Research On Feature Extraction Technologies Of Complex Components Internal Structure

In engineering practice, the complex product's reliability is closely related with the state of its internal component, and the correct or not of the internal components'condition will decide the success or fail of the whole system, especially for aviation, astronautics, military fields, therefore, it is needed to detect its internal structures. Because of internal structure state of the complex components is stacked on top of each other, that causing great difficulties in detecting rapidly and accurately. To solve the problem, an effective feature extraction method to get the condition of the internal structure of the complex component is proposed, which established the foundation for the correct and quickly detection of the complex products.Firstly, aiming at the internal structure of complex components, the paper used X-ray digital imaging system to capture image sequences of detected products, and analyzed the characteristics of X-ray image sequences.Secondly, Based on the summary of the traditional image preprocessing method, this paper discussed several different wavelet methods to remove image noise in view of the X-ray image; To background interference, adopting methods which integrates improved FCM and Two-Dimensional Histogram, combined with grey level transformation, through experimental results, it is confirmed that this method can divide the target image from the background actually, with effectively noise elimination.Finally, after the introduce of commonly used feature extracting methods, the KPCA method and the KPCA based on hybrid projection method are proposed in view of the overlapped and staggered internal structure. And with the compare between PCA, 2DPCA, KPCA and the KPCA based on hybrid projection by recognize actually and speed, the result shows that the KPCA has the highest actually, and the KPCA based on hybrid projection is the second one; However, the KPCA based on hybrid projection is faster than normal KPCA on the speed of recognition. So, on condition that the recognition rate of production, KPCA based on hybrid projection is more fit to the on-line detection.