The conception of 'main direction' of multi-dimensional wavelet is established in this paper, and the capabilities of several classical complex wavelets for representing directional singularities are investiga...The conception of 'main direction' of multi-dimensional wavelet is established in this paper, and the capabilities of several classical complex wavelets for representing directional singularities are investigated based on their main directions. It is proved to be impossible to represent directional singularities optimally by a multi-resolution analysis (MRA) of L2(R2). Based on the above results, a new algorithm to construct Q-shift dual tree complex wavelet is proposed. By optimizing the main direction of parameterized wavelet filters, the difficulty in choosing stop-band frequency is overcome and the performances of the designed wavelet are improved too. Furthermore, results of image enhancement by various multi-scale methods are given, which show that the new designed Q-shift complex wavelet do offer significant improvement over the conventionally used wavelets. Direction sensitivity is an important index to the performance of 2D wavelets.展开更多
Owing to the intrinsic nonlinearities of the system,a contracting mechanism,such as myogenic response,may induce different oscillatory patterns.Many specialists discussed the relations of oscillatory patterns with int...Owing to the intrinsic nonlinearities of the system,a contracting mechanism,such as myogenic response,may induce different oscillatory patterns.Many specialists discussed the relations of oscillatory patterns with intrinsic control system or some pathological condition,but there is no single,well-defined criterion to achieve the identification of regular,stochastic,and chaotic activities.In this paper,we focus on the Mallat algorithm of wavelet packet and use it in the identification of the regular periodic,stochastic,and chaotic fluctuations.According to the specific frequency configuration of the chaos activity,we select proper layers of decomposition of wavelet packet and did fine segments to the frequency of signals.The frequency band of energy convergence could be recognized.The signal of periodic,stochastic,and chaotic could be distinguished depending on it.Numerical experiment is given to show its efficiency.Experiments on 12 babies' lung data have been done.This identification by means of wavelet packet could support the cardiologist or cerebral specialist to do more observation and deeper analysis to physic signals.展开更多
On-line partial discharge(PD)detection still remains a very challenging task because of the strong electromagnetic interferences.In this paper,a new method of de-noising,using complex Daubechies wavelet(CDW)transform,...On-line partial discharge(PD)detection still remains a very challenging task because of the strong electromagnetic interferences.In this paper,a new method of de-noising,using complex Daubechies wavelet(CDW)transform,has been proposed.It is a relatively recent enhancement to the real-valued wavelet transform because of tow important properties,which are nearly shift-invariant and availability of phase information.Those properties give CDW transform superiority over other real-valued wavelet transform,and then the construction algorithm of CDW is introduced in detail.Secondly,based on the real threshold algorithm of real-valued wavelet transform,complex threshold algorithm is devised.This algorithm take the different characteristics of real part and imaginary part of complex wavelet coefficients into account,it modifies the real and imaginary parts of complex wavelet coefficients respectively.Thirdly,to obtain a real de-noised signal,new combined information series is devised.By applying different combination of real part and imaginary part of de-noised complex signal,a real de-noised signal can be restored with higher peak signal-to-noise ratio(PSNR)and less distortion of original signals.Finally,On-site applications of extracting PD signals from noisy background by the optimal de-noising scheme based on CDW are illustrated.The on-site experimental results show that the optimal de-noising scheme is an effective way to suppress white noise in PD measurement.展开更多
针对真实环境下采集的病害图像中存在的大量噪声和复杂背景干扰,导致作物病害识别准确性和泛化性低的问题,该研究提出基于自适应BayesShrink和频-空特征融合的作物病害识别方法(adaptive BayesShrink and frequencyspatial domain featu...针对真实环境下采集的病害图像中存在的大量噪声和复杂背景干扰,导致作物病害识别准确性和泛化性低的问题,该研究提出基于自适应BayesShrink和频-空特征融合的作物病害识别方法(adaptive BayesShrink and frequencyspatial domain features fusion, AFSF-DCT)。首先,设计了自适应BayesShrink算法(Ad-BayesShrink)以减少噪声干扰,同时保留更多细节,降低识别模型提取病害特征的难度。然后提出基于频-空特征融合和动态交叉自注意机制的作物病害识别模型(crop leaf disease identification model based on frequency-spatial features fusion and dynamic cross-self-attention,FSF-DCT)。为实现全面的频-空特征映射,设计了基于离散小波变换(discrete wavelet transform,DWT)和倒残差结构(bneck)的频-空特征映射(DWT-Bneck)分支以捕获多尺度病害特征。频域分支设计了基于2D DWT的特征映射模块(2D DWT-based frequency-features decomposition module, DWFD)以捕获病害细节和纹理,用于补充空间域特征在全局信息表达上的不足。空间域分支在bneck中引入CBAM(convolutional block attention module)和Dynamic Shift Max激活函数以实现全面的空间特征映射。最后设计了动态交叉自注意特征融合模块(multi-scale features fusion network based on dynamic cross-self-attention, MDCS-DF)融合频-空特征并增强模型对病害特征的关注。结果表明,Ad-BayesShrink获得了35.78的最高峰值信噪比,优于VisuShrink和SUREShrink。FSF-DCT在自建数据集和2个开源数据集(PlantVillage和AI challenger 2018)上分别获得了99.20%、99.90%和90.75%的识别精度,且具有较小的参数量(7.48 M)和浮点运算数(4.62 G),优于当前大部分的主流识别模型。AFSF-DCT可为复杂背景下的作物叶片病害的快速精准检测提供模型参考。展开更多
基金Supported by National Natural Science Foundation of P.R.China (10171109)the Special Research Fund for Doctoral Program of Higher Education of P. R. China (20049998006)
文摘The conception of 'main direction' of multi-dimensional wavelet is established in this paper, and the capabilities of several classical complex wavelets for representing directional singularities are investigated based on their main directions. It is proved to be impossible to represent directional singularities optimally by a multi-resolution analysis (MRA) of L2(R2). Based on the above results, a new algorithm to construct Q-shift dual tree complex wavelet is proposed. By optimizing the main direction of parameterized wavelet filters, the difficulty in choosing stop-band frequency is overcome and the performances of the designed wavelet are improved too. Furthermore, results of image enhancement by various multi-scale methods are given, which show that the new designed Q-shift complex wavelet do offer significant improvement over the conventionally used wavelets. Direction sensitivity is an important index to the performance of 2D wavelets.
