Fourier Near-infrared Determinate Soyabean, Rape-seed Meal Routine Chemical Composition And Granularity To Model Influence

The calibration models to predict the routine chemical composition inprotein feeds such as soybean meal and rapseed meal were developed by applying onFourier near infrared spectroscopy;also the influences of the feeds granularity on themodel were investigated. Seventy-nine soybean meal and 73 rapseed meal sampleswere collected and crude protein predication models were development under 40 and60 mesh sizes granularity based on DM condition in separately. The optimizationgranularity was confirmed based on these two feedstuff, the airdry postural CP, H₂O,EE, Ash, ADF and NDF near infrared prediction model of the soybean meal andrapeseed meal were established under 40 mesh sizes granularity. At the same time themodel were evalucated through the internal and external test.Finally the application ofthe model developed under the 40 mesh sizes granularity was evalucated by analyzingthe samples in difference granularty ,the samples were not crushed, crushes through20, 40, 60, 80 mesh sizes in separately. The results were as follows:1. Under the DM condition, there shows no difference in soybean meal CP modelestablished under 40 and 60 mesh sizes granunity condition, while the rape-seed mealmodel established under 60 mesh sizes shows slightly better than that of the 40 meshsizes. The soyabean calibration model decision coefficient (R²) were 97.87 and 97.45in respectively, correction mean-square deviation (RMSEE) and relative deviation(RSD) were 0.335, 0.359 and 0.669, 0.723 seperatety. Internal crossing test R² andRSD were 96.18, 96.53 and 0.810, 0.786 respectively. The rape-seed meal 40 and 60mesh sizes calibration model R² and RSD were 95.33, 98.0 and 0.565,0.388,respectively; internal crossing test R² and RSD were 87.81, 94.86 and 0.808,0.549,respectively. External inspecting of the soybean meal and rape-seed meal we foundthe model developed under 40 or 60 mesh sizes granularity could well function inpredicating the samples under the same granularity(R² were 98.53, 98.68 and 91.77,91.83, RSD were 0.479, 0.447 and 0.564, 0.672 respectively). The crossing forecastresult was not so good (R² were 96.76, 96.32 and 89.25, 91.25, RSD were 0.707, 0.771 and 0.561, 0.749 respectively) but have no difference.The result indicated whenthe feeds was crushed into some extend, for example the GB routine analysis standard40 mesh sizes, it has already meet the requirements of the near-infrared analysis.2. The calibration model of H₂O, CP, EE, NDF and the ADF in soybean mealunder 40 mesh and the airdry postural condition, the R² were 98.9, 99.0, 99.07, 97.93and 97.00, RMSEE were 0.07, 0.21, 0.07, 0.34 and 0.25, RSD were 0.62, 0.47, 6.37,2.92 and 3.12, respectively. The crossing and the external inspection R² were above95, and the correspondence standard deviations were small except that of the EE,which were 6.49 and 6.86 in separately. The R² and RSD in Ash calibration modelwere 90.69, 1.09 respectively, need further improve the model accuracy.The R² of rape-seed meal H₂O, Ash, NDF and ADF calibration model were 98.07,96.57, 98.09, and 97.73, RMSEE were 0.096, 0.135, 0.821 and 0.901, RSD were 0.91,1.79, 0.82, and 0.90, respectively. All of the internal crossing and external inspectionR² had achieved 95 except that of the Ash (93.59), each standard deviation (RMSECV,RMSEP: 0.10-1.2) and RSD (0.81-4.55) were also small in value, which may indicatethe well application of the models. The R² of the rape-seed meal CP, EE model were96.16 and 99.00, RSD were 0.71 and 9.21, after the rejection of these abnormalsamples the R² of the rape-seed meal CP, EE model could achieved above 85. Thecorssing and enteral inspection results showed the established models could apply onsample CP and EE content prediction, but the accuracy was not so good.3. The calibration models established for soybean meal and rape-seed meal wereapplied in separately on the same samples with different granularity, the acquirednear-infrared predicated values were pairs T-test with correspondence chemicalanalysis value. The results shows the models could well applied on 40 and 60 meshsizes samples. The predicated values for soybean meal with 40 mesh sizes (r=1.000,P=0.111)and repeseed meal with 60 mesh sizes(r=0.999, P=0.533) were much matchto the correspondence chemical analysis values, while the samples without crush orcrush for 20 mesh sizes shows signifantly difference between the predicated valuesand the chemical analysis values test by pairs T-test (r:0.000～0.019). When thesamples were crushed through 80 mesh sizes, the correlation coefficient haveimproved to some extend (r:0.981～1.000).Above findings indicated that, the soybean meal and rape-seed meal crushed 40 mesh sizes is well, further crushed 60 is not improvement; routine chemicalcomposition calibration model establishes which under 40 mesh sizes granularitybesides soybean Ash, rape-seed meal CP, EE, the model forecast effect is good(R2＞95, RSD＜3.7). The external inspection showed this model may analysis to 40item and 60 item between samples.