Research On The Technology Of Navel Orange Quality Automatic Detection And Classification Based On Machine Vision

The detection and classification of fruit quality is an important step before its maturity,because of the quality of fruit directly affects its Economic benefits and market competitiveness.In the fruit quality detection,the traditional classification methods mainly include manual detection classification and mechanical detection classification.Manual detection consumes a lot of manpower,and financial resources,and there are different degrees of difference between the specific classification criteria of each person.The efficiency and accuracy and classification is easily affected by people's emotion,fatigue and physical condition and other factors;Mechanical detection will cause a certain degree of mechanical damage to the fruit,and it is easy to bruise fruits in the process of detection.Once the fruit is damaged,it will shorten the shelf life and slowly become rotten.The detection effects of the existing traditional methods are are not ideal and will directly affect the maximum economic benefits that fruit should bring.Therefore,it is very necessary to find a technology that can achieve automatic detection and classification of fruit quality.Machine vision can accomplish the tasks that require observation and judgment by the human eyes,and is ideally suited for the judgment of a large number of repetitive movements that make eyes fatigue.This paper takes navel orange as an example to apply machine vision technology to the automatic detection and classification of navel oranges,which can improve the efficiency of detection and classification and ultimately achieve the automatic NDT classification of navel oranges.In order to overcome the shortcomings of traditional detection and classification methods,the main contents and innovations of this paper are as follows:(1)In order to overcome the shortcoming that will lose some useful information of traditional method of denoising,this paper studies the image denoising method based on dictionary learning in Chapter 3.It uses the K-SVD(K-Singular Value Decomposition)algorithm to complete dictionary training.At the same time.the principle of the algorithm and the denoising process are studied.This method uses two key techniques that called sparse coding and dictionary training,and making the final denoised image have a higher peak signal to noise ratio.Through the simulation analysis of different algorithms,the experimental results show that this method can preserve the original information of the image more completely,so as to achieve a better denoising effect.(2)The existing traditional methods are often only used for independent identification of the single feature,and the detection results are not ideal.In order to further improve the detection rate,this paper proposes a method to implement the detection and classification of navel orange based on feature fusion with the help of the HALCON platform In Chapter 4.This method extracts the three features of the navel orange that include the size of the lateral diameter,the color of the peel and the defects of the surface.First of all,these three features are individually detected and classified,and the corresponding classification results are given.Then the three features are classified together.Finally,the results of comprehensive detection and classification of navel orange are completed by using the method of feature fusion.(3)Most of the existing research focuses on the use of manual design features to complete the identification.These methods are time-consuming and effort-consuming.Therefore,it is necessary to find a method that can be used to automatically distinguish features.we proposes the classification of navel orange quality based on the deep learning method In Chapter 5,and collected 4 kinds of navel orange pictures,they are 1000 high quality pictures,1000 good pictures,1000 qualified pictures and 1000 unqualified pictures,and then completed the construction of navel orange data set.The supervised learning method is used to train the deep network of navel orange detection.a network model for automatically classifying the quality of navel oranges was constructed based on the improved residual network,The potential distribution of sample data was automatically learned.The supervised learning method was applied to complete the training of the model,and compared with the detection and classification results of other methods.