摘要
针对支持向量机算法在对天气雷达回波图像出现异常的识别中存在识别率低、时效性不高、对原始图像特征提取较复杂等问题,因此该文采用基于深度学习技术中的卷积神经网络(CNN)算法对雷达回波图像进行异常判别。研究中结合了传统经典的LeNet-5网络和在图像处理中比较有优势的VGG网络,将VGG网络中的小卷积核级联思路代入了LeNet-5浅层网络;每次卷积都对权重L2正则化处理,并且每层都使用了ReLU激活函数,大大增加了系统的泛化能力;在全连接层中添加了Dropout层,很好地解决了过拟合。实验证明,此算法对雷达回波图像识别率高、时效快,较传统识别方法具有明显优势。
Aiming at the problems of low recognition rate,low timeliness and the complex extraction of original images due to the recognition of the anomalous weather radar echo images by Support Vector Machine(SVM)algorithm,a method is proposed though the Convolutional Neural Network(CNN)algorithm for the anomalous recognition of the radar echo images in this paper.The combination of the traditional LetNet-5 network and the advantages of the VGG network dealing with the pictures,the cascade idea of small convolution kernel in VGG network is applied in LeNet-5 shallow network.Each convolution is regularized with the weight L2,and each layer uses the ReLU activity function,which greatly increases the generalization ability of the system.Adding the Dropout layer to the full connection layer is to effectively solute the overfitting problem.Thought experimental results,the algorithm for the recognition of radar echo images has the advantages of high recognition rate and fast time efficiency compared with the traditional recognition method.
作者
刘昉
李奇临
蒋涌
杨永毅
张俊
赵思亮
LIU Fang;LI Qilin;JIANG Yong;YANG Yongyi;ZHANG Jun;ZHAO Siliang(Chongqing Meteorological Information Center,Chongqing Meteorological Service,Chongqing 401147,China)
出处
《电子设计工程》
2021年第6期74-78,87,共6页
Electronic Design Engineering
基金
重庆市气象部门智慧气象技术创新团队项目(ZHCXTD-201922)。
