Research On The Identification Method Of Feedstuff Species Based On Spectral And Image Information

In the process of feed processing,raw materials may not enter the fixed batching bin due to the mechanical failure or operation error of raw material conveying equipment,rotary distributor,etc.Currently,most feed mills still use manual sampling at the sampling port of the slip pipe for sensory identification of incoming raw material types.This study collects spectral and image information of feed materials,sets up the feed applied to the structure recognition model,recommends the feed material type recognition method based on spectral and image data,as well as the material type recognition method based on the fusion of spectral and image information,and provides a reference for the automation of incoming feed materials to meet the development needs of automation and intelligence in the feed production process.The following are the key findings and findings of the study:(1)Ten types of bulk feed ingredients were collected,including 60 copies each of corn flour,wheat flour,rice flour,bran,flour,puffed corn,soybean meal,cottonseed meal,peanut meal,and fish meal.Using the independent visible and near-infrared spectral information acquisition system,the raw spectral data of the ten types of raw materials were collected separately,and the average spectrum was obtained as their spectral data.The images of the ten types of raw materials were collected separately using the MV-VS1200 S machine vision experimental development platform,and the six R,G,B,H,S,and V values indicating the color characteristics of the images were extracted using Matlab and Python.The six texture feature values of contrast,dissimilarity,homogeneity,energy,correlation,and angular second-order moment(asm)were extracted using Python,and the average of each was used as the texture feature parameter.For the establishment and validation of the raw material species identification model,the SPXY algorithm divides the calibration group and the prediction group into a 7:3 ratio.(2)Research on the identification method of feed ingredient species based on spectral information was conducted.Four spectral preprocessing methods,namely,Smoothing,Normalizing,Multiple Scattering Correction(MSC),and Derivatie,and three feature wavelength extraction algorithms,namely,Competitive Adaptive Weighting Algorithm(CARS),Uninformative Variable Elimination(UVE),and Continuous Projection(SPA),were used to establish partial least squares discriminant analysis(PLS-DA),support vector machine(SVM),artificial neural network(ANN),and baseline model,respectively.The optimal preprocessing,feature wavelength extraction,and modeling methods were determined using the accuracy of raw material species identification as the evaluation index.The results show that the accuracy of the SVM recognition model built with normalized preprocessing is 93.10% for the correction set,91.11% for the prediction set,and 98.33%for the cross-validation;the accuracy of the SVM recognition model built with the competitive adaptive weighting algorithm(CARS)feature extraction method is 93.56%,92.78% for the prediction set,and 90.00% for the cross-validation accuracy.The raw material type recognition model built based on spectral information can achieve raw material type recognition,but the recognition accuracy needs to be improved.(3)To research the identification method of feed ingredient types based on image information.Support vector machine(SVM)and artificial neural network(ANN)models were established using color,texture,and color and texture features as input variables,respectively,and the optimal feature variables were determined using the accuracy of raw material species recognition as an evaluation index.The results show that the SVM model established with color and texture features as input variables can achieve 99.78% accuracy in the correction set and 100% accuracy in the prediction set.The raw material type recognition model based on the image information can achieve the raw material type recognition very well.(4)To investigate a method for recognizing feed ingredient kinds using a combination of spectral and image data.The information is fused into two forms: the feature layer and the decision layer.In the feature layer,the partial least squares discriminant analysis(PLSDA),support vector machine(SVM),artificial neural network(ANN),and baseline models were built using three features: spectral combined color,spectral combined texture,and spectral combined color and texture as input variables,respectively.The results reveal that the ANN model created with a spectrum combination with color and texture characteristics has a correction set accuracy of 94.05 % and a prediction set accuracy of 100 %;in terms of decision layer fusion,the accuracy of the SVM recognition model based on the D-S evidence fusion theory is 94.76% for the correction set and 91.11% for the prediction set.The feature layer level fusion recognition effect is better,and the feed ingredient type recognition method based on the fusion of spectral and image information can accomplish the qualitative recognition task well.By comparing the recognition effects of the models based on spectral information,image information,and information fusion,the information fusion method can well handle feed materials with different color and texture characteristics,and the model results are more stable,which can well solve the recognition problem of multiple types of feed materials and provide a new idea and method for solving the recognition of feed materials before they are put into the warehouse.