Black Fungus Variety Identification Method Based On Near Infrared Spectroscopy Characteristic

The rapid recognition of phenotypic characteristics of edible mushroom fruiting bodies has important theoretical significance and practical value for the evaluation,breeding,and cultivation of edible germplasm resources.The existing morphological and physiological phenotypic detection of the growth process of edible fungi is not systematic enough,and there is a lack of fast and accurate detection methods for Auricularia auricula varieties.At present,the operation process of DNA molecular marker technology is complex,time-consuming,and expensive.However,manual methods rely on visual observation and experience to determine the variety of fruiting entities,which are easily affected by subjective factors,and have shortcomings such as being prone to errors and difficult to ensure objective accuracy.Therefore,this article proposes a Auricularia auricula variety recognition method based on near-infrared spectral features.This study used four Auricularia auricula varieties,Cheng D,Heifeng,Heishan,and Xu1,as experimental subjects to collect raw spectral data of Auricularia auricula.By analyzing the multidimensional spectral characteristics of pre processed black fungus,the characteristic wavenumbers of Auricularia auricula near-infrared spectral data were extracted,and a fast identification model for Auricularia auricula varieties was established.A Auricularia auricula variety recognition system was integrated and developed,Completed the research on the identification method of Auricularia auricula varieties based on near-infrared spectral features.The main research content is as follows:(1)Spectral data collection of Auricularia auricula.Four Auricularia auricula varieties mainly cultivated in the Lindian area of Heilongjiang Province were used as research objects.They were dried in a natural environment and ground into powder form using a laboratory crusher.Samples were filtered into particles of uniform size using a 6-mesh experimental sieve.During the scanning process,a volume of 30ml of the sample to be tested is taken each time,and data is collected using the Tango near-infrared spectrometer from the German company Brooke to obtain the original spectral data of Auricularia auricula.At the same time,accurately corresponding the spectral data of Auricularia auricula to its labeled varieties,a total of 600 sets of spectral data were collected,with 150 sets of spectral data for each variety.(2)Pre processing of Auricularia auricula spectral data.For the obtained Auricularia auricula spectral data,SPXY algorithm was selected to divide the sample set,and multivariate scattering correction(MSC),standard normal variable transformation(SNV),and de trend(DT)algorithms were applied to preprocess the original spectral data.Four PLSR models were constructed.Among them,the R_p~2 of the DT algorithm increased by 0.0207,0.0159,and 0.0179compared to the RAW,MSC,and SNV algorithms,respectively.The R_p values increased by0.0856,0.0081,and 0.0108,respectively.The RMSEP values decreased by 0.0987,0.0358,and0.046,respectively.The SEP values decreased by 0.0528,0.0505,and 0.056,respectively.The results showed that the DT algorithm had the best performance in processing the original spectral data of Auricularia auricula,effectively highlighting the feature information of the spectral data.(3)Extraction and optimization of characteristic wavenumbers from Auricularia auricula spectral data.Based on the spectral data processed by the DT algorithm,competitive adaptive reweighted sampling(CARS),continuous projection(SPA),and uninformation variable elimination algorithm(UVE)were used to extract the feature wavenumbers of Auricularia auricula spectral data.150,6,and 509 wavebands were extracted,respectively,reducing the total number of original 1845 bands by 91.87%,99.67%,and 72.41%,greatly reducing the dimensionality of spectral data.Through the established PCR model,the R_p~2 value of the CARS algorithm was 0.9886,which increased by 0.0206 and 0.0064 compared to SPA and UVE,respectively.The RMSEP value was 0.1193,which decreased by 0.0807 and 0.0299 compared to SPA and UVE,respectively.The results indicate that the CARS algorithm can effectively select spectral data features of Auricularia auricula,providing reliable wavenumber indicators for constructing a Auricularia auricula variety recognition model.(4)Build a Auricularia auricula variety recognition model.Taking the characteristic variables selected by CARS algorithm as the input data index of the recognition model,a variety recognition method of Auricularia auricula was established by combining the back propagation neural network(BP),radial basis function neural network(RBF)and convolutional long short-term memory neural network(CNN-LSTM).The recognition accuracy of the constructed CNN-LSTM neural network is 99.32%,with a root mean square error value of 0.0089.Compared with BP and RBF neural networks,it has improved by 1.35%and 0.67%respectively,and the root mean square error value has decreased by 0.091.CNN can extract features from spectral data through convolutional and pooling layers,while LSTM can use sequence relationships for spectral information memory and forgetting,enabling the CNN LSTM dual stream fusion network to better capture data features and relationships,establishing a new method for rapid identification of Auricularia auricula varieties.(5)Development and integration of a Auricularia auricula variety recognition system.The system is programmed using Python language and designed with a user interface based on the Py Qt5 development framework.The system design and development are completed by combining key technologies for spectral analysis and processing of different varieties of Auricularia auricula.This system can analyze and process spectral data,covering classic processing methods commonly used in near-infrared spectroscopy technology,mainly including file module,preprocessing module,data calibration module,feature extraction module,model training module,and variety recognition module.This achievement can achieve variety recognition of Auricularia auricula sample data and provide a visual analysis platform for the detection of edible mushroom phenotypic characteristics.