基于遗传算法的带钢表面缺陷特征降维优化选择

Feature Dimensions Reduction and Its Optimization for Steel Strip Surface Defect Based on Genetic Algorithm

针对带钢表面的划痕、黑斑、翘皮、辊印、褶皱和压印6种典型缺陷,提取了样本图像的灰度、纹理和几何形状特征等32维特征向量。基于遗传算法对32维特征向量进行降维优化选择,选择了其中的20维以进行缺陷图像类型的分类。利用BP神经网络对降维前后的6种典型带钢表面缺陷分类进行对比识别,并同主成分降维方法进行了对比,验证了所提取的带钢表面缺陷图像特征及其遗传算法降维的有效性。

The 32-dimensional feature vectors of intensity,texture and geometry characteristics for six kinds of steel strip surface typical defects images were extracted.The 32-dimensional feature vectors of steel strip surface defect images were reduced and optimized based on genetic algorithm,and 20-dimensional feature vectors were selected to classify types of the defects images.The recognition and classification experiments were carried out to contrast the 32-dimensional feature vectors with 20-dimensional feature ones with BP neural network for six kinds of steel strip surface typical defects images.Furthermore,the results using genetic algorithm was contrasted with principal component analysis.It is shows that the algorithms of features extraction and genetic algorithm are effective.