| It cannot satisfy the need of various fields depending on the human visual inspection.In order to improve the bullet appearance of nondestructive inspection level in our countryand the product quality and production efficiency of bullet, it urgent for higher automationlevel of cartridges for visual check.In this paper，the framework is the algorithm of multi-source information fusion. It’susing a variety of image processing technology, and there are analysis and extraction forappearance characteristics of bullets. It researches and experiments the data fusion of datalevel and decision making level, eventually constructs a decision-making model, has beenintelligent detection method. The main research contents:(1) Image feature extraction. No obvious defects for bullet appearance image, alongwith the noise characteristics, using the histogram equalization, and the smoothing filter onthe bullet image preprocessing, projecting the blemish edge image, and reduce the effectsof noise; An edge detection algorithm based on mathematical morphology, bullets imagefor image segmentation. Specially constructed for the specific characteristics of the bulletimage scars, scratches, multiple structural elements to be complete, clear outline of theimage; Finally, the histogram for the global features area measurement method based onthe area marked for local area characteristics, and local perimeter of the characteristics ofadjacent pixels notation, then GLCM method to obtain the local texture.(2) Data layer fusion. Problem for multi-source heterogeneous information cannot beunified calculated16-merge method and priority information of interest, the gray leveldistribution characteristic dimension reduced from256to5-dimensional, and makes itsdata in the form and other features consistent; for the properties of the sample set is toolarge, the redundant attribute reduction algorithm based on the consistency criterion ofeigenvalues data processing. Not significantly reduce the classification ability to reduce theattribute dimension.(3) Decision fusion. The paper presents a method based on support vector machinebullets look intelligent classification model, and for training time will increase according tothe number of samples, the formation of the curse of dimensionality problem, an improvedsupport vector machine sequential minimal optimization algorithm, the algorithm cansignificantly reduce training time.In this paper, information fusion algorithm successfully applied to the field of visualinspection of the bullets, the correct classification rate of95.6%, little manual inspection gap on the classification accuracy. Therefore, in view of the excellent performance inefficiency, cost, stability, the algorithm has a relatively high superiority and applicability. |
