| With the rapid development of wood processing industry,people’s demand for wood veneer quality has been rising,how to improve the quality of wood veneer has become an important issue of concern for wood product processing enterprises,and the wood defect detection technology based on machine vision as an important means to improve the quality of wood veneer is of great significance to promote production efficiency and save production costs.However,in practical production,the use of machine vision to detect defects is susceptible to the influence of the acquisition environment,equipment,calculation methods and other factors,which may result in low recognition rate,poor accuracy and weak stability of veneer defect images,therefore,it is particularly important to improve the recognition efficiency of wood veneer defect images,reduce time consumption and enhance the analysis and utilization of wood veneer defect artificial intelligence.In this thesis,based on the traditional support vector machine,we propose a recognition method based on Kernel Principal Component Analysis(KPCA)and Multiclass Support Vector Machines(LIBSVM)to identify live knots,dead knots and cracks in wood veneer.The pre-processed defect images were used as the basis for the identification of the defects.The pre-processed defect images were used as recognition samples,and three types of recognition parameters,namely color moments,Gabor wavelet variations,and Hu-invariant moments,were identified by using the KPCA-LIBSVM recognition method optimized by Gravitational Search Algorithm(GSA).The main elements of this thesis study are as follows.1.The theoretical basis of adaptive Gaussian filtering,adaptive gamma correction and multi-scale detail enhancement methods are introduced,and the three defects of wood veneer are pre-processed,and the enhanced veneer image quality is improved based on three evaluation indexes: mean square error,peak signal-to-noise ratio and structural similarity.2.On the basis of pre-processing,color features,texture features,and shape features of veneer defect images are extracted using color moments,Gabor wavelet variations,and Hu invariant moments to form the feature set required for recognition,and the feature set is dimensionalized using kernel principal component analysis,and the central kernel matrix is formed by normalizing the feature set,and the feature values and feature vectors corresponding to the kernel matrix are calculated,and the cumulative variance of the feature values according to the The main features of wood veneer defect image information after dimensionality reduction are obtained as recognition data.3.Based on the theoretical basis of multiclassification support vector machine,the Gaussian radial basis kernel function is derived as the best kernel function applicable to the recognition of live,dead,and cracked veneer,in terms of recognition rate and elapsed time.Then the Gaussian radial basis kernel function is brought into the recognition model formed by kernel principal component analysis and multiclassification support vector machine,and compared with LIBSVM without feature reduction,and it is concluded that KPCA-LIBSVM is more effective in improving the recognition efficiency and reducing the time cost of wood veneer defects.4.The principles and methods of three parameter optimization methods,genetic algorithm(GA),particle swarm algorithm(PSO)and gravitational search algorithm,are introduced,and the optimal penalty coefficient C and Gaussian kernel coefficient g are derived from the adaptation curves of different methods.5.The veneer defect image recognition system is established,combining the method of this thesis with genetic algorithm,particle swarm algorithm and KPCA-LIBSVM model to form a complete recognition system,and using Accuracy,Precision,Recall and F1(F1-score)as objective evaluation indexes for live knots,dead knots The research on the identification of three kinds of defects: live knots,dead knots and cracks.The results show that the KPCA-LIBSVM model based on GSA optimization proposed in this thesis shows a better recognition effect in the recognition of different veneer defects of live knots,dead knots and cracks,in which the average accuracy reaches94.71%,the accuracy,recall and F1 reach more than 90% on average,and the overall recognition effect has been improved compared with the comparison algorithm. |
