基于Gabor小波和神经网络的布匹瑕疵检测

Fabric defect detection based on Gabor wavelet and neural network

为了实现布匹表面瑕疵的在线视觉检测,利用Gabor小波函数与神经网络的结合,提出了一种有效提取Gabor滤波最优参数的方法。该方法通过离线构建Gabor小波神经网络,结合Levenberg-Marquardt算法优化得到最优解,重构无瑕疵的布匹图像,以削弱在线检测时布匹纹理对瑕疵检测的影响,从而能够于在线实时监测过程中凸显布匹瑕疵,最终从融合图像中得到瑕疵区域。通过对霉点、断经、油污、破洞四种常见的布匹瑕疵图像进行检测,表明该方法能够满足对瑕疵的实时分割要求。

To achieve the fabric defects in online visual detection, an effective method is proposed by combining the Gabor wavelet function and neural network to extract optimal Gabor filter parameters. Through building a Gabor wavelet neural network model, with the Levenberg-Marquardt algorithm to find the optimal solution, this method reconstructs the non-defect fabric image offline, to weaken the impact of the texture of the fabric defect detection during online testing, so that the defects will be highlighted during the online real-time testing, and then the defect area can be segmented from the fused image. The experimental results of four typical defect images including stain, broken warp, oil stain, and hole prove that this method is effective.