Research On Image Recognition Of Ethnic Medicinal Plants Based On Deep Learning

Image recognition is one of the research directions in the field of computer vision,and as computer technology continues to move forward and the world is moving towards an intelligent era,image recognition technology is gradually being applied to all areas of life.Ethnic medicine is an important part of China’s pharmaceutical industry,and improving the accuracy and efficiency of ethnic medicine plant recognition in practical applications is of great research value to the development of ethnic medicine health care.Therefore,the use of computer image recognition technology to construct a model suitable for the image recognition of ethnic medicinal plants and establish an ethnic medicinal plant recognition system is of great practical significance in promoting the dissemination and popularisation of ethnic medicinal knowledge.This paper proposes a recognition model for ethnic medicinal plant images based on deep learning image recognition methods,and completes an automatic ethnic medicinal recognition system.When constructing a dataset of ethnic medicinal plant images,due to the specificity of its data sources,there are often problems such as unbalanced dataset categories,small sample size and more complex image backgrounds,which add difficulty to the image recognition task and affect the classification performance and prediction effect of the model.In response to the above problems,a series of research works have been carried out in this paper,the main contents are as follows:(1)An improved Res Net-based method for the recognition of ethnic medicinal plant images is proposed to address the class data imbalance problem in the ethnic medicinal plant recognition task.The method proposes an improvement strategy in the network structure and loss function.The first convolutional kernel of Res Net34 is replaced with a cascaded small convolutional kernel,and a non-linear activation layer is added to improve the discriminative ability of the network;a self-adaptive normalisation method is used to improve the robustness of the model;a Dropout layer is added to set an appropriate discard rate to reduce the overfitting phenomenon.The optimized Focal loss function is used instead of the cross-entropy loss function,the balance parameters are improved,and the loss weights of various types of ethnic medicinal samples are dynamically adjusted by adding weighting factors.The effectiveness of the method was demonstrated through experiments,which improved the overall classification performance of the network model and the classification accuracy of a few classes,and better solved the class imbalance problem of the dataset.(2)To address the problems of small sample size of ethnic medicinal plant image datasets and difficulties in image feature extraction due to complex image backgrounds,the number of categories was increased on the basis of the datasets in the previous sections to construct the Tibetan MP dataset,and an image recognition method(SE-Res Net34-Transfer)that embeds the Squeeze-and-Excitation mechanism Res Net and combines migration learning was proposed.The method performs migration learning on the pre-trained model of Res Net34 on Image Net to reduce the overfitting phenomenon caused by the insufficient amount of data,while introducing the Squeeze-and-Excitation mechanism in the shallow layer of the network to focus the network on the key features in the complex images of the background,and finally fine-tuning the model.Experiments show that the proposed method achieves better convergence accuracy and lower training error on the Tibetan medicine dataset,and the effectiveness of the model is verified on several publicly available plant and non-plant image datasets,while the SE-Res Net34-Transfer has higher recognition accuracy when compared with other network models.This demonstrates that the method can effectively improve the performance of ethnic medicinal plant image recognition and can be used for the automatic recognition of ethnic medicinal drugs,providing a new idea for the modernization and development of ethnic medicinal drugs.(3)Based on the above research,this paper uses the Python programming language and tools such as Pycharm and Hbuilder X to design and develop an image recognition system for ethnic medicinal plants,which realizes the visual display of the image recognition results.The purpose is to provide a reference for workers in related fields to identify ethnic medicines and help.