Inspection Of Melon’s Sugar Content Based On Marginal Fisher Analysis And Artificial Neural Network

Nondestructive testing is a detection method that can test the internal quality aswell as the structure of an object. A major advantage of nondestructive testing is itwill not damage to the bodies. At present, most of the nondestructive testingequipment is dependent on the foreign technology and the price of the equipment ishigh and has low accuracy analysis. In More Recent Years, the progress ofnondestructive testing science research is very rapid. In recent years, domestic andforeign scholars have been studying nondestructive testing of fruit method, but themethod rarely can apply to in people’s life and production.Melon is one of the crops in Xinjiang and northwest of China witch is enjoyedby a lot of people because of its high sugar content and good taste. Because of thereason of the thick rind and large body, the research of nondestructive testing of sugarcontent in melon is more difficult. Based on a variety of current nondestructive testingmethod, the nondestructive testing method based on dielectric characteristics wasdeeply researched in this paper. The nondestructive testing method was researchedrespectively from the relationship between sugar content and dielectric property in thesingle frequency and the frequency.Firstly, a series of experiments were performed to get the linear functionrelationship of dielectric property parameters and sugar content of pulp and theexperimental results show that the fitting accuracy is very high, so it can be used toprove the dielectric characteristics of melon have a complicated relationship with itssugar content. Neural network is established to predict the sugar content of melon,Quantum particle swarm optimization algorithm was adopted to optimize the RBFneural network to improve the performance of the network. With the benefits ofquantum particle swarm is to prevent fall into local minimum in the optimizationprocess.In more than one frequency, the research of sugar content in melon based on neural network was very difficult because the input dimension is too big. Marginalfisher analysis combined with neural network was proposed to overcome the problemof the training and learning is complex, the time is too long and the forecast precisionis too low. The algorithm can not only make multidimensional data dimensionreduction and more conducive to the neural network prediction, but also to Keep thelocal structure of the data.The results show that the nondestructive testing method based on the marginalfisher analysis method and the quantum particle swarm optimization improved radialbasis function neural network (QPSO-RBF) to forecast the sugar melon has higherprediction accuracy and provide a new effective prediction method for the predictionof sugar content in melon.