Research On Flower Species Recognition Based On Deep Learning

In the fields of plant science research,agro-forestry investigation,and production management,plant identification is a crucial basic work,and flower identification is an important part of plant identification.The artificial flower species information query and traditional computer vision methods have the problems of high cost,low efficiency and low accuracy.Applying convolutional neural network to the task of flower recognition,the accuracy rate of flower recognition has a qualitative leap compared with traditional methods.However,due to the different angles of the pictures,the different orientations of the subjects in the flower pictures make the convolutional neural network model have a low recognition accuracy.The following researches have been done to solve the problems:Firstly,through the research of flower image recognition and deep learning,it is found that the flower recognition process is divided into the following basic modules,which are the recognition results for flower image data acquisition,preprocessing,feature extraction and flower classification.When quantifying flower-shaped images in traditional flower recognition,the three most important attributes are color,texture,and shape.Convolutional layers,pooling layers,activation functions,loss functions,basic principles of back propagation and gradient descent,and transfer learning in convolutional neural networks are of great significance for flower recognition.Then,I have studied the three well-known neural networks: LeNet,CapsNet,and ResNet’s algorithm principles.It is analyzed that due to its own limitations,the convolutional neural network has the problem of losing a lot of position information,which leads to a decline in the image recognition rate after rotation transformation.The capsule network model retains location information during image recognition,setting a new record on the handwritten digital data set.However,if the capsule network is directly applied to flower recognition,the effect is not good,so an improved capsule network model is proposed to solve the problems in flower recognition,and the recognition accuracy is 94.1%.The Oxford 17 and Oxford 102 data sets are introduced and 50 representative flower varieties are selected from them to form a new flower data set.Three neural network models are compared and analyzed with the improved capsule network model on the new dataset.Finally,a flower recognition server built on the Python-based Django framework is used to build the mysql flower database.The mui framework is used to set up the flower recognition mobile front-end interface,which enables the mobile phone and computer to upload flower pictures to complete the flower recognition task.Figure 31;Table 4;Reference 51...