Study Of Non-destructive Detection Of Wood Species And Density Based On Visible/near Infrared Spectroscopy

Wood is the basic wood industry product,and different wood have different properties(such as wood species,density,moisture content,roughness and strength),which leads to great differences in the use,physical property and price.Therefore,the detection of the above wood properties is very important in wood quality inspection industry.Visible/near-infrared spectroscopy technology has the advantages of low cost,high efficiency,high speed,non-destructive,easy detection,and good test reproducibility.In this paper,a new exploration has been carried out by using them to detect the wood species and density.Firstly,aiming at the problem that the wood detection models are usually based on specific wood species and can not reject the abnormal wood species,and that the traditional wood species detection models need positive and negative samples,but the abnormal wood species are too numerous to be obtained,this paper proposes a method of detecting abnormal wood species by using one-class classifiers which are constructed for the samples in the normal condition(target samples)and the abnormal samples are not needed.The unknown samples are abnormal or not can be judged by the threshold.The detection results of three one-class classifiers constructed by BP neural network,self-organizing feature mapping network and support vector data description are compared.The result shows that the one-class classifier constructed by BP has the best detection ability.What's more,this paper proposes a one-class classifier established by compound network that composed of RBF neural network and BP neural network,which can increase the difference between positive wood species and abnormal wood species,improve the anomaly detection rate and reduce the false alarm rate very well.Secondly,aiming at the problem that several models are needed to be established if multiple properties are required in the current wood detection researches,this paper proposes a method of simultaneous detection of wood species and density by using visible/near infrared spectroscopy.It can output two properties simultaneously with only one model.Two methods of dimensionality-reduction of principal component analysis and wavelet transform are combined with BP neural network and least squares support vector machine respectively which have the characteristic of multi-output to establish the models that can detect wood species and density simultaneously.Then,the accuracies of the models are compared.The result shows that the combination of wavelet transform and least squares support vector machine has the best effect of detection,which can accurately detect the wood species and density simultaneously,output the analysis results of qualitative and quantitative at the same time,and improve the efficiency of wood detection.Then,this paper proposes an automatic detection method for the clustering number of unknown wood species based on visible/near infrared spectroscopy.Firstly,t-SNE is used to reduce the dimension of spectral data,and then Clustering by fast search and find of density peaks(CFSFDP)is combined with internal index of CH to detect the clustering number of unknown wood species.The result shows that CFSFDP combined with CH can accurately detect the clustering number of unknown wood species when t-SNE is used to reduce the spectral dimension to 3-D.At the same time,according to the evaluation result of four external index of Rand,Adjusted Rand,Jaccard and Fowlkes-Mallows,we can know that the proposed method in this paper can accurately classify the unknown wood species into the corresponding categories.At last,based on the results of above researches,a prototype system of non-destructive detection for wood species and density based on visible/near infrared spectroscopy is designed by using MATLAB programming software.When a sample to be tested is imported into the system,the first step is to detect whether it is an abnormal wood species,if not,it will show its detection results of the wood species and density simultaneously.Otherwise,the system will refuse it.If the number of rejected samples reaches the threshold,the system will cluster them and give us the result of the clustering number.