基于PSO-Gabor特征增强的钢板表面缺陷识别研究

Study on Defect Recognition of Steel Plate Surface Based on PSO-Gabor Feature Enhancement

为了解决钢板表面类内缺陷形态差异大、类间缺陷相似度高且传统方法难以准确识别的问题,提出了基于PSO-Gabor特征增强的钢板表面缺陷识别方法。首先采用粒子群优化算法(PSO)对Gabor滤波器的频率、方向、尺度和滤波窗口尺寸进行迭代寻优,根据得到的最优参数构造Gabor滤波器;然后利用该滤波器与钢板表面缺陷图像进行卷积操作,并计算其能量响应值获得对应的能量图;最后将能量图输入到卷积神经网络中进行训练和测试。实验结果表明,该方法对钢板表面缺陷识别准确率达到97.5%,检测时间约为50 ms,与传统的模式识别方法相比,识别准确率大幅提升,识别速度快。

In order to solve the problems that there are great differences in defect shape within classes and high similarity between classes in steel plate surface, and the traditional recognition methods are difficult to classify accurately, a defect recognition method of steel plate surface based on PSO-Gabor feature enhancement was proposed.Firstly, particle swarm optimization(PSO) algorithm was used to optimize the frequency, direction, scale and filter window of Gabor filter iteratively, the Gabor filter was constructed according to the optimal parameters, then the defect image of steel plate surface was convoluted with the Gabor filter, and the energy map were obtained by calculating the energy response value of filtered image.Finally, the energy map was used as input to train and test the convolution neural network.The experimental results show that the accuracy of the proposed defect recognition method for steel plate surface defect is 97.5% and the detection time is about 50 ms.Compared with the traditional pattern recognition methods, the recognition accuracy is greatly improved and the detection speed is fast.