Design Of Intelligent Composition Matching System In Coffee Vending Machine

With the popularity of coffee drinks and the development of an unmanned economy,many companies have begun research on self-service coffee vending machines.However,at present,most of the self-service coffee vending machines in our country are functional machines,which can neither adjust the appropriate proportion according to the health of customers,nor adjust the proportion according to the needs of customers.Therefore,this paper mainly studies from three aspects of face attribute estimation,speech recognition and fuzzy control,and designs an intelligent proportioning system of ingredients in coffee vending machines.First,this paper studies the face attribute estimation,and proposes a cascade structure of gender recognition and attribute estimation,first judges the gender of customers,and then estimates the age and BMI of customers of different genders through multi-task convolutional neural network.The gender cross experiment has verified the necessity of the cascade structure.When there is no transgender,the errors of age and BMI estimation are smaller.By comparing different network models with the improved VGG-Face model studied in this paper,it is concluded that the improved VGG-Face model has a higher accuracy in estimating age and BMI.The two methods of single-task learning and multi-task learning were compared and experimented.It was concluded that the accuracy of multi-task learning in the experiment was higher in age and BMI.Then,this paper studies the speech recognition when the customer interacts with the coffee vending machine.Because the current scene of the coffee vending machine is in China,the speech recognition tends to study Chinese.In this paper,in order to obtain more accurate MFCC features,first,the residual structure is used to deepen the traditional convolutional neural network layers,secondly,the activation function after the convolutional layer is optimized,and finally the improved convolutional neural network and CTC The structure is combined tooptimize the network model.In order to verify the improved network performance,a comparison experiment of four network models is conducted to prove that the improved network has the lowest word error rate.Finally,this paper uses a fuzzy control system to implement the ratio estimation algorithm.The input of the fuzzy control system includes two parts:the customer's age and BMI(body mass index)estimated by the face attribute estimation model;when the customer interacts with the coffee vending machine The customer's demand for ingredient content obtained by speech recognition.The input variables are inferred according to suitable fuzzy rules,and the content of ingredients that are suitable for the customer's body and meet the customer's needs is obtained.During the experiment verification,the system collected the facial images and requirements of the five testers.After reasoning by the component ratio estimation algorithm,the coffee sugar content in line with the prior knowledge was obtained,which proved the good performance of the system in simple scenarios.