Research On The Algorithms For Seed Identification Of Gramineous Grass Based On Textural Features

Gramineous grass,also called gramineae,is one of the main groups of forage grass in China.Seeds are relatively stable and vital organ for grass reproduction,so accurate and rapid identification of them helps to realize the automation and digital management of grassland.Therefore,in order to increase efficiency as well as reduce labor costs,the study has been conducted on the seed identification for gramineous grass based on textural features in this study.Firstly,the images of gramineous grass seed were captured to set up an image database.And then,some textural features extracted from image were used for comparison and fusion.Finally,we proposed some fit identification approach of gramineous grass seed.The main reserch and conclusions are as follows:(1)Gramineous grass seed images were captured manually to set up an image database.12 species digital images of gramineous grass seeds collected in north China and northwest China were captured by a common digital camera.Owing to some redundant data in original images and chaotic direction of seeds,some pretreatment operations are needed.Firstly,the direction of rotating seeds was treated by geometrical normalization.And then,the area-of-interest was obtained by image segementation.Finally,the image database were set up of the gramineous grass seed.(2)We proposed a seed identification approach based on local similarity pattern(LSP)and linear discriminant analysis(LDA)for gramineous grass.The LSP feature preserved the characteristics of the rotation invariance and the uniform of the local binary pattern(LBP)feature,and could more flexibly extract image texture features by adjusting the SRR.Meanwhile,when performing feature matching,the LDA classifier could ensure that the sample has the largest inter-class distance and the smallest intra-class distance,resulting in a better recognition effect.(3)We proposed a feature extraction approach which fused LSP with GLCM for gramineous grass seeds.LSP was able to describe the local texture details.Moreover,GLCM focused on more global texture information,which fits for large and discrete textural features.So the fusion of LSP and GLCM can get better results in identifying seeds where textural features are highly similar.