基金Supported by the National Natural Science Foundation of China (60102002)the Doctoral Foundation of Hebei Province of China(B2004522)
文摘Owing to the intrinsic nonlinearities of the system,a contracting mechanism,such as myogenic response,may induce different oscillatory patterns.Many specialists discussed the relations of oscillatory patterns with intrinsic control system or some pathological condition,but there is no single,well-defined criterion to achieve the identification of regular,stochastic,and chaotic activities.In this paper,we focus on the Mallat algorithm of wavelet packet and use it in the identification of the regular periodic,stochastic,and chaotic fluctuations.According to the specific frequency configuration of the chaos activity,we select proper layers of decomposition of wavelet packet and did fine segments to the frequency of signals.The frequency band of energy convergence could be recognized.The signal of periodic,stochastic,and chaotic could be distinguished depending on it.Numerical experiment is given to show its efficiency.Experiments on 12 babies' lung data have been done.This identification by means of wavelet packet could support the cardiologist or cerebral specialist to do more observation and deeper analysis to physic signals.
基金Project Supported by National Natural Science Foundation China（50577069）, National Grid Company （2004-SGKJ）.
文摘On-line partial discharge(PD)detection still remains a very challenging task because of the strong electromagnetic interferences.In this paper,a new method of de-noising,using complex Daubechies wavelet(CDW)transform,has been proposed.It is a relatively recent enhancement to the real-valued wavelet transform because of tow important properties,which are nearly shift-invariant and availability of phase information.Those properties give CDW transform superiority over other real-valued wavelet transform,and then the construction algorithm of CDW is introduced in detail.Secondly,based on the real threshold algorithm of real-valued wavelet transform,complex threshold algorithm is devised.This algorithm take the different characteristics of real part and imaginary part of complex wavelet coefficients into account,it modifies the real and imaginary parts of complex wavelet coefficients respectively.Thirdly,to obtain a real de-noised signal,new combined information series is devised.By applying different combination of real part and imaginary part of de-noised complex signal,a real de-noised signal can be restored with higher peak signal-to-noise ratio(PSNR)and less distortion of original signals.Finally,On-site applications of extracting PD signals from noisy background by the optimal de-noising scheme based on CDW are illustrated.The on-site experimental results show that the optimal de-noising scheme is an effective way to suppress white noise in PD measurement.
文摘针对真实环境下采集的病害图像中存在的大量噪声和复杂背景干扰,导致作物病害识别准确性和泛化性低的问题,该研究提出基于自适应BayesShrink和频-空特征融合的作物病害识别方法(adaptive BayesShrink and frequencyspatial domain features fusion, AFSF-DCT)。首先,设计了自适应BayesShrink算法(Ad-BayesShrink)以减少噪声干扰,同时保留更多细节,降低识别模型提取病害特征的难度。然后提出基于频-空特征融合和动态交叉自注意机制的作物病害识别模型(crop leaf disease identification model based on frequency-spatial features fusion and dynamic cross-self-attention,FSF-DCT)。为实现全面的频-空特征映射,设计了基于离散小波变换(discrete wavelet transform,DWT)和倒残差结构(bneck)的频-空特征映射(DWT-Bneck)分支以捕获多尺度病害特征。频域分支设计了基于2D DWT的特征映射模块(2D DWT-based frequency-features decomposition module, DWFD)以捕获病害细节和纹理,用于补充空间域特征在全局信息表达上的不足。空间域分支在bneck中引入CBAM(convolutional block attention module)和Dynamic Shift Max激活函数以实现全面的空间特征映射。最后设计了动态交叉自注意特征融合模块(multi-scale features fusion network based on dynamic cross-self-attention, MDCS-DF)融合频-空特征并增强模型对病害特征的关注。结果表明,Ad-BayesShrink获得了35.78的最高峰值信噪比,优于VisuShrink和SUREShrink。FSF-DCT在自建数据集和2个开源数据集(PlantVillage和AI challenger 2018)上分别获得了99.20%、99.90%和90.75%的识别精度,且具有较小的参数量(7.48 M)和浮点运算数(4.62 G),优于当前大部分的主流识别模型。AFSF-DCT可为复杂背景下的作物叶片病害的快速精准检测提供模型参考。
