Research On Greengage Grade Intelligent Cognition Method Based On Deep Learning And Semi-supervised

As people's living standards improve,consumers pay more attention to the quality of fruits.However,the low level of commercial processing after fruit picking has become a major factor limiting the added value of domestic fruits and the competitiveness of the international market.There are many commercialization steps for fruit postpartum,and grading is one of the key steps.Achieving accurate and automatic grading of fruit grades has become a prerequisite for the modernization of the domestic fruit industry.As a kind of medicine and food resources with multiple health functions,greengage is well received by the masses.In order to avoid the influence of subjective factors such as operator experience and responsibility,the performance of manual sorting mode is the main technology of fruit grading.In view of the difficulty in the calibration of fruit samples in the supervised learning cognition method,the feature space abundant representation and the lack of generalization ability of the classifier,and the posterior statistics of cognitive results,imitating human repeated reinterpretation of information interaction cognitive mode,this paper explores a kind of semi-supervised intelligent cognition method with cognitive result entropy measure index constraint based on deep learning,in order to improve the greengage grade recognition rate.The main work of this dissertation is list as follows:(1)The semi-supervised learning mechanism based on collaborative training calibrates the unlabeled greengage image samples with strong reliability,expands the training of greengage image sample set,and constructs a abundant representation adaptive architecture convolutional neural network.The establishment of the greengage image abundant representation of multi-layered features space from the local to the global multilayered surface.(2)Introducing decision attribute information of greengage grade image Based on the variable precision rough set theory,under the uncertainty condition of finite universe,from the perspective of information theory,a cognitive intelligent decision information system model is developed to train the multi-dimensional surface of greengage grade to abundant representation of multi-layered features space.A stochastic configuration network pattern classifier with tens of local approximation capability is designed to construct a classification criterion for multi-level minimal feature space.(3)Based on the generalized error and generalized entropy theory,a latent error and generalized entropy theory is used to define the cognitive semantic error entropy measure of greengage grade cognitive error.The index measures the credibility of the uncertain cognitive results of the greengage grade in real time,and provides a quantitative basis for the operational mechanism of the intelligent feedback cognition.(4)Based on the constraint conditions of the uncertainty entropy error entropy measure index,the feature space adjustment mechanism of the greengage grade is constructed,and the self-optimization adjusts the inner cognition feature efficiency,the middle layer cognition feature level and the outer unlabeled sample calibration reliability,and updates.the greengage grade multi-layered features space,classification criteria and calibration of the unlabeled sample set are graded and re-cognized for the low-reliability greengage image samples.In order to validate the superiority of the proposed greengage grade semi-supervised intelligent cognitive model,5400 images of greengage are selected as sample library,and the effectiveness and feasibility of this method are verified by MATLAB simulation.The experimental results show that the average recognition rate of the proposed method reaches 98.32%,which is better than other cognitive methods.