Chinese Orchid Species Recognition Based On Convolutional Neural Network And Transfer Learning

Orchid has a wide variety of species and a wide distribution.With the improvement of people’s living standards,growing orchids has become a hobby of more and more people after life and work.Due to the large number of varieties and similar appearances,the value and research significance of different types of orchids are quite different.Failure to accurately identify orchid species has caused unnecessary losses for many orchid lovers and related orchid workers.The existing national orchid identification methods have the defects of relying on personal experience,incomplete accuracy,or requiring very professional botany knowledge and related experimental equipment,and cannot be popularized.In order to facilitate orchid lovers to better distinguish the species of orchids,and at the same time provide help for the classification and identification of scientific researchers,an accurate,effective,convenient and quick identification method for Chinese orchids is needed.Based on the above background and current situation,this paper uses the technology based on Inception-Res Net-v2 network and migration learning to complete the research on the classification and recognition of Chinese orchid images.The main workflow and content of this article are as follows:(1)In response to the current lack of high-quality and available open Chinese orchid image data sets,the Chinese orchid images were collected and sorted through network downloads supplemented by on-site shooting and book scanning.A total of 5 traditional types of 1982 were established.Chinese orchid data set A and 8 kinds of common Chinese orchid data set B containing a total of 2779 pieces of other varieties.(2)A method based on Inception-Res Net-v2 network and migration learning technology to complete the classification and recognition of Chinese orchid images is proposed.The method Inception-Res Net-v2 network is a pre-training model,designed to explore the impact of experimental results by combining migration learning Factors and improvements.(3)Based on the Inception-Res Net-v2 network,design a structure that integrates shallow features to obtain an improved Inception-Res Net-v2 recognition model for multi-channel feature fusion,which improves the accuracy of recognition and can achieve better recognition The purpose of national orchid classification.(4)Based on the Chinese orchid image recognition model proposed in this article,design and implement a Chinese orchid image recognition system on the Android platform.The system can upload pictures or mobile photo albums to select orchid images.The function of identifying and displaying related types and displaying corresponding maintenance precautions.The experimental results show that the top-1 accuracy of the five most traditional types of national orchid image recognition models proposed in this paper is 91.51%,and the top-1 accuracy of the eight types of national orchid image recognition models including the remaining varieties is87.23%.It has a high recognition rate for common Chinese orchid images,and has reached the expected goal of this article,which has certain research significance.And the Chinese orchid identification mobile terminal application designed in this paper can analyze and identify the images uploaded by users in time,which is convenient and fast,has a wide range of applications,and has certain practical significance and social value.