Shape Complex Invariant Descriptor Of Object Image And Its Application In Plant Leaf Recognition

Plants are an important part of the earth's ecological environment.They play an important role in maintaining the balance of the earth's ecological balance.However,with the great expansion of human civilization,human's productive activities have made a great negative effect on plants,as a result,a great deal of plants were destroyed or even extinct.Therefore,it's time of us to take some measures to protect plant diversity.However plant classification and identification is a very challenging task,even experienced botanists may not be able to correctly identify it.In addition,plant leaves collected under natural conditions are vulnerable to light,occlusion,shape distortion and other factors.For these reasons,the key to directly affect the accuracy of plant recognition is to extract the invariant shape features from plant leaves.In this thesis,firstly,the research background and significance of the subject are described.Secondly,the research situation of plant recognition at home and abroad is introduced,and the current popular plant recognition methods based on leaf shape and some technical difficulties are mainly introduced.Owing to the shape of the leaf is prone to deformation while they are being collected,this paper makes a thorough study of the image changes.On this basis,the current popular shape descriptors are classified and introduced,and the advantages and disadvantages of each algorithm are analyzed.In order to solve these problems,a shape complex invariant descriptor is proposed in this thesis.The method has the following characteristics:(1)The method satisfies the invariance of rotation,scaling and translation.This means,when the shape changes in rotation,scaling and translation,the features extracted by this method are unchanged.(2)The method achieves high recognition accuracy.In order to illustrate the superiority of this method,we tested it on Swedish,CVIP100,ICL and subICL datasets,and compared it with the current methods of obtaining high recognition rate on these datasets.The experimental results indicate that the proposed method achieves the stateof-art performance among all these datasets.(3)The proposed method is insensitive to the self-occlusion of plant leaves.The problem of self-occlusion on plant leaves restricted the improvement of the accuracy on plant recognition.Up to now,many methods have been proposed to solve this problem,however the results are not good.In this thesis,a shape descriptor is proposed to solve this problem effectively.In order to verify it,a dataset of self-occluding leaves is selected on the ICL dataset for testing.The recognition rate of the proposed method on this data set is much higher than other methods.