摘要
目前图像识别是计算机领域的重要任务,其目标是自动定位和识别图像中特定目标的位置。但目前主流的基于深度学习的目标检测算法计算量较大、模型参数量较高,导致在资源受限的嵌入式平台上部署过程繁琐、检测速度较低。该文针对以上问题,提出基于YOLOv5-Lite搭建图像识别系统,该算法是基于YOLOv5的轻量化改进。改进后的模型在保持一定检测精度的同时,体积仅为原模型的24.11%,参数量仅为原模型的22.68%,计算复杂度大幅下降,在RK3566开发板上的实时推理速度达到8.67 FPS,对嵌入式平台部署深度学习算法具有重要的研究意义。
At present,image recognition is an important task in the field of computer science,with the goal of automatically locating and recognizing the position of specific targets in images.However,the current mainstream target detection algorithms based on deep learning have a large amount of computation and a high amount of model parameters,resulting in cumbersome deployment processes and low detection speeds on resource-limited embedded platforms.Aiming at the above problems,this paper proposes to build an image recognition system based on YOLOv5-Lite.This algorithm is a lightweight improvement based on YOLOv5.While maintaining a certain detection accuracy,the improved model is only 24.11%of that of the original model,and the parameter quantity is only 22.68%of that of the original model.The computing complexity is greatly reduced.The real-time reasoning speed on the RK3566 development board reaches 8.67 FPS,which is of great research significance for deploying deep learning algorithms on embedded platforms.
出处
《科技创新与应用》
2025年第22期6-10,15,共6页
Technology Innovation and Application
