| Apples suffering from watercore disease are commonly known as "sugar heart apple".C Compared with normal apples,the diseased tissue became significant sweetening due to the accumulation of sorbitol.Watercore apples’ prices are higher in the market because they are harmless to humans and their taste are better than that of disease-free apples.Watercore disease will shorten the shelf life of apples,so when apples are picked,apples with watercore disease should be picked first.In order to reduce the losses caused by the increase of water core disease and increase the economic benefits by selling watercore apples separately,it is urgent to have a fast,accurate and non-destructive detection method to distinguish them.Near-infrared spectroscopy technology has unique advantages in apple watercore disease detection and is a convenient and effective method.Convolutional neural network(CNN)can autonomously extract valid feature structures from complex spectral data to learn and has stronger model expression ability than traditional models.Whereas,applying CNN to the analysis of apple near-infrared spectroscopy data has not yet been studied.In order to solve the above problems,this paper combined CNN with near-infrared spectroscopy and performed the following work:First,the spectral data collected from water core apples and normal apples were simply preprocessed to reduce spectral noise.Then,based on the CNN algorithm,the network structure and parameters were optimized through experiments,and the convolutional neural network model for apple near-infrared spectral data was finally obtained.The convolutional neural network watercore apple prediction model for near-infrared spectroscopy data was compared with the traditional model,the results showed that the prediction accuracy of the convolutional neural network classification is better than the traditional model,the rate of accuracy could reach to 98%,and the dependence on the preprocessing was smaller.Therefore,the convolutional neural network can use the near-infrared spectroscopy data to make an effective identification of apple watercore disease.Finally,a set of visual,real-time watercore apple online detection software was designed and developed,the software calls the packaged CNN discriminant model interface to perform real-time online analysis of the collected apple spectral data.Software test results showed that the rate of discriminant accuracy reached to 83.3%and the detection speed reached to 1.5 seconds per apple,which can meet the needs of the industrial apple water core disease online detection.In this paper,the convolutional neural network was applied to detect near-infrared spectroscopy of watercore apple.An effective convolution neural network modeling method for spectral classification was proposed.Based on the classification model,a software that can meet the needs of industrial for watercore apple detection was developed.The paper provided new ideas for watercore apple online detection,and it has a positive significance for increasing the added value of watercore apple and promoting the fruit industry classification level. |
