Wavelet, a powerful tool for signal processing, can be used to approximate the target func-tion. For enhancing the sparse property of wavelet approximation, a new algorithm was proposed by using wavelet kernel Support...Wavelet, a powerful tool for signal processing, can be used to approximate the target func-tion. For enhancing the sparse property of wavelet approximation, a new algorithm was proposed by using wavelet kernel Support Vector Machines (SVM), which can converge to minimum error with bet-ter sparsity. Here, wavelet functions would be firstly used to construct the admitted kernel for SVM according to Mercy theory; then new SVM with this kernel can be used to approximate the target fun-citon with better sparsity than wavelet approxiamtion itself. The results obtained by our simulation ex-periment show the feasibility and validity of wavelet kernel support vector machines.展开更多
A new filtering method for SAR data de-noising using wavelet support vector regression （WSVR） is developed. On the basis of the grey scale distribution character of SAR imagery, the logarithmic SAR image as a noise ...A new filtering method for SAR data de-noising using wavelet support vector regression （WSVR） is developed. On the basis of the grey scale distribution character of SAR imagery, the logarithmic SAR image as a noise polluted signal is taken and the noise model assumption in logarithmic domain with Gaussian noise and impact noise is proposed. Based on the better per- formance of support vector regression （SVR） for complex signal approximation and the wavelet for signal detail expression, the wavelet kernel function is chosen as support vector kernel func- tion. Then the logarithmic SAR image is regressed with WSVR. Furthermore the regression distance is used as a judgment index of the noise type. According to the judgment of noise type every pixel can be adaptively de-noised with different filters. Through an approximation experiment for a one-dimensional complex signal, the feasibility of SAR data regression based on WSVR is con- firmed. Afterward the SAR image is treated as a two-dimensional continuous signal and filtered by an SVR with wavelet kernel function. The results show that the method proposed here reduces the radar speckle noise effectively while maintaining edge features and details well.展开更多
为提高月径流时间序列预测精度,改进最小二乘孪生支持向量回归机(Least Squares Twin Support Vector Regression,LSTSVR)性能,首次基于高斯核函数、多项式核函数、线性核函数构建混合核函数,提出4种季节趋势分解(Seasonal and Trend de...为提高月径流时间序列预测精度,改进最小二乘孪生支持向量回归机(Least Squares Twin Support Vector Regression,LSTSVR)性能,首次基于高斯核函数、多项式核函数、线性核函数构建混合核函数,提出4种季节趋势分解(Seasonal and Trend decomposition using Loess,STL)-小波包变换(Wavelet Packet Transform,WPT)-裂狐优化(Rüppell's Fox Optimizer,RFO)算法-混合核最小二乘孪生支持向量回归机(Hybrid Kernel Least Squares Twin Support Vector Regression,HLSTSVR)模型,并构建STL-WPT-RFO-LSTSVR、STL-WPT-RFO-混合核最小二乘支持向量回归机(Hybrid Kerllel Least Squares Twin Suppart Vector Regression,HLSSVR)、STL-WPT-RFO-最小二乘支持向量回归机(Least Squares Support Vector Regression,LSSVR)等17种对比分析模型,通过云南省高桥、凤屯水文站月径流时间序列预测实例对21种模型进行验证。首先利用STL-WPT二次分解技术对月径流序列进行分解处理,合理划分训练集和验证集;然后基于高斯核函数、多项式核函数、线性核函数,采用“三三”线性组合和“两两”线性组合的方式构建4种混合核函数对月径流分解分量进行空间映射;最后利用RFO寻优HLSTSVR/LSTSVR/HLSSVR/LSSVR最佳超参数,利用最佳超参数建立21种模型对实例月径流序列各分解分量进行训练、预测和重构。结果表明:①4种STL-WPT-RFO-HLSTSVR模型能适应不同尺度的月径流数据分布,具有较好的模型性能和较小的预测误差,其中STL-WPT-RFO-HLSTSVR(高斯+多项式+线性)模型对高桥、凤屯站月径流预测的平均绝对百分比误差MAPE分别为2.85%、2.19%,决定系数R2均为0.9994,预测精度最高、效果最好;②混合核函数兼顾了不同核函数优势,能在模型复杂度与泛化能力之间取得平衡,显著提升模型性能和预测精度;③STL-WPT二次分解技术能有效解决复杂时间序列的非平稳性、非线性和多尺度特征,较STL更具分解优势;④组合模型融合了STL-WPT、RFO和HLSTSVR优点,具有较好的普适性和参考价值。展开更多
文摘Wavelet, a powerful tool for signal processing, can be used to approximate the target func-tion. For enhancing the sparse property of wavelet approximation, a new algorithm was proposed by using wavelet kernel Support Vector Machines (SVM), which can converge to minimum error with bet-ter sparsity. Here, wavelet functions would be firstly used to construct the admitted kernel for SVM according to Mercy theory; then new SVM with this kernel can be used to approximate the target fun-citon with better sparsity than wavelet approxiamtion itself. The results obtained by our simulation ex-periment show the feasibility and validity of wavelet kernel support vector machines.
