Visual Quality Recognition of Nonwovens using Wavelet Texture Analysis and Robust Bayesian Neural Network

An approach to recognize the visual quality of nonwovens by combining wavelet texture analysis, Bayesian neural network and outlier detection is proposed in this paper. Nonwoven images (625) of five different grades, 125 of each grade, are decomposed at four levels with wavelet base sym6 and two energy-based features, norm-1 L1 and norm-2 L2, are calculated from wavelet coefficients of each high frequency subband to train and test Bayesian neural network. To detect the outlier in the training set, scaled outlier probability of training set and outlier probability of each sample are introduced. When the nonwoven images are decomposed at level 3, with 500 samples to train the Bayesian neural network, the average recognition accuracy of test set is 98.4%. Experimental results on the 625 nonwoven images indicate that the energy-based features are expressive and powerful in characterizing texture of nonwoven images and the robust Bayesian neural network has excellent recognition performance.