| Apple anthracnose is one of the most common diseases of apple.In the early stage of infection,the infection character is not obvious.But with the longer infection time,the character of the infected area is similar to that of penicillium disease and bruised fruit.The damaged area loses water and its color becomes darker.If apples infected with Anthracnose in the early stage cannot be detected in time,the infected fruit will quickly infect other high-quality fruits.According to the different time of apple anthracnose fruit infection,the cause of infection can be inferred.Pesticides with different concentrations and capacities can be used more scientifically and environmentally.In this paper,hyperspectral imaging technology was used to collect the sample data of apple anthracnose,penicillium,bruised fruit and apple anthracnose at different infection times.Combined with chemometric methods,qualitative discrimination models were established to distinguish the infection and infection time of early apple Anthracnose.The main research contents and conclusions of this paper were summarized as follows:(1)The spectra of apple anthracnose,penicillium,bruised fruit and healthy fruit were extracted,and then the average spectra of all the spectra of the four kinds of samples were obtained.The spectral differences of the four kinds of samples were seen from the waveform of the average spectra.The spectra were pretreated with baseline,MSC,SNV,normalize,first derivative and S-G smoothing spectra respectively.The modeling set and prediction set were divided by K-S algorithm in the ratio of 3:1.PLS-DA,PCR,LS-SVM and RF models were established for the data of different spectral pretreatment methods.All models were compared to obtain the results of early apple anthracnose,penicillium.The optimal model for the discrimination of bruises and healthy fruits is the RF model pretreated by the first derivative spectrum,and the overall discrimination accuracy is 99.1%.(2)The color and texture features of apple anthracnose samples infected for 24 h,48h,72 h and 96 h were extracted,and different qualitative discrimination models were established based on different data preprocessing methods.Comparing the judgment accuracy of all models,the discrimination effect of different infected days of Anthracnose using image features is not very ideal.The best model for judging the infected time of apple anthracnose based on color and image texture features is the PCR model pretreated by normalize,and the judgment accuracy of the model is 65%.(3)The average spectra of four types of samples were compared by extracting the spectra of anthrax at different infection times,and then the spectra were pretreated with different types of spectra.K-S algorithm is used to divide the spectral data into modeling set and prediction set,The characteristic bands of the pretreated spectra were screened by CARS,UVE and SPA.Finally,LS-SVM,PLS-DA and PCR models were established.Compared with all models,it was concluded that the optimal discrimination model of different apple anthracnose infection time based on hyperspectral characteristics was UVE-RAW-LS-SVM model,and the overall discrimination accuracy was 95.8%.The results show that hyperspectral can better distinguish and detect the early apple anthracnose infection and different infection time,and is of great significance to the sorting of apple quality and the improvement of fruit quality. |
