无目标食品掺假的近红外光谱单类分类检测研究

近红外光谱技术可以满足食品掺假的大规模快速无损检测需求。现有的方法多是基于掺假信息已知的情形，通过传统的监督分类方法构造判别模型。随着食品掺假手段的多样化和水平的提高，这种基于“有目标掺假”的检测方法难以处理掺假信息不确定的“无目标掺假”检测问题。本项目拟利用单类分类器实现无目标食品掺假的近红外光谱检测。单类分类器只需要目标类别的训练样本，十分符合无目标食品掺假的检测需求，但是现有的单类分类器具有不适于高维数据和参数难以优化的缺陷。本项目旨在解决上述缺陷并开展无目标食品掺假的近红外光谱检测，具体研究内容包括：1. 基于“知己知彼”的学习法则，从目标类的自我特征出发，对目标类进行特征降维，解决近红外光谱的高维问题；2. 利用生成对抗网络，生成位于目标类低密度区域的样本，解决单类分类器参数难以优化的问题。本项目研究成果有助于实现无目标食品掺假的快速、自动鉴别，对加强食品监控具有重要意义。

Near-infrared spectroscopy can meet the needs of large-scale and rapid non-destructive detection of food adulteration. Most of the existing methods are based on the situation that the adulteration information is known, and the discriminant model is constructed by the traditional supervised classification method. With the diversification and improvement of food adulteration methods, the detection methods based on targeted adulteration are difficult to deal with the problem of untargeted adulteration with uncertain information. This project intends to use the one-class classifier to achieve the near-infrared spectroscopy detection of non-target food adulteration (NFA). The one-class classifier only needs the training samples of the target class, which meets the requirements of NFA detection, but the existing one-class classifier is not suitable for high-dimensional data and the parameters are difficult to optimize. This project aims to solve the above defects and carry out near-infrared spectroscopy detection of NFA. The specific research contents include two aspects. On the one hand, based on the learning rule of knowing oneself and knowing the enemy and the self-characteristics of the target class, the feature dimension of the target class is reduced to solve the problem of the high dimension of near-infrared spectrum. On the other hand, using the generative adversarial network to generate samples in the low-density area of the target class to solve the problem of parameter optimization of the one-class classifier. The research results of the project contribute to the rapid and automatic identification of NFA, which is of great significance to the strengthening of food monitoring.