Research On Butterfly Image Classification Based On Two-level Texture Feature Extraction Method

Nowadays,the development of computer vision technology is getting faster and better,the development and design of automatic recognition system for different types of images has become an important development direction of the technology in combination with the application requirements of different research fields.Among them,combining with the morphological characteristics of different kinds of insects,the image of insects is used for automatic recognition.Butterflies are an important branch of insects,with a wide range of species,so it is very difficult to classify and identify them.The research on the automatic identification method of butterfly species can not only protect the environment,but also be used in border quarantine.In this paper,30 species of butterflies were taken as research objects,and each butterfly contained 20 sample images.A total of 600 digital images of butterfly specimens were collected to construct the research data set.In the aspect of feature extraction,texture feature is used as classification feature in this paper.The calculation of this feature is firstly realized by "Grey-level Co-occurrence Matrix" and KNN algorithm,so as to make the first recognition of all kinds of butterfly images.Then,on the basis of the initial recognition results,this paper takes "local texture feature" as the classification feature of the secondary recognition,and carries out fine-grained image recognition of butterfly species based on the convolutional neural network.The whole experiment not only describes the texture rules and characteristics of the whole image,but also reflects the differences between the local fine textures of butterflies.The initial recognition can greatly reduce the computation,improve the recognition rate,and ensure the integrity of global and local features.The system environment in this paper is windows platform,using python programming software,to complete an automatic identification of the butterfly system.Through experiments,the first recognition accuracy rate of 30 species of butterflies was78%,and that of the top ten was 100%.Through the system's secondary recognition,the identification accuracy is 95.67%,improved the overall recognition accuracy.The results of experiment show that the two-stage texture features not only effectively describe the global texture changes and rules of the butterfly wing surface,but also reflect the local differences of the butterfly wing surface.Combining with the two-level classification model,different species of butterflies can be effectively classified at different granularity levels.Therefore,the relevant methods of feature extraction and the design idea of classification model are effective,which also provide some references for the automatic identification of other insect species in the future.