Plant Recognition Method Based On Leaf Images

Plant identification plays an important role in the protection of plant species diversity,the development and utilization of plant resources and other fields.The traditional plant identification requires the operator to have rich professional knowledge,and it is heavy and time-consuming.Automatic identification of plants using digital image processing technology has become a hotspot of plant identification.Plant leaves as an important organ of plants,it has a two-dimensional structure and is easy to get.Therefore,blade was used in the research of automatic plant recognition.This paper has conducted the research of plant identification method based on leaf images from image pre-processing,feature extraction and classifier design.In the stage of leaf image pre-processing,the paper introduces leaf image gray processing,leaf image segmentation,leaf image morphology,leaf image boundary extraction and location normalization method.This paper use an interactive image segmentation method to segmentate the leaf images with complex background.Compared with the traditional threshold segmentation method,interactive image segmentation algorithm can obtain an ideal segmentation of the leaf images with complex background.In the feature extraction stage,the characteristics of the plant were extracted by the shape features and texture features.Shape features of the leaves are captured using invariant moments together with geometric parameter of leaves.Texture of the leaf is modeled using Gray Level Co-occurrence Matrix(GLCM).In the classifier design stage,the ensemble learning classifier is applied to the classification of plant leaves.To solve the classification of multiple leaf categories,an improved neural network ensemble algorithm is proposed.Experimentations to demonstrate the efficacy of the proposed approach are conducted on a dataset of 600 images divided into 20 classes with 30 images per class,collected from Flavia.And the accuracy was 91%.By contrast with other methods,this method can significantly improve the accuracy of the system.At the end,a plant identification software based on random forest algorithm is designed.And it achieves satisfying result on the leaves collected from campus.