Identification Of Different Degrees Of Ginkgo Leaf Disease Based On Deep Learning

Ginkgo biloba leaves have high economic value,medicinal value and ornamental value.The diseases and insect pests of Ginkgo biloba leaves can lead to the death of Ginkgo biloba and bring economic losses.Ginkgo Biloba Leaves Extract Tablets is unable to take measures for the late illness.Therefore,early,automatic identification and prediction of Ginkgo Biloba Leaves Extract Tablets is of great significance for reducing the incidence rate and mortality of Ginkgo Biloba Leaves Extract Tablets.At present,many scholars apply deep learning to the identification of crop species and plant diseases and insect pests,but there are few researches on the same plant disease level,especially on the automatic identification of the disease degree of Ginkgo leaves.Therefore,The innovations and contributions of this paper are as follows: the blank of the study on the recognition of the disease degree of Ginkgo leaves is supplemented;cnnlstm model is applied to the recognition of the disease degree of Ginkgo leaves for the first time.The cnnlstm model can recognize time-dependent sequence images.This paper studies the application of Ginkgo biloba leaf disease recognition based on deep learning,using convolution neural network and cyclic neural network for training.The main research work is as follows:(1)Two methods based on CNN were studied to identify the prevalence of Ginkgo biloba leaves in single background and complex background.The first method is based on vggnet-16 model.The recognition rate of ginkgo leaf in a single background is 98.44%,and the recognition rate of ginkgo leaf in a complex background is 92.19%.Finally,the best model is selected.The second method is to study the recognition method of ginkgo leaf disease degree based on the fine-tuning of perception V3 network model.The recognition rate of ginkgo leaf in single background was 92.30%,and that in complex background was 93.20%.Vggnet-16 model has the highest recognition rate for the disease degree of Ginkgo biloba leaves with single background,and the network model with fine adjustment of perception V3 has the best recognition effect for the disease degree of Ginkgo biloba leaves with complex background.For the two different backgrounds of Ginkgo biloba leaves under the same model,the recognition rate of the fine-tuning network model of perception V3 is relatively stable and the generalization ability is better.(2)In this paper,we also studied the method of identifying the diseased degree of Ginkgo biloba leaves based on cnn-lstm.The sequence images in complex background are studied and trained,and the accuracy is 93.06%.The effects of data enhancement and super parameter adjustment on the performance of cnn-lstm are studied.It is verified that increasing the sample size of the data set can improve the accuracy of the test set,reduce the over fitting,and improve the performance of the model,so as to achieve a good effect of sequence image recognition in complex background.It is also verified that the effect of adjusting the super parameters on the model when the sample size of the dataset is constant.After using the technique to adjust the super parameters,the recognition rate and the performance of the model are improved to a certain extent.