With the transformation of education ideas and technology innovation,people pay more and more attention to outdoor learning which aim at expanding knowledge and cultivating practical ability.The mobile visual search technology is a kind of leading-edge information retrieval mode and knowledge interaction mode.It can be used to retrieve the visual informatioa And it also provides a strong technical support for outdoor learning.But now the main problem is that the image matching technology of this kind of learning system is not mature enough,and this problem leads to poor accuracy in practical applications.It has become one of the important research topics in the field of outdoor learning that how to effectively improve the identification accuracy of the system and how to raise system’s reliability to improve learners’ learning efficiency.In the field of image identification,the method of deep learning that can overcome the interference factors such as light,shade,angle in a certain degree is of good robustness.Compared with the traditional method of image identification,the method of deep learning has a significant improvement in identification accuracy.Therefore,this paper hopes to improve the weakness of the previous systems by the method of deep learning.In view of the characteristics of biology and the technical advantages of deep learning,this paper selects flower as a identification object,and proposes a method of flower identification based on deep learning algorithm.This method not only will improve system’s reliability and provide a better user experience,but also enhance learners’ learning efficiency when it applied to the outdoor learning system.Therefore,the major works are as follows:1)This paper proposes a method for modeling a flower identification model based on deep learning.It consists of constructing a data set containing 10 types of 57600 flower images,and based on the data set design comparative experiments to study the influence of different parameters on the model performance.These parameters include activation function,receptive field,number of convolution layers,and so on.2)Through modifying and adjusting the network structure and parameters repeatedly by a lot of experiments,this paper builds a flower deep learning identification model and optimize the model.As a result this model achieve a good effect on new sample data.3)Design and implement a prototype system for outdoor plants knowledge expands based on flower recognition algorithm.Through system design,implementation and inspection,this paper verifies the feasibility of this method,and confirms the application value of this system. |