Light-YOLOv4: An Edge-Device Oriented Target Detection Method for Remote Sensing Images

Most deep-learning-based target detection methods have high computational complexity and memory consumption, and they are difficult to deploy on edge devices with limited computing resources and memory. To tackle this problem, this article proposes to learn a lightweight detector named Light-YOLOv4,...

Descripción completa

Guardado en:

Detalles Bibliográficos
Autores principales:	Xiaojie Ma, Kefeng Ji, Boli Xiong, Linbin Zhang, Sijia Feng, Gangyao Kuang
Formato:	article
Lenguaje:	EN
Publicado:	IEEE 2021
Materias:	Edge device model compression NVIDIA Jetson TX2 remote sensing target detection YOLOv4 Ocean engineering TC1501-1800 Geophysics. Cosmic physics QC801-809
Acceso en línea:	https://doaj.org/article/1baa97f168b744baa380b29a853ff08d
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

id	oai:doaj.org-article:1baa97f168b744baa380b29a853ff08d
record_format	dspace
spelling	oai:doaj.org-article:1baa97f168b744baa380b29a853ff08d2021-11-09T00:00:15ZLight-YOLOv4: An Edge-Device Oriented Target Detection Method for Remote Sensing Images2151-153510.1109/JSTARS.2021.3120009https://doaj.org/article/1baa97f168b744baa380b29a853ff08d2021-01-01T00:00:00Zhttps://ieeexplore.ieee.org/document/9573455/https://doaj.org/toc/2151-1535Most deep-learning-based target detection methods have high computational complexity and memory consumption, and they are difficult to deploy on edge devices with limited computing resources and memory. To tackle this problem, this article proposes to learn a lightweight detector named Light-YOLOv4, and it is obtained from YOLOv4 through model compression. To this end, first, we perform sparsity training by applying L1 regularization to the channel scaling factors, so the less important channels and layers can be found. Second, channel pruning and layer pruning are enforced on the network to prune the less important parts, which could significantly reduce network's width and depth. Third, the pruned model is retrained with a knowledge distillation method to improve the detection accuracy. Fourth, the model is quantized from FP32 to FP16, and it could further accelerate the model with almost no loss of detection accuracy. Finally, to evaluate Light-YOLOv4's performance on edge devices, Light-YOLOv4 is deployed on NVIDIA Jetson TX2. Experiments on the SAR ship detection dataset (SSDD) demonstrate that the model size, parameter size, and FLOPs of Light-YOLOv4 have been reduced by 98.63%, 98.66%, and 91.30% compared with YOLOv4, and the detection speed has been increased to 4.2×. While the detection accuracy of Light-YOLOv4 is only slightly reduced, for example, the mAP has only reduced by 0.013. Besides, experiments on the Gaofen Airplane dataset also prove the feasibility of Light-YOLOv4. Moreover, compared with other state-of-the-art methods, such as SSD and FPN, Light-YOLOv4 is more suitable for edge devices.Xiaojie MaKefeng JiBoli XiongLinbin ZhangSijia FengGangyao KuangIEEEarticleEdge devicemodel compressionNVIDIA Jetson TX2remote sensingtarget detectionYOLOv4Ocean engineeringTC1501-1800Geophysics. Cosmic physicsQC801-809ENIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, Vol 14, Pp 10808-10820 (2021)
institution	DOAJ
collection	DOAJ
language	EN
topic	Edge device model compression NVIDIA Jetson TX2 remote sensing target detection YOLOv4 Ocean engineering TC1501-1800 Geophysics. Cosmic physics QC801-809
spellingShingle	Edge device model compression NVIDIA Jetson TX2 remote sensing target detection YOLOv4 Ocean engineering TC1501-1800 Geophysics. Cosmic physics QC801-809 Xiaojie Ma Kefeng Ji Boli Xiong Linbin Zhang Sijia Feng Gangyao Kuang Light-YOLOv4: An Edge-Device Oriented Target Detection Method for Remote Sensing Images
description	Most deep-learning-based target detection methods have high computational complexity and memory consumption, and they are difficult to deploy on edge devices with limited computing resources and memory. To tackle this problem, this article proposes to learn a lightweight detector named Light-YOLOv4, and it is obtained from YOLOv4 through model compression. To this end, first, we perform sparsity training by applying L1 regularization to the channel scaling factors, so the less important channels and layers can be found. Second, channel pruning and layer pruning are enforced on the network to prune the less important parts, which could significantly reduce network's width and depth. Third, the pruned model is retrained with a knowledge distillation method to improve the detection accuracy. Fourth, the model is quantized from FP32 to FP16, and it could further accelerate the model with almost no loss of detection accuracy. Finally, to evaluate Light-YOLOv4's performance on edge devices, Light-YOLOv4 is deployed on NVIDIA Jetson TX2. Experiments on the SAR ship detection dataset (SSDD) demonstrate that the model size, parameter size, and FLOPs of Light-YOLOv4 have been reduced by 98.63%, 98.66%, and 91.30% compared with YOLOv4, and the detection speed has been increased to 4.2×. While the detection accuracy of Light-YOLOv4 is only slightly reduced, for example, the mAP has only reduced by 0.013. Besides, experiments on the Gaofen Airplane dataset also prove the feasibility of Light-YOLOv4. Moreover, compared with other state-of-the-art methods, such as SSD and FPN, Light-YOLOv4 is more suitable for edge devices.
format	article
author	Xiaojie Ma Kefeng Ji Boli Xiong Linbin Zhang Sijia Feng Gangyao Kuang
author_facet	Xiaojie Ma Kefeng Ji Boli Xiong Linbin Zhang Sijia Feng Gangyao Kuang
author_sort	Xiaojie Ma
title	Light-YOLOv4: An Edge-Device Oriented Target Detection Method for Remote Sensing Images
title_short	Light-YOLOv4: An Edge-Device Oriented Target Detection Method for Remote Sensing Images
title_full	Light-YOLOv4: An Edge-Device Oriented Target Detection Method for Remote Sensing Images
title_fullStr	Light-YOLOv4: An Edge-Device Oriented Target Detection Method for Remote Sensing Images
title_full_unstemmed	Light-YOLOv4: An Edge-Device Oriented Target Detection Method for Remote Sensing Images
title_sort	light-yolov4: an edge-device oriented target detection method for remote sensing images
publisher	IEEE
publishDate	2021
url	https://doaj.org/article/1baa97f168b744baa380b29a853ff08d
work_keys_str_mv	AT xiaojiema lightyolov4anedgedeviceorientedtargetdetectionmethodforremotesensingimages AT kefengji lightyolov4anedgedeviceorientedtargetdetectionmethodforremotesensingimages AT bolixiong lightyolov4anedgedeviceorientedtargetdetectionmethodforremotesensingimages AT linbinzhang lightyolov4anedgedeviceorientedtargetdetectionmethodforremotesensingimages AT sijiafeng lightyolov4anedgedeviceorientedtargetdetectionmethodforremotesensingimages AT gangyaokuang lightyolov4anedgedeviceorientedtargetdetectionmethodforremotesensingimages
_version_	1718441366254518272

Light-YOLOv4: An Edge-Device Oriented Target Detection Method for Remote Sensing Images

Ejemplares similares