英文摘要 (English Abstract):
This study addresses the problem of training deep learning models with limited datasets, a significant challenge in sectors like medical imaging and food quality analysis. To tackle this issue, generative adversarial networks (GANs) will be employed to augment the available data and improve model performance. An innovative approach is introduced here, integrating semi-supervised learning and generative modeling to maximize the use of small datasets in developing robust models. The method involves reversing the conventional distribution of training and testing data to focus on model evaluation and generalization from limited samples. Wasserstein GANs (WGANs) and Semi-Supervised GANs (SGANs), are utilized to supplement datasets with synthetic but realistic examples, enhancing the training process in scenarios of data scarcity. These techniques are applied in the context of visible reflectance spectroscopy to analyze tomato sauces, demonstrating the method’s effectiveness in non-invasively assessing key quality parameters such as oil content, °Brix, and pH. The results show significant improvements in model performance metrics: for %Oil content, overall accuracy increased from 0.47 to 0.66; for °Bx, it rose from 0.65 to 0.71; and for pH measurement, accuracy improved from 0.43 to 0.62. These outcomes highlight the model’s improved capability to generalize and maintain accuracy with limited data.
摘要翻译 (中文)：
本研究针对在数据集有限的情况下训练深度学习模型这一难题展开，该问题在医学影像和食品品质分析等领域尤为突出。为应对这一挑战，研究采用生成对抗网络（GANs）对现有数据进行增强，以提升模型性能。本文提出了一种创新方法，将半监督学习与生成建模相结合，以最大限度地利用小规模数据集构建鲁棒的模型。该方法通过反转传统的训练集与测试集划分方式，聚焦于在样本有限条件下对模型评估与泛化能力的优化。具体而言，采用Wasserstein GAN（WGAN）和半监督GAN（SGAN）生成逼真的合成样本，以扩充原始数据集，从而在数据稀缺场景下有效增强模型训练过程。该技术被应用于可见光反射光谱分析番茄酱的案例中，用于无损评估油含量、可溶性固形物（°Brix）和pH值等关键品质参数。实验结果表明，模型性能指标显著提升：油含量（%Oil）的总体准确率从0.47提高至0.66；°Brix的准确率从0.65提升至0.71；pH值的准确率则从0.43提升至0.62。这些结果充分证明，所提出的方法能够有效提升模型在小样本条件下的泛化能力和预测准确性，为数据受限场景下的光谱分析与质量评估提供了可行且高效的解决方案。

英文摘要 (English Abstract):
The application of Visible-near infrared spectroscopy (Vis-NIRs) to internal quality detection of pomelo with large sizes and thick peels is still challenging. Considering that the peel has a large mass ratio in pomelo and a great added value in industrial processing, a postharvest processing mode of grading after peeling without affecting consumption of pomelos may be an approach to avoid adverse influence of peel on soluble solids content (SSC) evaluation. In this study, we investigated the performance variation of Vis-NIRs for SSC online determination of pomelos with and without peel for the first time. An online detection system for pomelo spectra acquisition was developed, and the spectral characteristic difference of the intact and peeled pomelos was analyzed. Subsequently, the performance of calibration models was gradually improved and comparatively analyzed by various spectral processing and wavelength selection methods. Furthermore, convolutional neural network (CNN) was utilized to explore its ability for feature extraction. The results showed that combined with standard normal variate (SNV) and second order detrending and changeable-size moving window algorithms, the partial least squares regression (PLSR) model using spectra of peeled pomelos achieved the best prediction results. Specifically, the model attained a determination coefficient of prediction (R²p) of 0.88, a root mean square error of prediction (RMSEP) of 0.294%, and a residual predictive deviation (RPD) value of 2.57. This study demonstrates that the peel has a significantly negative effect on the prediction performance and the CNN could be an alternative to conventional PLSR method. Our work may open new avenues for the internal quality assessment of agro-products with complex tissue structure.
