Research On Methods Of Plant Leaf Recognition

Plants play an important role in nature.The orderly circulation of nature and the daily production and life of people are inseparable from plants.There are many kinds of plants in the world,and the number is huge.Especially in our country,plant species are rich and diverse,and widely distributed.How can we quickly and effectively identify plant species?Due to the wide variety of plants,relying on manual identification is low efficiency,and automated identification is particularly important.Different plant leaves have large differences in shape,color and vein texture,so that we can use the visual characteristics of plant leaves to distinguish different species.With the rapid development of current digital computer processing technology and artificial intelligence technology.the visual feature extraction of plants does not rely on labor,computer technology and programming software play an important role in such fields,not only can save manpower and material resources,but also Can effectively improve the accuracy of image recognition.This paper is based on the UCI public data set.A total of 281 plants were selected from the UCI dataset,with an average of 12 images per class,and each image corresponds to 14 features.First,the characteristics of all pictures are classified by the first method. The second method is to directly train and identify 281 pictures using convolutional neural networks,as follows:The first category is the traditional method of machine learning,including decision trees,support vector machines,and random forests.The traditional method in machine learning is based on the external features of plant leaves.The results of classification are as follows:first:decision tree classification,plant leaf recognition rate:64.4%;second random forest,plant leaf The recognition rate is:78.3%;the third:support vector machine.the recognition rate of plant leaves:89.3%.And the ten-fold cross-validation is carried out,and the error value is getting smaller and smaller.It can be seen that the support vector machine has a good effect in the machine learning methodThe second method is to use the convolutional neural network in deep learning.The convolutional neural network has high accuracy in image information recognition.It does not need to manually define the leaf texture features to reduce the labor cost.The network construction and training process is based on this paper.Keras deep learning framework,the data set of the photo library is tailored to a uniform size of 299 × 299,and the final recognition rate is 83.67%.In order to improve the recognition rate,this paper further pre-processes and expands the original image.Firstly,the dataset of the image library is pre-processed in batches.and then the original image is expanded in direction and grayscale.In the end,2.880 images will be expanded by 281 images.Then 2880 pictures were divided into 9:1,90%of which were used as training sets,and the remaining 10%pictures were trained and tested.After network training,the plant leaf recognition rate reached 95.14%.It can be concluded from the experiment that the convolutional neural network is superior to the traditional method of machine learning in terms of accuracy and efficiency.It can be seen that the convolutional neural network has a good application prospect in the identification of plant leaves.