The Detection Of Fish Meal Freshness Index And Mold Counts Based On NIRS

Fish meal is a valuable animal-derived protein feed ingredient that contains significant amounts of protein,fat,calcium,phosphorus,and other essential nutrients.However,during storage and transportation,fish meal is susceptible to spoilage,which not only diminishes the quality of feed products but also poses a potential health hazard to the feeding animals.As such,the freshness and mold count of fish meal are critical indicators that can be utilized to evaluate its quality.The development of rapid and nondestructive testing methods for assessing the freshness,health and safety indices of fish meal holds substantial practical significance.This paper focuses on the use of domestic fish meal and imported fish meal(Peru)as the primary research objects.The spectral information of the fish meal samples was collected through near-infrared technology.Various technical methods such as spectral preprocessing and characteristic wavelength extraction were employed.Furthermore,five modeling methods were utilized to establish quantitative prediction models for fish meal freshness indicators,including volatile basic nitrogen,acid value,p H value,and the total number of molds in fish meal.The optimal prediction model for each index was determined through comparative analysis.The main research findings and contributions are as follows:(1)Sample preparation and testing.Domestic fish meal and imported fish meal(Peru)were collected from feed production enterprises.The fish meal samples were placed in artificial climate incubators under varying temperature and relative humidity conditions,including 15°C,20°C,25°C,30°C,35°C and relative humidity of 40%,50%,60%,70%and 80%RH.Sampling was conducted every two days,and the storage period lasted for 30days.A total of 210 fish meal samples were collected.Based on the applicable national standard methodologies,the levels of volatile basic nitrogen,acid value,p H value,and the total number of molds were assessed individually in the samples of fish meal.Upon statistical analysis,it was observed that,except for the p H value,which exhibited a relatively low coefficient,the coefficients of variation for volatile basic nitrogen(VBN),acid value(AV),and the total number of molds exceeded 20%.(2)The optimal quantitative prediction model of fish meal freshness was established.The Monte Carlo cross-validation(MCCV)technique was utilized to identify and remove any outliers present in the volatile basic nitrogen(VBN),acid value(AV),and p H value measurements of fish meal,and the KS method was used to partition the sample set.The spectra were preprocessed using ten methods,including multivariate scatter correction(MSC),derivative processing(Derivative),and SG smoothing(SG),as well as their combinations.Partial least squares regression(PLSR),support vector regression optimized by particle swarm optimization(PSO-SVR),support vector regression optimized by genetic algorithm(GA-SVR),support vector regression optimized by grey wolf optimization(GWO-SVR),and BP neural network models were separately constructed to predict the values of VBN,AV,and p H value in fish meal,and the best spectral pretreatment method for each model was selected.Furthermore,characteristic wavelengths associated with VBN,AV,and p H value were extracted using competitive adaptive reweighted sampling(CARS),uninformative variable elimination(UVE),genetic algorithm(GA),and successive projection algorithm(SPA).The study findings indicate that the optimal spectral pretreatment method for VBN in fish meal was SG+1st+SNV.The characteristic wavelength extraction method that yielded the best results was CARS,and the optimal prediction model was PLSR.The R_c~2 and R_p~2 were 0.9151 and 0.8423,respectively.The RMSEC and RMSEP values were 4.8689 and 5.8910,respectively,and the residual predictive deviation RPD was 2.0892.The optimal spectral pretreatment method for acid value in fish meal was SG+1st+SNV.The characteristic wavelength extraction method that yielded the best results was CARS,and the optimal prediction model was PLSR.The R_c~2and R_p~2 were 0.9424 and 0.9268,respectively.The RMSEC and RMSEP values were0.2677 and 0.3131,respectively,and the residual predictive deviation RPD was 3.5899.The optimal spectral pretreatment method for p H value in fish meal was SG+SNV.The characteristic wavelength extraction method that yielded the best results was SPA,and the optimal prediction model was GWO-SVR.The R_c~2 and R_p~2 were 0.9616 and 0.9403,respectively.The RMSEC and RMSEP values were 0.0406 and 0.0504,respectively,and the residual predictive deviation RPD was 4.0064.These results indicate that the developed quantitative prediction models for VBN,AV,and p H value in fish meal demonstrated high accuracy.(3)The optimal quantitative prediction model for the total number of molds in fish meal was established.The Monte Carlo cross-validation(MCCV)technique was utilized to identify and remove any outliers present in the total number of molds measurements of fish meal,and the KS method was used to partition the sample set.The spectra were preprocessed using ten methods,including multivariate scatter correction(MSC),derivative processing(Derivative),and SG smoothing(SG),as well as their combinations.Partial least squares regression(PLSR),support vector regression optimized by particle swarm optimization(PSO-SVR),support vector regression optimized by genetic algorithm(GA-SVR),support vector regression optimized by grey wolf optimization(GWO-SVR),and BP neural network models were separately constructed to predict the values of the total number of molds in fish meal,and the best spectral pretreatment method for each model was selected.Furthermore,characteristic wavelengths associated with the total number of molds were extracted using competitive adaptive reweighted sampling(CARS),uninformative variable elimination(UVE),genetic algorithm(GA),and successive projection algorithm(SPA).The study findings indicate that the optimal spectral pretreatment method for the total number of molds in fish meal was SNV.The characteristic wavelength extraction method that yielded the best results was CARS,and the optimal prediction model was PLSR.The R_c~2 and R_p~2 were 0.8755 and 0.8279,respectively.The RMSEC and RMSEP values were 0.1953 and 0.2173,respectively,and the residual predictive deviation RPD was 2.4721.The results indicate that the developed quantitative prediction models for the total number of molds in fish meal demonstrated high accuracy.