【摘要】 带钢表面缺陷存在着噪声、光照不均匀、纹理复杂以及局部区域对比度弱的问题。为此，提出了一种新的带钢表面缺陷识别算法。首先提出了ILLBP特征提取算法，在LBP算法中引入LTP的低阈值模式克服一定的噪声和光照影响，为了更好表征带钢表面缺陷复杂的纹理特征引入了ILBP中3种新的纹理结构，同时将LBP值的频率直方图改为了LBP局部梯度幅值和局部梯度方向的频率直方图，使其能更好地表征局部区域对比度强弱的关系；最后为了进一步提高带钢缺陷识别的准确率和减少冗余特征的影响，提出了一种改进的樽海鞘特征选择算法(ISSA)。在NEU数据集上仿真实验结果表明：算法(ILLBP+ISSA)能够克服光照不均匀、局部区域对比度弱、纹理复杂多样的影响，以及对噪声具有一定的鲁棒性。在高斯噪声信噪比为50 dB时带钢表面缺陷识别准确率能达到99.10%,40 dB时准确率能达到97.60%。更多还原

【Abstract】 The surface defects of strip steel have the problems of noise, uneven illumination, complex texture and weak contrast in local areas. Therefore, a new algorithm of strip surface defect recognition was proposed. Firstly, an improved lower local binary patterns(ILLBP) feature extraction algorithm was proposed, the low-threshold pattern of local ternary patterns(LTP) was introduced into the local binary patterns(LBP) algorithm to overcome a certain amount of noise and illumination effects. In order to better characterize the complex texture features of strip surface defects, three new texture structures in improved LBP(ILBP) were introduced. At the same time, frequency histograms of LBP local gradient amplitude and local gradient direction were used to replace the frequency histograms of LBP values, which could better characterize the relationship between the contrast of local areas. Finally, in order to further improve the accuracy of strip defect recognition and reduce the influence of redundant features, an improved salps feature selection algorithm(ISSA) was proposed. The simulation results on the NEU dataset show that the proposed algorithm(ILLBP+ISSA) can overcome the effects of uneven illumination, weak local contrast, complex and diverse textures, and has a certain amount of robustness to noise. When the Gaussian noise signal to noise ratio is 50 dB, the recognition accuracy of strip surface defects can reach 99.10%, and the accuracy can reach 97.60% at 40 dB.更多还原