Method And Emprical Analysis Of Gemstones Identification Based On Convolution Neural Network

In recent years,with the widespread use of deep learning such as convolutional neural networks,breakthrough have been made in the field of artificial intelligence,especially in the field of image recognition.Therefore,based on the summary of the basic theoretical results and application of convolutional neural networks at home and abroad,and combines the popular TensorFlow and Keras as deep learning framework to construct a convolutional neural network model for image classification of natural gemstones.This paper designs different network structure for small data sets,and have an improved model framework.The main tasks are completed as follows:(1)Firstly,the status of research and trend of convolutional neural network model based on deep learning in the field of image classification are discussed.The basic algorithms needed to understand the network model are introduced.On this basis,the classical models ZFNet,VGGNet13 and VGGNet16 of the convolutional neural network are trained on the established computer platform to compare the rate of recognition and rate of several algorithms.Analyze and study the structure and method that affect the recognition performance of convolutional neural networks,and finally choose VGGNet13 as the benchmark model.(2)In order to train the small gem dataset,an improved algorithm based on VGGNet13 is proposed.By adjusting the parameters of complexity,such as the size of the filter,the image size and the number of layers of the convolutional layer,the network is reduced.The complexity reduces the training difficulty and speeds up the picture recognition speed,which makes the Top-1 accuracy rate reach 93.52%,which is 8.27% higher than the benchmark model's Top-1 accuracy rate,and the model complexity is also reduced by 5 times.At the same time,the adjustment of the improved model parameters,such as Padding fill,Batch_size and optimizer algorithm,makes the Top-1 accuracy rate reach 98.30%,which is 13.81% higher than the accuracy of benchmark model,which is 5.11% higher than the accuracy of complex model.The analysis and data show that the improved convolutional network model established by deep learning can obtain efficient,accurate and reliable automatic classification results based on gemstone images,and provides theoretical and practical reference for the research and demonstration of specific image classification.(3)Based on the trained model,combined with the iOS system platform,an intelligent mobile application of image recognition was developed,which includes for take photo,import photos,download classification.At present,the application implements a series of complete functions from extracting picture data generation classifier to mobile phone using the classifier for classification,and opens up a new method of identification in the field of natural gem image recognition,which is no longer limited.In traditional professional equipment,no cumbersome human intervention is required.Breaking through the limitations of the individual's experience,it provides an auxiliary tool for the identification of jewelry gems.