摘要翻译 (中文)：
可见-近红外光谱（Vis-NIR）技术在柚子这类体积大、果皮厚的水果内部品质检测中的应用仍面临挑战。考虑到柚子果皮在其整体质量中占比较大，且在工业加工中具有较高的附加价值，一种在不影响食用的前提下、于去皮后进行分级的采后处理模式，可能有助于避免果皮对可溶性固形物含量（SSC）评估造成的不利影响。本研究首次系统比较了Vis-NIR光谱技术对带皮与去皮柚子SSC在线检测性能的差异。为此，我们开发了一套适用于柚子的在线光谱采集系统，并分析了完整柚子与去皮柚子的光谱特征差异。随后，通过多种光谱预处理方法和波长选择策略，逐步优化并对比了校准模型的性能。此外，还引入卷积神经网络（CNN）以探索其在特征提取方面的潜力。结果表明：结合标准正态变量变换（SNV）、二阶去趋势（second-order detrending）以及可变窗口大小的移动平滑算法，基于去皮柚子光谱构建的偏最小二乘回归（PLSR）模型取得了最优预测效果。具体而言，该模型的预测决定系数（Rp²）达到0.88，预测均方根误差（RMSEP）为0.294%，残差预测偏差（RPD）为2.57。本研究证实，果皮对SSC预测性能具有显著负面影响，而去皮后检测可显著提升模型精度；同时，CNN可作为传统PLSR方法的有效替代方案。该工作为具有复杂组织结构的农产品内部品质评估开辟了新路径。

英文摘要 (English Abstract):
The quality of edible oils is closely related to their chemical compositions. Antioxidants have widespread application in edible oil production. In this study, a pioneering detection approach involving the use of a one-dimensional convolutional autoencoder (1D-CAE) was introduced to compress spectral data for assessing antioxidant levels in edible oils. Fourier-transform near-infrared (FT-NIR) characterisation of edible oil samples with varying antioxidant concentrations was also conducted. An 1D-CAE model was developed to compress different pre-processed spectra into a condensed representation. These compressed features were then integrated with a support vector machine and partial least squares regression models to establish correlations for each target. The study examined the influence of pre-processing steps and feature engineering methods on near-infrared spectral analysis through independent or combined model analysis. The findings revealed that features derived from the 1D-CAE model demonstrated remarkable repeatability and can be utilised to construct robust detection models. The experimental results showed that the optimal detection model derived based on the 1D-CAE compression features has an average R2, RPD and RMSE of 0.9953, 15.1664 and 1.2035, respectively, on the prediction set. FT-NIR spectroscopy can be used to accurately detect butylated hydroxytoluene in edible oils. Therefore, autoencoders are an effective tool in spectroscopic analysis, offering promising avenues for future research and application.
摘要翻译 (中文)：
食用油的品质与其化学成分密切相关，抗氧化剂在食用油生产中具有广泛应用。本研究提出了一种开创性的检测方法，利用一维卷积自编码器（1D-CAE）对光谱数据进行压缩，以评估食用油中的抗氧化剂含量。研究首先对含有不同抗氧化剂浓度的食用油样品进行了傅里叶变换近红外（FT-NIR）光谱表征。随后构建了一个1D-CAE模型，将经过不同预处理的光谱压缩为紧凑的特征表示。这些压缩后的特征被分别与支持向量机（SVM）和偏最小二乘回归（PLSR）模型结合，建立与各目标抗氧化剂含量之间的关联关系。本研究通过单独或联合建模的方式，系统考察了预处理步骤和特征工程方法对近红外光谱分析性能的影响。结果表明，由1D-CAE模型提取的特征具有优异的重复性，可用于构建稳健的检测模型。实验结果显示，基于1D-CAE压缩特征构建的最优检测模型在预测集上平均决定系数（R²）、相对分析误差（RPD）和均方根误差（RMSE）分别为0.9953、15.1664和1.2035。研究表明，FT-NIR光谱技术能够准确检测食用油中的丁基羟基甲苯（BHT）等抗氧化剂。因此，自编码器在光谱分析中是一种高效工具，为未来相关研究与实际应用提供了富有前景的新路径。

摘要英文：
The most frequent kind of damage to strawberries is bruising. However, most of the bruises are so barely perceptible at an early stage on the surface, that detection of them with the human eye is quite challenging. This study proposes a method for accurately detecting and classifying the damage using reflectance imaging spectroscopy. In order to carry out the study, an experiment was devised to artificially induce bruises and a dataset was generated at different bruise intervals. A model for detecting and classifying bruises at their latent stage was developed using machine learning classifiers, including support vector machines (SVM), k-nearest neighbors (KNN), linear discriminant analysis (LDA), random forest (RF), and decision tree (DT), to investigate the changes over time after bruise occurrence on the detection performance. Regression models for the prediction of bruising time were developed using partial least square regression (PLSR), RF, gradient boosting (GB), support vector regression (SVR), and DT. Among the compared models, both SVM and LDA could achieve 99.99 % classification accuracy. RF was regarded as being the most advisable for detection and prediction jobs due to its high performance. It achieved MSE of 0.052 and R2 of 0.989 for prediction.
