A Method for Optimizing Deep Learning Object Detection in Edge Computing

Kim, Ryangsoo; Kim, Geonyong; Kim, Heedo; Yoon, Giha; Yoo, Hyung-Joun

doi:10.1109/ictc49870.2020.9289529

Cited by 10 publications

(7 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For the collected power vision image, it can complete the image analysis and calculation closer to the perception terminal. At present, there are two kinds of technical ideas of power edge intelligence, one is the model compression strategy through channel pruning and parameter reduction [19], and the other is the hardware-based model acceleration strategy [20]. The main methods of model compression strategy include network pruning, quantification, low-rank decomposition and knowledge distillation [21].…”

Section: Power Vision Edge Intelligencementioning

confidence: 99%

Identification of Key Links in Electric Power Operation Based-Spatiotemporal Mixing Convolution Neural Network

Feng¹,

Wang²,

Ma³

et al. 2023

Computer Systems Science and Engineering

View full text Add to dashboard Cite

As the scale of the power system continues to expand, the environment for power operations becomes more and more complex. Existing risk management and control methods for power operations can only set the same risk detection standard and conduct the risk detection for any scenario indiscriminately. Therefore, more reliable and accurate security control methods are urgently needed. In order to improve the accuracy and reliability of the operation risk management and control method, this paper proposes a method for identifying the key links in the whole process of electric power operation based on the spatiotemporal hybrid convolutional neural network. To provide early warning and control of targeted risks, first, the video stream is framed adaptively according to the pixel changes in the video stream. Then, the optimized MobileNet is used to extract the feature map of the video stream, which contains both time-series and static spatial scene information. The feature maps are combined and non-linearly mapped to realize the identification of dynamic operating scenes. Finally, training samples and test samples are produced by using the whole process image of a power company in Xinjiang as a case study, and the proposed algorithm is compared with the unimproved MobileNet. The experimental results demonstrated that the method proposed in this paper can accurately identify the type and start and end time of each operation link in the whole process of electric power operation, and has good real-time performance. The average accuracy of the algorithm can reach 87.8%, and the frame rate is 61 frames/s, which is of great significance for improving the reliability and accuracy of security control methods.

show abstract

Section: Power Vision Edge Intelligencementioning

confidence: 99%

Identification of Key Links in Electric Power Operation Based-Spatiotemporal Mixing Convolution Neural Network

Feng¹,

Wang²,

Ma³

et al. 2023

Computer Systems Science and Engineering

View full text Add to dashboard Cite

show abstract

“…In this section, we introduce an overview of the proposed object detection offloading framework that adopts an object detection-specific task-level pipeline parallelism proposed in our previous work [12] to improve the computing resource utilization on both mobile edge devices and an edge server. Afterward, we discuss preliminary experimental results that show possible problems arising in real-world object detection offloading scenarios.…”

Section: Proposed Object Detection Offloading Frameworkmentioning

confidence: 99%

“…Our DNN-based object detection offloading framework is designed to parallelize the workload execution by distributing heavy and lightweight workloads to the edge server and mobile edge devices simultaneously to improve the object detection performance. The proposed framework is adopted from our previous work [12], including the offloading DNN inference and NMS filtering workloads to a remote server. Figure 2 shows a block diagram that represents the proposed object detection offloading framework, which applies tasklevel pipeline parallelism to improve the computing resource utilization on both edge devices and a server.…”

Section: A Overview Of Object Detection Offloading Applying Task-level Pipeline Parallelismmentioning

confidence: 99%

“…The proposed framework is designed to execute computationally intensive workloads, including the DNN inference and non-maximum suppression (NMS) filtering on the edge server, while relatively lightweight workloads, including the resizing images and rendering detection results, are running on mobile edge devices. Furthermore, we apply the task-level pipeline parallelism presented in [12] to improve the computing resource utilization on both the edge devices and edge server, reducing the overall object detection latency.…”

Section: Introductionmentioning

confidence: 99%

“…• We implement an object detection offloading framework, which is designed to parallelize the workload execution by distributing heavy and lightweight workloads to the edge server and mobile edge devices simultaneously to improve the object detection performance, in a real-world WLAN environment. In addition, we apply object detection-specific task-level pipeline parallelism proposed in our previous work [12] to improve the computing resource utilization on both the mobile edge devices and edge server, reducing the overall object detection latency. • We derive an object detection offloading decision problem as a simple binary optimization problem and propose a greedy algorithm to find the optimal solution in polynomial time.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Implementing Practical DNN-Based Object Detection Offloading Decision for Maximizing Detection Performance of Mobile Edge Devices

et al. 2021

Self Cite

View full text Add to dashboard Cite

In the last decade, deep neural network (DNN)-based object detection technologies have received significant attention as a promising solution to implement a variety of image understanding and video analysis applications on mobile edge devices. However, the execution of computationally intensive DNN-based object detection workloads in mobile edge devices is insufficient in fulfilling the object detection requirements with high accuracy and low latency, owing to the limited computation capacity. In this paper, we implement and evaluate a DNN-based object detection offloading framework to improve the object detection performance of mobile edge devices by offloading computation-intensive workloads to a remote edge server. However, preliminary experimental results have shown that offloading all object detection workloads of mobile edge devices may lead to worse performance than executing the workloads locally. This degradation is obtained from the inefficient resource utilization in the edge computing architectures, both for the edge server and mobile edge devices. To resolve the aforementioned problem with degradation, we devise a device-aware DNN offloading decision algorithm that is aimed to maximize resource utilization in the edge computing architecture. The proposed algorithm decides whether or not to offload the object detection workloads of edge devices by considering their computing power and network bandwidth, and therefore maximizing their average object detection processing frames per second. Through various experiments conducted in a real-life wireless local area network (WLAN) environment, we verified the effectiveness of the proposed DNN-based object detection offloading framework.INDEX TERMS Deep learning offloading, object detection, wireless edge computing, resource optimization.

show abstract