Image-based Hand Feature Recognition System For Hand-foot-mouth Disease

With the rapid increase in the level of computer hardware and the proliferation of medical images,it has gradually become a trend to convert three-dimensional actual scenes into two-dimensional planar images suitable for computer to process and analyze.Different from traditional diagnostic methods,the computer software system can complete the auxiliary diagnosis of children's hand-foot-mouth disease through the recognition of children's hand feature images.It can not only make full use of medical images,greatly save medical resources,but also realize off-site diagnosis and reduce the burden on children.In this paper,children's hand features are used as recognition objects.Using the convolutional neural network model to recognize the children's hand feature images,a handfoot-mouth disease hand feature recognition system was designed and implemented.The main research contents of this article are as follows:1.Hand feature image recognition based on Inception-v3 migration model: pre-process the collected child hand feature images,complete image recognition through Inception-v3 model migration learning.Aiming at the image recognition results,the training parameter adjustment strategy of the convolutional neural network model is proposed,and the advantages and disadvantages of the Inception-v3 migration model in the recognition of hand feature images are analyzed.2.Hand feature image recognition based on small-scale convolutional network concatenation: in view of the overfitting of the Le Net-5 model on the hand feature image data set,the Le Net-5 model was optimized and improved in four points: the feature map before convolution is filled with 0,the pooling layer is implemented with a local response normalization operation,the fifth convolutional layer is changed to a fully connected layer,and a Dropout layer is added before the last Softmax layer.Analyze the advantages and disadvantages of the Inception-v3 migration model and the Le Net-5 improved model,add the inception module to the Le Net-5 improved model,build a new series model Le Net-5 + inception model,increase the depth and width of the model,improve the model recognition accuracy,and improve the recognition accuracy of hand feature images to 91%.3.Hand-foot-mouth disease hand feature assisted recognition system design and implementation: for small-scale data sets such as children's hand feature images,compared with Inception-v3 model transfer learning,the series connection of small-scale convolutional neural networks,and the addition of the inception module to the Le Net-5 improved model have higher recognition accuracy.The Le Net-5 + inception model is internally encapsulated,and a hand-foot-mouth disease hand feature recognition system is designed and implemented.The entire recognition process is fast,smooth,and efficient,requiring only 1-2 seconds.