TheClassification Algorithm ofDefects inWeld Image based onAsymmetrical SVMs

Thispaperfirstly analyzes theClassification principle ofSVM andindicates thatSVM cannotobtain favorable classification ability whenthenumbers ofallclasses ofsamples varygreatly. Thealgorithm ofasymmetrical SVM isputforward basedontheanalysis ofthereason whythe classification inclination comesintobeing, whichcan compensate theeffect oftheunevenclass sizes andadvance the classification accuracy ofthesmaller samplesize. The experimental results ofdefect recognition inweldimageshow thatthis algorithm canimprove theaccuracy ofsmallclass effectively.