Application Of Image Recognition In The Classification Of The Roadside Trees

With the improvement of urban environmental construction standards,roadside trees as the backbone of the urban road greening tree species are massively planted.In the daily management,roadside trees are often transplanted for road reconstruction.The species and numbers of roadside trees are changed requently,so roadside trees need to be recounted.Based on this demand this paper studies the application of image recognition in the classification of roadside trees from image segmentation,feature analysis,and classifier structure.In this paper,the specific research process is divided into three steps,the application of threshold segmentation,regional growth and edge detection and other segmentation methods for tree image segmentation,and both the principle and the application results are analyzed.After comparing various segmentation methods,the effect of the adaptive threshold segmentation method is considered better than other methods of threshold segmentation.On the basis of traditional region growing segmentation method based on gray value,this paper puts forward the principles and steps of the color region growing segmentation based on color similarity.Through observing the segmentation effect of various gradient operators on tree image segmentation in the segmentation method experiment based on edge detection,it founds that the second-order gradient operators of Laplace operator can carry on the edge detecting on tree images under a certain amount of noise interference with satisfactory results comparing other methods.After the tree image is separated from the background,the image of the tree itself is further extracted and analyzed.In this paper,the characteristics of both the tree shape and the tree trunk texture are selected.On the aspect of the feature extraction of tree shapes,it carries on the corrosion and expansion operation for different tree species by the use of mathematical morphology,analyses various regional properties,and obtains the shape characteristic parameters for the next step analysis.On the aspect of the feature extraction of tree trunks,it extracts the texture feature of bark by using radon transforms,gray probability density function,gray level co-occurrence matrix and HSV model.The results show that the texture of the bark is more effective in the use of the gray level co-occurrence matrix and the HSV model to identify the polluted texture of the bark.After the feature extraction,the BP neural network is used to construct the classifier,and the multiple eigenvalues of the trees are selected as the input signals of the neuron,and the classification results of the tree species are output by the classifier.Experimental results show that the BP neural network trained by multiple sets of training data is ideal for the classification of the test samples in this paper.This shows that the results of this study are feasible in the application of the roadside tree classification.