Study On Image Classification Technology Of Succulent Plants Based On Deep Convolution Network

In recent years,succulents are widely concerned and favored by people because of the highly ornamental value and easy planting.However,it is difficult to rely solely on manual identification and the error rate is high,because there are many kinds of succulents and their appearances are similar.In view of the ten most common succulents in China,a large number images of succulents are collected,the Tensorflow deep learning framework is used to study the succulents recognition methods,and the image-based succulents automatic recognition model is implemented through training.The main work of this thesis is as follows:1.Building a succulents data set.The data set is made up of ten kinds of images of succulents commonly found in China.The succulents data set has a total of 9,348 images,with an average of 934 in each category.The lithops data set has a total of 5,988 images,nearly 600 in each category.The resolution of each image is 300x300.2.Using self-constructed convolutional neural network to realize the image recognition of succulent plants.The multilayer convolution neural network is constructed and the convolution neural network model is trained by data enhancement,dropout,stochastic gradient descent and other techniques.The trained model achieves classification accuracies of 94.7%and 86.7%to recognize succulent plants and lithops.After testing,it costs 6s to recognize an image.3.Realizing image classification of succulent plants based on the AlexNet deep model and the transfer learning technology.①Using fine-tuning technique to retrain the last three full connection layers and get the classification model.The model can achieve 95.3%and 87.3%of the classification accuracy for the succulents and lithops,with an average of 7.5s per image test.② After fusion of features extracted from the fully connected layer,the last three full connection layers are retrained,and the accuracy of the final model are 96.3%and 87.3%to recognize the images of succulents and lithops.The average test time of each image is 8s.4.In this thesis,the image classification model of succulent plants is transplanted to Android operation system to achieve the portability of the recognition of succulent plants,which means that off-line recognition of succulent plants can be done directly by terminal equipment.After testing,the classification model can run normally on mobile phones and has good classification performance.The test results show that the average time for the system to recognize an image of succulent plants is 1Os.