基于近红外光谱特征提取的花椰菜灰霉病早期检测

Early Detection of Cauliflower Gray Mold Based on Near-InfraredSpectrum Feature Extraction

花椰菜在生长过程中容易感染灰霉病而导致产量减少,现有的分选方法难以在早期检测到感染灰霉病的花椰菜。应用近红外光谱技术实现花椰菜灰霉病的早期判别检测,对花椰菜病害防治意义重大。以接种灰霉菌孢的花椰菜为研究对象,首先,采集对照组和处理组花椰菜的近红外光谱曲线并进行去噪处理,获取4个批次共608个样本(接菌0.5,1,2和3 d每日的健康和染病花椰菜各76朵)在500~2400 nm波段范围内的光谱曲线。同时测量花椰菜样本的多酚氧化酶(polyphenol oxidase,PPO)、过氧化物酶(peroxidase,POD)与丙二醛(malondialdehyde,MDA)的活性值,采用单因素方差分析(analysis of variance,ANOVA)对单一批次的健康和染病花椰菜品质指标进行统计分析。然后,采用K-S算法(Kennard-Stone)将单天的样本划分为校正集(114个样本)与预测集(38个样本),使用竞争性自适应重加权算法(competitive adaptive reweighted sampling,CARS)进行4个批次的花椰菜样本的光谱数据特征波段提取,并基于偏最小二乘回归(partial least square regression,PLSR)算法建立单一批次判别模型和组合批次判别模型。结果表明:在接菌早期,用肉眼无法实现染病花椰菜样本的识别,仅在染病第3 d后部分染病样本病害特征明显时可实现判别。测定对照组和处理组花椰菜品质指标后发现:染病2 d后,对照组和染病组样本的所有品质指标均存在显著性差异(p<0.05),但在第0.5 d时各项指标均无显著性差异,仅在第1 d时MDA值出现显著性差异,说明从品质指标上无法判别早期染病花椰菜。建立单一批次下的PLSR判别模型后表明:第一批次样本(0.5 d)所建模型的判别准确率达到了94.74%,预测集均方根误差为0.835,第二至第四批次(接菌1~3 d)所建判别模型准确率达到100%,表明PLSR模型可以实现单一批次下早期染病花椰菜样本的检测;PLSR组合判别模型在第0.5 d和第1 d判别准确率分别达到了92.11%与97.37%,可以判别出大部分的患病花椰菜,但是PLSR组合批次建模效果低于PLSR单一批次建模。结果表明,基于近红外光谱技术,通过CARS算法提取特征波段结合PLSR模型的建立,可以在早期检测出感染灰霉病的花椰菜,为花椰菜灰霉病的早期检测提供参考,具有一定的实际应用价值。

Gray mold easily occurs during cauliflower growth,thereby leading to reduced output.Cauliflower infected with gray mold at an early stage is difficult to detect with existing methods.In this study,near-infrared spectroscopy was used to distinguish and detect cauliflower with gray mold,which is highly significant for the disease control of cauliflower.Taking cauliflower with Botrytis cinema spore inoculation as the research object,this study obtained the near-infrared spectra of cauliflower in control and treatment groups and performed de-noising.The spectra of 608 samples in four batches(76 healthy and infected cauliflowers at 0.5,1,2 and 3 d old each)were acquired within the waveband range of 500~2400 nm.After measuring the activity of polyphenol oxidase,peroxidase and malondialdehyde in the cauliflower samples,and one-way ANOVA was used to statistically analyze the quality indices of a single batch of healthy and infected cauliflowers.The Kennard-Stone algorithm was used to divide each day’s samples into a calibration(114 samples)and a prediction(38 samples)set.Competitive adaptive reweighted sampling(CARS)was then used to extract the feature waveband of the spectroscopic data of the four batches of cauliflower samples,and the discrimination models of single and combination batches were established based on of partial least square regression(PLSR).Results indicated that the naked eye could not identify infected cauliflower samples at the early stage of inoculation and could identify them only 3 d after infection when some infected samples showed evident disease characteristics.The measurement of quality indices of the cauliflower in the control and treatment groups showed significant differences in all quality indices between these groups 2 d after infection(p<0.05);however no significant differences existed in all quality indices at 0.5 d,and a significant difference in MDA value existed only at 1d.These findings suggested that the quality indices of infected cauliflower cannot be discriminated at an early stage.The PLSR discrimination model of a single batch was established,and it showed the following:the discrimination accuracy of the model established for the first batch(0.5 d)reached 94.74%,the root-mean-square error of the prediction set was 0.835,and the discrimination accuracy of the models established for the second to fourth batch(1~3 d)reached 100%.These findings indicated that the PLSR model could detect infected cauliflower samples under a single batch at an early stage.The discrimination accuracy of the PLSR combination discrimination model reached 92.11%and 97.37%at 0.5 and 1 d,respectively,to discriminate a large proportion of infected cauliflower.However,the effect of PLSR combination-batch modeling was inferior to that of PLSR single-batch modeling.Therefore,using near-infrared spectroscopy,extracting the feature waveband through CARS,and establishing a PLSR model can detect cauliflower infected with gray mold at an early stage,thereby providing a reference for the early detection cauliflower with gray mold and has some practical value.