摘要
提出一种阈值均匀局部二值模式(TULBP)算法。该算法使用均一模式以减少特征值的数量,在保证特征描述准确性的同时,可大大加快计算速度。在该算法基础上,提出基于BP神经网络的缺陷检测算法,使用矩形窗口提取特征值,将复杂的缺陷模式判断转化为神经网络模式识别问题。实验结果表明,该算法在使用单层BP神经网络时,即可达到较高的准确性,抗噪声能力强,适用范围广。
In order to detect solar cell defects precisely and break through the limitation of low resolution, an algorithm called Threshold Uniform Local Binary Pattern （TULBP） was proposed. The algorithm used uniform patterns to reduce quantity of characteristic values so that it could ensure accuracy of feature description and quicken computing speed at same time. A BP neural network method with a rectangle testing window to detect defects was also proposed based on TULBP, which transferred the defects detection problems to a pattern recognition problem. The experiment results show that this method can reach high accuracy rate with single layer BP neural work.
出处
《太阳能学报》
EI
CAS
CSCD
北大核心
2014年第12期2448-2454,共7页
Acta Energiae Solaris Sinica
