【作者】 刘新鑫；

【导师】 韩东海；

【作者基本信息】 中国农业大学 ， 农产品加工及贮藏工程， 2004， 硕士

【摘要】 本文利用自制差分仪对苹果水心病进行无损伤检测分级研究，同时也利用可见近红外连续光谱仪对苹果水心病、内部褐变进行了发病面积比例预测和分级研究。对苹果不同放置位置、不同方法检测水心病和褐变的效果进行了比较。研究结果发现密度法只能分离好果与严重水心病果(4级)，无法对相邻的级之间进行分离。而根据自制差分仪所得的OD810、OD760值做判别分析，果柄向上位置200个样品分到5个组的正确率达到68.5％；果柄向下时达到69％。霉心病果影响OD值，去除霉心病果后，果柄向上位置的总误判率减少到28.19％。连续光谱判断5个级别的水心病可达到83％准确率，果柄向上和向下有相似的结果，果柄水平位置的判断结果较其它两个位置差。逐步多元回归(SMLR)和偏最小二乘(PLS)预测水心病发病面积相关系数为0.932、0.925。水心病果在贮藏过程中700～900nm光谱随时间OD值变大，到第11周时开始有下降的趋势。连续光谱预测褐变面积比例时，果柄水平位置预测效果SMLR和PLS(R~2=0.898、0.765；RMSEP=9.35、11.9)模型都稍好于果柄向上位置(R~2=0.893、0.745；RMSEP=9.65、13.9)，而ΔOD(715-750)和ΔOD(715-810)非线性模型的预测效果几乎相同(RMSEP=9.74、9.75；RMSEP=10.0、10.07)。两位置SMLR模型预测效果最好，ΔOD(715-750)和ΔOD(715-810)非线性模型次之，PLS模型预测效果相对较差。若选择715nm、750nm、810nm作为特征波长判别分析结果可以正确分离97.32％的褐变果和91.38％的好果。更多还原

【Abstract】 In this thesis, some primary work was done for the study of detecting the watercore-apple by using the self-made difference-meter, and predicting the internal disorder degreed of watercore and breakdown by using visible-near-infrared spectroscopy. The result displayed that density-methods can classify sound apple and serious watercore apple (class4) only. A classifier based on the discriminant analysis both OD810nm and OD760nm was used to segregate 200 samples to each grade. It is corrected classified 68.5% at the place of stem-up, and 69% at the place of calyx-up. Apple mould core affected the optical density. Excluded the mould core apple, total misclassification decreased to 28.19%. The continuous spectrum correctly classified the five grades watercore was 83%. The stem-up and calyx-up have the similar result, but the result of the stem-level place is worse than them. SMLR and PLS calculation models applied to independent prediction sets gave R=0.932, 0.925. Spectral OD value increase was observed in the rage 700~900nm during the watercore apple storage, but it became to decline after 11 th weeks. Using the continuous spectrum predicted the internal browning disorder rate, SMLR and PLS models at stem-up place (R2=0.898, 0.765; RMSEP=9.35, 11.9) were better than another place(R2=0.893, 0.745; RMSEP=9.65, 13.9). However, the A OD(715-750) and OD(715-810) non-linear regression models had the similar predicted result at both places (RMSEP=9.74, 9.75; RMSEP=10.0, 10.07). SMLR model was the best at all of calculation models. Two non-linear models took second place. If 715nm, 750nm, 810nm three waves were selected as the character wave, 97.32% of the breakdown apple and 91.38% of sound apple were correctly classified by discriminant analysis.更多还原