Characteristic Identification Of Bark Based On Gray Level Co-occurrence Matrix And SOM Neural Network

Texture is the natural attribute of bark which can be used as important basis to distinguish the different tree species.The bark of some tree species(such as Cortex Phellodendri,the Fraxinus mandshurica,etc.)has precious medicinal value.Forestry workers by girdling regeneration technology acquisition bark ensure that the economy is in need to effectively protect the trees at the same time.But the tree bark close and not easy to distinguish,at the same time gathering the site conditions,the light interference factors,such as manual collection and identification efficiency is low.For forestry workers professional quality request is higher,the result of the acquisition is not ideal,and caused great waste of human resources.Therefore,how to seek a highly efficient,can realize the tree species classification and recognition method,furthermore has both theory and practice value.In this study,the image processing technology and pattern recognition theory were used to construct the SOM neural network model,and the tree species were effectively identified.First,300 images were collected from the three major hardwoods of the Northeast,and the images of ROI(region of interest)were interfered with each image,and the histogram was balanced,There by obtaining an image with a definite area,a strong contrast,and a characteristic value.Then,the characteristic parameters of the gray-level co-occurrence matrices of the processed images are analyzed with the three structural factors(the generation step size,the image gray level and the generation direction),and the characteristics of the tree bark skin texture are repeated,The construction method of gray co-variance matrix suitable for describing the skin texture of tree bark was established,and the construction d=2;g=128;?=0,45,90 and 135 were established.Based on the 12 texture feature parameters of the gray co-variance matrix,the tree number is W1-W12,and the bark texture feature storage GUI interface is established to effectively preserve the 12 eigenvalue data of the bark texture information.The eigenvalues of the two eigenvalues of the three tree bark skin tattoos were analyzed.Eight eigenvalues were selected and analyzed.The second moment,entropy,moment of inertia,correlation,variance,mean sum,maximum probability,and entropy.The six eigenvalues are taken as SOM neural network input feature vector,and the recognition rate is 83.33%.The experimental results show that the method is more accurate,convenient and efficient than the traditional method,and can help the forestry workers to use the image processing technology to automatically distinguish the three kinds of trees.