Research On Deep Learning Method For Chinese Herb Planting

Chinese medicinal planting industry is the source of the traditional Chinese medicine industry chain,which directly affects the quality and curative effect of Chinese medicinal materials and impeds the inheritance and development of traditional Chinese medicine.This paper conducts model research on the core problems of the planting of Chinese medicinal materials,in order to provide auxiliary decision-making for the cultivation of Chinese medicinal materials,realize efficient management,and promote the scientific and modernization of Chinese herb planting.Based on the ecological suitability of traditional Chinese medicinal materials,the paper discovers potential suitable areas,identifies leaf diseases in the field management of traditional Chinese medicinal materials,and combines the time series data of traditional Chinese medicinal materials to predict yields.The work of this paper is as follows:1.Aiming at the problem of discovering potential suitable areas of Chinese medicinal materials urgently,predicting suitable growth areas for Chinese medicinal materials using deep support vector data description model.Because the data format of the ecological environment factors is not uniform,the t-SNE algorithm is used for unified transformation and fusion,using a deep support vector data description model to map the fused data to the high-dimensional feature space in a non-linear manner and find the smallest hypersphere(optimal hypersphere),when the location sample point falls within the optimal hypersphere,the location is considered suitable for the growth of this Chinese medicinal material;otherwise,it is not suitable for that.The experimental results for data of Salvia miltiorrhizae show that the AUC(Area Under Curve)value is 99.65%.2.Aiming at the problem that manual identification cost of medicinal plants diseases in the leaves is large and high recognition error rate,identifying leaf diseases of Chinese medicinal materials using convolutional neural network with depth-separable scheme.Firstly,the network structure such as Inception and ResNet with PlantVillage data sets are trained in advance;secondly,which is migrated to the leaf disease data sets and finetuning its parameters;finally,the depth separable convolution operation and random pooling layer are integrated to reduce the number of parameters and prevent overfitting to realize leaf disease recognition.The experimental results show that the recognition accuracy rate is 97.03%.3.In order to solve the problem of high computational complexity and low accuracy of planting yield prediction model for traditional Chinese medicinal materials,a convolutional neural network and long short-term memory network based on the attention mechanism for Chinese herbal medicines production is proposed.Based on convolutional neural networks and long short-term memory networks,an attention mechanism is used to realize automatic weight distribution to solve the problem of information redundancy.The experimental results show that the root mean square error in corn and potato yield data sets are 30.26 and 10.34 respectively,and the Pearson correlation coefficients are 83.26% and 80.73% respectively.4.Designing and developing a service platform for traditional Chinese medicinal planting.The platform adopts a Browser/Server(B/S)architecture,and chooses Java open source lightweight framework Spring Boot for platform development.The platform includes user management,data management and planting management functions,so as to realize the assisted decision of traditional Chinese medicine planting for the prediction of the distribution of suitable areas,identification of leaf diseases and yield prediction,providing auxiliary decision-making for the planting process of traditional Chinese medicine from three aspects: suitable area distribution prediction,leaf disease identification and yield prediction.