Research On Potato Identification And Classification Based On Hyperspectral Imaging Technology And ACGAN Model

Potato is an important food crop after rice,corn and wheat.It is rich in nutrients.It is a crop that combines grain,vegetables,feedstuff and industrial raw materials.Its cultivation is mainly distributed in Northeast China,North China and East China.The potato industry is of great significance for ensuring food security in China.The quality of potatoes directly affects the quality of their products,and the inspection of potatoes' quality is the key to the potato staple food strategy.At present,potato quality testing is mainly based on traditional physical and chemical testing,experienced recognition,biological testing and other testing methods.These methods have the problems of time-consuming,laborious,long time,high cost,etc.,which cannot meet the needs of rapid testing of potato quality.Hyperspectral imaging technology has the advantages of multiple bands and high resolution ratio.It has great advantages in the non-destructive and rapid detection of potatoes.It can not only reflect the characteristics of the surface of the measured object,but also reflect its internal quality and material composition.However,the hyperspectral imaging data is three-dimensional cube data,which has many wave bands and large information redundancy.If it is directly calculated and processed,the running speed is slow,and the accuracy of the calculation results will be reduced.In order to further improve the speed and accuracy of non-destructive testing of potato quality hyperspectral imaging,it is necessary to continuously study the data processing method of hyperspectral imaging,and then promote the further development of potato quality testing.The thesis takes a variety of potato stem as the object,and uses hyperspectral imaging technology,comprehensively uses spectral imaging analysis technology,chemometrics and machine learning to study the classification and identification of potatoes.To explore the feasibility of identifying classification of potato hyperspectral imaging data based on auxiliary classifier generative adversarial nets(ACGAN)network model with a small sample.The main research contents and conclusions are as follows:(1)A total of 429 potato samples were selected from Chuanyin,Daxiyang,Feiwuruita,Houqihong,Kexin,Qingshu 9 and 226,which were washed,dried and stored for 7 days.And collect hyperspectral imaging data in different varieties.(2)The thesis adopts multiplicative scatter correction(MSC),standard normal variate(SNV),SG(Savizkg-Golag)smoothing method and derivative method to smooth and reduce the noise of spectral dimension data for pretreatment.(3)Research on the use of convolutional neural networks(CNN)model to perform classification and recognition experiments on the processed data.After repeated experiments and optimization of model parameters,the average of the test results is finally taken.The results are better than other methods,and the classification accuracy can reach 92.97%.(4)The thesis explores the method of potato identification and classification.The spectral dimension data of pre-processed 125-band ROI area is converted into 64 x 64 single-channel images as input,and the ACGAN model is used for potato classification and identification experiments.At the same time,CNN,partial least squares discrimination analysis(PLS-DA),principal component analysis(PCA)+support vector machines(SVM)and semi-supervised generation of adversarial networks(SGAN)are models used simultaneously for classification comparative experiments.The experimental results show that the ACGAN model proposed in this paper has a classification accuracy rate of 98.45%,which is higher than other models.The research results show that the ACGAN model is feasible in terms of accuracy of the potato classification study with a tiny sample.