Abstract: In order for the support vector machine to accurately identify the surface defects of wood veneer and improve the quality of wood veneer, an efficient and accurate recognition model of kernel principal component analysis（KPCA） gravity search algorithm-support vector machine（GSA-SVM） for veneer defects was proposed. Considering the redundant effect between image feature data, KPCA method was used to reduce the dimension of original feature data, and GSA optimized the penalty factor C and kernel parameter g of support vector machine（SVM） to establish KPCA-GSA-SVM wood veneer defect recognition model. Based on the three features of color, texture and shape, the raw data set of samples with live knots, dead knots and cracks as the research objects, 8 main features（1 color feature, 1 texture feature and 6 shape features） were selected as the basis for wood veneer recognition. The wood veneer identification model was learned, trained, predicted and analyzed, and the identification effect was compared with the KPCA-PSO-SVM identification model composed of the traditional particle swarm parameter optimization algorithm（PSO）. The results showed that the recognition rate of live knots, dead knots and cracks of the wood veneer recognition model based on KPCA-GSA-SVM was 100%, 96.78% and 100%, which were 21.62%, 0.63% and 7.41% higher than that of KPCA-PSO-SVM, and the overall time was shortened by 7.26 s, it can be seen that the prediction recognition rate, recognition speed and stability were higher than the former. The research conclusions identify the veneer defects from a new perspective, which is helpful for the identification and development of wood veneer defects.