Research On Automatic Detect Ion & Recognition Method For SMD Resistor Surface Defect

SMD resistor has been widely used in such high-density electronic products as phones, PC(Personal Computer)etc, for small volume, high reliability, applying to the flow welding, wave soldering and SMT facilities, all these make it adapt to the development of integration and complanation on electronic products. The method, artificially basing on microscope is not suitable for the drawback of slow speed, high cost for long and high false detection rate. It is important for improving the quality of SMD resistor to have research on method for fast and high-precision automatic flaw detection.Simple structure, orderliness and few gray level of SMD resistor image is the prominent feature of SMD resistor units array, this paper proposes a method for resistor flaw detection based on subgraph projection matching. Three-pixel-wide edge of the SMD resistor image would be get by improved Sobel method from the binary resistor image; the longest line where the resistor edge exists would be detected by rough-elaborate Hough within the bounds of 2°, the speed of which is a order of magnitude higher than traditional one; the correlation coefficient of adjacent resistor units is calculated to distinguish the resistor flaw. The detection result based on the method proposed is not affected by such problem as uneven illumination, instability illumination and constraint zoom. The result of experiment indicates that the detection rate employed the proposed method is 92.5%. Besides, the detection speed is far faster than the traditional block matching.The feature of resistor flaw is extracted on basis of the method of PCA(Principal Component Analysis). PCA removes the correlation and redundant information of samples, gaining image compression information with smallest covariance error. The research on description and recognition precision effected by transformation space consisted of different training samples set and different principal component is done. The result shows that the information of resistor flaw image is compressed to 50D from 123×246D when the principal component constituting the transformation space takes up 78.57% of all. Using the PCA feature as input of SVM cuts down tremendously the calculated amount on flaw classification.The resistor flaw would be recognized by SVM(support vector machine) based on the theory of statistical learning. The flaw categories classifier is built through one against the rest strategy and studies 70 samples to found optimal classification face for flaw recognition. This paper has researches on generalization performance of SVM classification model based on different kernel function and the effect of different kernel parameters and gives the principles of setting kernel parameters. Experiments verify that the best classification model for resistor flaw recognition is SVM based on linear kernel.At last, the software program for flaw detection and identification algorithms is written, and the experimental platform is built. The experiments on threshold for flaw detection, flaw detection and recognition are taken. The result of the experiments verifies that the method proposed is high efficient for SMD resistor flaw detection. The test on identification speed and accuracy of the method is done. The outcome proves that the method meets the requirements on high accuracy and speed.