Marine Objects Detection Using Deep Learning on Embedded Edge Devices

Abstract

Intelligence techniques based on convolution neural networks (CNNs) are now dominant in the field of object detection and classification. The deployment of CNNs on embedded edge devices targeting real-time inference sets a challenge due to the limited computing resources and power budgets. Several optimization techniques such as pruning, quantization and use of light neural networks enable the realtime inference but at the cost of precision degradation. However, using efficient approaches to apply the optimization techniques at training and inference stages enable high inference speed with limited degradation of detection performance. In this paper, we revisit the problem of detecting and classifying maritime objects. We investigate different versions of the You Only Look Once (YOLO), a state-of-the-art deep neural network, for real-time object detection and compare their performance for the specific application of detecting maritime objects. The trained YOLO networks are efficiently optimized targeting three recent edge devices: Nvidia Jetson Xavier AGX, AMD-Xilinx Kria KV260 Vision AI Kit, and Movidius Myriad X VPU. The proposed deployments demonstrate promising results with an inference speed of 90 FPS and a limited degradation of 2.4% in mean average precision.

FPGA-Based Real-Time Object Detection and Classification System Using YOLO for Edge Computing

“…Corcoran et al [32] and Valadanzoj et al [34] denote elevated throughput albeit accompanied by increased power consumption and accuracy. Conversely, Heller et al [31] and Valadanzoj et al [34] showcase a harmonized trade-off among throughput, power consumption, and efficiency. For an equitable comparison, the proposed system is compared with the work of Heller et al [31], given the congruent utilization of FPGA boards and test image sizes in the experimentation.…”

“…However, to discern the state-of-the-art among these systems, several notable implementations warrant attention. Heller et al [31] introduced an object detection system utilizing deep learning on embedded edge devices, focusing on maritime object detection with the Xilinx Kria KV260 Vision AI Kit. Their study involved training and evaluating multiple YOLO neural networks of varying sizes and architectural specifications, incorporating structured pruning techniques such as sparsifying to reduce network size while preserving detection performance.…”

A review of intelligent ship marine object detection based on RGB camera

“…Secondly, pruning techniques can be employed to eliminate redundant parameters from the original model, thus achieving lightweight goals. For instance, Heller et al [133] utilized sparsifying approach to modify network models, resulting in a reduction of parameter count by over 80% and a simultaneous acceleration in computational speed. Subsequent researchers can devote efforts to both aspects or adopt a fusion approach to enhance real-time performance and reduce parameter count in maritime object detection.…”

Multimodal perception for obstacle detection for flying boats - Unmanned Surface Vehicle (USV)

“…In our previous article [14], we proposed a deep learningbased solution for detecting and classifying maritime objects using convolutional neural networks (CNNs). During our study, we evaluated the performance of several networks and selected YOLOv4 due to its high detection performance and rate.…”