基于图像傅里叶变换纹理特征和概率神经网络的气固流化床流型识别

Identification Method of Flow Regime Based on Images Texture Features by Fourier Transform and Probabilistic Neural Network in Gas-solid Fluidized Bed's

提出一种基于图像傅里叶变换纹理特征和概率神经网络相结合的气固流化床流型识别的新方法。该方法利用高速摄影系统获取流型图像。首先对流型图像进行组合滤波去噪,然后运用长方环傅里叶周向谱能量百分比法来计算图像频率分布特征,从而建立流型图像的纹理特征向量,并结合概率神经网络进行训练,实现流型的识别。实验结果表明,该方法能有效地识别气固流化床中鼓泡床、节涌床、湍动床、快速流化床、稀相输送五种典型流型,整体识别率达到98%,为流型识别开辟一条新途径。

A flow regime identification method based on images texture features by Fourier transform and probabilistic neural network was proposed. Flow images were captured by a high speed photography system. First, combined filter method was used to eliminate noises in those images. Then, the algorithm with rectangle loops Fourier spectral energy percentage was used to calculate the feature of image frequency distribution. Thus the images texture eigenvectors of flow regime were established. The probabilistic neural network was trained using above eigenvectors and flow regime i- dentification was realized. The test results show this method can effectively identify five typical flow regimes which are bubbling bed, slugging bed, turbulent bed, fast fluidized bed and dilute phase transfer of gas-solid two-phase flow in fluidized bed. The whole identification accuracy is 98% ,opening up a new avenue for the flow pattern recognition.