基于视觉显著性与RBF神经网络融合的织物瑕疵检测

Fabric Defect Detection Based on Fusion of Visual Saliency and RBF Neural Network

为提升织物瑕疵检测准确率,避免出现漏检或误检,提高织物检测智能化程度,提出基于视觉显著性与RBF神经网络融合的织物瑕疵检测方法,将织物图像经中值滤波降噪后,通过织物瑕疵区域对比度、纹理粗糙度和纹理方向的异常显著性模型计算获得显著图,并经RBF神经网络训练获得的映射函数和网络特征字典重构织物图像提取特征,之后采用最大熵自动阈值法定位分割,有效获得织物瑕疵检测结果。实验结果表明:此方法能实时有效对多种异常特征较弱的织物瑕疵进行检测,适应性强,检测准确率大幅提升,能满足实际工业织物瑕疵检测要求。

In order to improve the accuracy of fabric defect detection,avoid missing detection or false detection and improve the intelligence level of fabric detection,a fabric defect detection method based on the fusion of visual saliency and RBF neural network was proposed.After the fabric image was denoised by median filter,the abnormal saliency model including fabric defect area contrast,texture roughness and texture direction was adopted to calculate to obtain the saliency map.The mapping function and network feature dictionary trained by RBF neural network were used to reconstruct the fabric image and extract the features.Then the maximum entropy automatic;threshold method was utilized to locate and segment in order to obtain the fabric;defect detection results effectively.The experiment results showed that the method could effectively detect a variety of fabric;defects with weak abnormal characteristics in real time,it had strong adaptability and greatly improved the detection accuracy,which could meet the requirements of actual industrial fabric;defect detection.