摘要
随着时代发展,传统的图像识别算法在面对大量复杂图像时,其计算能力和效果方面存在一定的限制。鉴于此,研究首先分析了现有深度学习中的卷积神经网络,对其运算方面进行了改进。其次,加入了现场可编程门阵列加速器进行优化。最后在嵌入式系统实现了布局,并提出了一种新型图像识别模型。实验结果表明,研究提出的新模型识别准确率最高为93%,Loss值最低为0.4,且迭代时间最快为200次。其单张图片处理时间最短为11.5 ms,平均资源利用率为80.5%。该模型在花卉图像仿真测试中的整体识别率为83.3%,其中玫瑰类的识别率最高可达92.3%。综上所述,研究提出的新模型能够在嵌入式环境中高效地完成图像识别任务,同时为后续的技术研究提供准确、高效的图像识别解决方案。
With the development of the times,traditional image recognition algorithms have certain limitations in terms of their computational ability and effectiveness when facing a large number of complex images.In view of this,the study first analyzes the existing convolutional neural network in deep learning and improves its computational aspects.Secondly,a field programmable gate array gas pedal was added for optimization.Finally,the layout is implemented in an embedded system and a novel image recognition model is proposed.The experimental results show that the proposed new model has the highest recognition accuracy of 93%,the lowest Loss value of 0.4,and the fastest iteration time of 200 times.Its shortest processing time for a single image is 11.5 ms,and the average resource utilization is 80.5%.The overall recognition rate of this model in the flower image simulation test is 83.3%,with the highest recognition rate of 92.3%for the rose category.In summary,the new model proposed in the study is able to efficiently accomplish image recognition tasks in embedded environments,while providing accurate and efficient image recognition solutions for subsequent technological research.
作者
杨东红
YANG Donghong(Xi’an Jiaotong University City College,Xi’an 710018,China)
出处
《自动化与仪器仪表》
2024年第6期241-245,251,共6页
Automation & Instrumentation
基金
西安交通大学城市学院2023年度校级科研项目(2023Y01)。
关键词
卷积神经网络
可编程门阵列
图像识别
嵌入式系统
识别准确率
convolutional neural network
programmable gate array
image recognition
embedded system
recognition accuracy
