研究目的
To develop an untargeted detection method based on local anomaly detection (LAD) using near infrared (NIR) imaging for detecting contaminants in soybean meal, addressing the limitations of traditional targeted detection methods.
研究成果
The untargeted LAD method demonstrated acceptable performance for detecting contaminants in soybean meal without the need for a 'clean' library, with a high coefficient of determination for quantitative analysis. The method shows potential for moving food and feed safety control from passive to active by enabling early detection of suspicious contaminants.
研究不足
The study acknowledges the challenges of multi-pixel targets in the local window and the absence of edge detection, which were addressed through randomization and edge expansion techniques. The method's performance is dependent on the size of the window filter and the level of confidence set for anomaly detection.
1：Experimental Design and Method Selection
The study employed an untargeted detection method based on local anomaly detection (LAD) using near infrared (NIR) imaging. The methodology included the use of a sliding local window filter to analyze every pixel in the NIR microscopic image, calculating the modified Mahalanobis distance (Global H) for outlier detection.
2：Sample Selection and Data Sources
Three different soybean meal samples (hulls, full-fat and de-hulled soybean meal) and one pure soybean meal sample were used. Adulterated samples were prepared with non-protein nitrogen compounds at various concentrations.
3：List of Experimental Equipment and Materials
Fourier transform infrared (FT-IR) imaging system (Spotlight 400 FTIR Imaging system, PerkinElmer Ltd.), line-scan NIR hyperspectral imaging system (BurgerMetrics SIA), Retsch mill (Ultra centrifugal Mill ZM100; Retsch GmbH), mixer (REAX 20/8; Heidolph).
4：Experimental Procedures and Operational Workflow
Samples were prepared by grinding and mixing, then scanned using NIR imaging systems. Data were processed using Matlab with the PLS_Toolbox for chemometric analysis.
5：Data Analysis Methods
The LAD method was applied to detect anomalous spectra, followed by k-means clustering for classification. Quantitative analysis was performed using partial least squares discriminant analysis (PLS-DA).
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
