modules are used to reconstruct the backbone network ofYOLOv5s to reduce model complexity and computational load for lightweight. Secondly,knowledge distillation training which combinesthe feature layer and the output layer is used to improve the accuracy of the reconstructed target detection network. Experiments onPASCAL VOC show that the size of DID-YOLO model is 3. 63 MB,which is 48. 65% smaller than that of the original network model.After distillation of the feature layer and the output layer,mAP@ 0. 5 of DID-YOLO increased to 73. 83% . DID-YOLO achieved a real-time image processing speed of 31. 2 frames per second on Jetson AGX Xavier. The DID-YOLO proposed has remarkable performanceand can meet the requirements of real-time and high-precision detection of embedded mobile robot platforms.

常用功能