摘要中文：
草莓最常见的损伤类型是碰伤（bruising）。然而，在损伤初期，大多数碰伤在果实表面几乎难以察觉，仅凭肉眼很难准确识别。本研究提出了一种基于反射式成像光谱技术的方法，用于精准检测并分类此类损伤。为开展研究，实验设计了人工诱导碰伤的方案，并在不同碰伤时间间隔下构建了一个数据集。研究采用多种机器学习分类器——包括支持向量机（SVM）、k近邻（KNN）、线性判别分析（LDA）、随机森林（RF）和决策树（DT）——构建模型，以检测和分类处于潜伏阶段的碰伤，并探究碰伤发生后随时间推移对检测性能的影响。同时，还建立了回归模型用于预测碰伤发生的时间，所用方法包括偏最小二乘回归（PLSR）、RF、梯度提升（GB）、支持向量回归（SVR）和DT。在所有对比模型中，SVM 和 LDA 的分类准确率均高达 99.99%。综合考虑检测与预测任务的性能，随机森林（RF）被认为是最优选择：其在碰伤时间预测任务中取得了均方误差（MSE）为 0.052、决定系数（R²）为 0.989 的优异表现。

When buying apples, there are several characteristics that are important to the consumer. These include taxonomic variety, which influences the desired flavor and the allergenicity, and, increasingly, whether the cultivation is regional and organic. In this study, we analyzed 217 apple samples with ¹H NMR spectroscopy and verified the authentication of the apples in terms of country of origin, local origin within Germany, taxonomic variety and whether they were organically produced. The obtained random forest models showed classification accuracies ranging from about 73 % for the discrimination of seven taxonomic varieties to 88.5 % for the separation of German and non-German samples. The results demonstrate that ¹H NMR spectroscopy is promising for comprehensive authentication of apples. This is achieved by using different parts of the NMR fingerprint depending on the desired outcome, which was confirmed by the application of variable selection methods and the comparison of the selected variables.
摘要翻译：
在购买苹果时，消费者关注多个重要特征，包括分类学品种（影响期望的风味和致敏性），以及日益受到重视的是否为本地种植和有机栽培。本研究利用¹H核磁共振（NMR）波谱技术对217份苹果样品进行了分析，旨在验证苹果在以下四个方面的真伪：原产国、德国境内的本地来源、分类学品种以及是否为有机生产。所构建的随机森林模型在不同分类任务中表现出良好的性能：在区分七个分类学品种时准确率约为73%，而在区分德国产与非德国产样品时准确率高达88.5%。研究结果表明，¹H NMR波谱技术在苹果的综合溯源与真实性鉴别方面具有广阔前景。该方法通过根据具体鉴别目标选择NMR“指纹”图谱中的不同区域实现高精度分类，这一点也通过变量选择方法的应用及所选变量的对比分析得到了进一步验证。