基于改进LeNet-5网络的堆芯燃料组件编码识别
Encoding recognition for core fuel assembly based on improved LeNet-5 model
吕伽奇 1丁帅 1庞静珠 1许小进2
作者信息
- 1. 东华大学机械工程学院,上海
- 2. 国核电站运行服务技术公司,上海
- 折叠
摘要
在核电站堆芯核燃料组件水下组装作业中,需要通过视觉技术进行组件编码的识别以便准确定位组件的安装位置.针对水下环境中弱光照等问题导致了图像质量的降低,本文通过乘方增强算法、OSTU算法、CLAHE算法和拉普拉斯变换的方法来实现堆芯燃料组件编码字符水下图像的增强.为了提高编码识别效果,提出了一种整合LeNet-5网络和支持向量机(SVM)的模型,在网络中添加BN(Batch Normalization)层与Dropout层来加速网络的运行速度,并改进Sigmoid函数,增加函数的平滑性,以此来减少梯度消失.实验表明,在自定义数据集上的验证准确率为99.82％,识别率为100％,相比于其他模型有显著的提升.
Abstract
In the underwater assembly of core fuel assemblies for nuclear power plants,visual technology is used to identify component codes to accurately locate the installation position of the nuclear fuel assemblies.In response to the problem of reduced image quality caused by weak lighting in underwater environments,the underwater image enhancement is achieved by means of multiplicative enhancement algorithm,OSTU algorithm,CLAHE algorithm and Laplace transform.To improve the performance of encoding recognition,a model that integrates LeNet-5 network and support vector machine(SVM)is proposed in this paper.BN(Batch Normalization)layer and Dropout layer are added to the network to accelerate the running speed of the network,and the Sigmoid function is improved to increase the smoothness of the function to reduce the gradient vanishing.Experiments show that the validation accuracy on the customized dataset is 99.82％and the recognition rate is 100％,which is a significant improvement compared to other models.
关键词
编码识别/图像处理/CLAHE算法/LeNet-5/支持向量机(SVM)Key words
encoding recognition/image processing/CLAHE algorithm/LeNet-5/support vector machine(SVM)引用本文复制引用
基金项目
国家科技重大专项(2019ZX06002001-003)
出版年
2024
NETL
NSTL
维普
万方数据