基于组合优化神经网络的航空发动机叶片损伤图像分割

Segmentation of Blade Damage Image of Aero-Engine Based on Combined-Optimization Neural Networks

采用PNN网络和RBF网络相融合的方法对航空发动机叶片损伤图像进行分割,选取损伤图像80个像素点的RGB值和HSV值分别作为PNN网络和RBF网络的输入样本;针对PNN网络和RBF网络的不足,采用GA算法优化PNN网络和RBF网络的输入参数;考虑到叶片损伤图像采集过程中不确定因素对分割结果的影响,采用D-S证据理论将两种网络分割结果进行融合,进而得到最终的叶片损伤图像分割结果;在30组测试样本中正确识别组数为29,识别率高达96.67%,实践表明,该方法有效地克服了凭借单一识别网络和单一信息源进行叶片损伤图像分割的不足,实现了对叶片损伤图像的高效分割。

Fusion method which is based on PNN neural networks and RBF neural networks is used for segmentation the blade damage image of aero--engine. Select 80 pixel RGB and HSV values of the image as train samples of PNN neural networks and RBF neural net- works. According to the shortage of PNN neural networks and RBF neural networks, the genetic algorithm is used to optimize the input pa- rameters of PNN neural networks and RBF neural networks. Take into account the uncertainty of aero--engine blades damage image in acqui- sition process, the D--S evidence theory was applied to fuse two kinds of neural networks and get the finally segmentation results. This method correctly recognize 29 groups of sample when all the samples is 30. So, the recognition rate is as high as 96. 67%. The results show that this method effectively overcome the shortage of the single recognition network and single source of information for image segmen- tation of blade damage and this segmentation the blade damage image method is high--efficiency.