基金supported by Shanghai Science and Technology Commission Innovation Action Plan(08DZ1205708)
文摘A new filtering method for SAR data de-noising using wavelet support vector regression （WSVR） is developed. On the basis of the grey scale distribution character of SAR imagery, the logarithmic SAR image as a noise polluted signal is taken and the noise model assumption in logarithmic domain with Gaussian noise and impact noise is proposed. Based on the better per- formance of support vector regression （SVR） for complex signal approximation and the wavelet for signal detail expression, the wavelet kernel function is chosen as support vector kernel func- tion. Then the logarithmic SAR image is regressed with WSVR. Furthermore the regression distance is used as a judgment index of the noise type. According to the judgment of noise type every pixel can be adaptively de-noised with different filters. Through an approximation experiment for a one-dimensional complex signal, the feasibility of SAR data regression based on WSVR is con- firmed. Afterward the SAR image is treated as a two-dimensional continuous signal and filtered by an SVR with wavelet kernel function. The results show that the method proposed here reduces the radar speckle noise effectively while maintaining edge features and details well.
文摘为提高月径流时间序列预测精度,改进最小二乘孪生支持向量回归机(Least Squares Twin Support Vector Regression,LSTSVR)性能,首次基于高斯核函数、多项式核函数、线性核函数构建混合核函数,提出4种季节趋势分解(Seasonal and Trend decomposition using Loess,STL)-小波包变换(Wavelet Packet Transform,WPT)-裂狐优化(Rüppell's Fox Optimizer,RFO)算法-混合核最小二乘孪生支持向量回归机(Hybrid Kernel Least Squares Twin Support Vector Regression,HLSTSVR)模型,并构建STL-WPT-RFO-LSTSVR、STL-WPT-RFO-混合核最小二乘支持向量回归机(Hybrid Kerllel Least Squares Twin Suppart Vector Regression,HLSSVR)、STL-WPT-RFO-最小二乘支持向量回归机(Least Squares Support Vector Regression,LSSVR)等17种对比分析模型,通过云南省高桥、凤屯水文站月径流时间序列预测实例对21种模型进行验证。首先利用STL-WPT二次分解技术对月径流序列进行分解处理,合理划分训练集和验证集;然后基于高斯核函数、多项式核函数、线性核函数,采用“三三”线性组合和“两两”线性组合的方式构建4种混合核函数对月径流分解分量进行空间映射;最后利用RFO寻优HLSTSVR/LSTSVR/HLSSVR/LSSVR最佳超参数,利用最佳超参数建立21种模型对实例月径流序列各分解分量进行训练、预测和重构。结果表明:①4种STL-WPT-RFO-HLSTSVR模型能适应不同尺度的月径流数据分布,具有较好的模型性能和较小的预测误差,其中STL-WPT-RFO-HLSTSVR(高斯+多项式+线性)模型对高桥、凤屯站月径流预测的平均绝对百分比误差MAPE分别为2.85%、2.19%,决定系数R2均为0.9994,预测精度最高、效果最好;②混合核函数兼顾了不同核函数优势,能在模型复杂度与泛化能力之间取得平衡,显著提升模型性能和预测精度;③STL-WPT二次分解技术能有效解决复杂时间序列的非平稳性、非线性和多尺度特征,较STL更具分解优势;④组合模型融合了STL-WPT、RFO和HLSTSVR优点,具有较好的普适性和参考价值。
