Abstract:Dehydrated milk powder serves as the fundamental raw material for formula milk, and its place of origin affects the quality of terminal dairy products. In this paper, a method for discriminating the source of dehydrated milk powder by employing electronic nose technology and utilizing a one-dimensional Laplacian convolution kernel is proposed. Sample data is collected using an electronic nose, and after temporal signal alignment, various orders of one-dimensional Laplacian convolution kernels are applied to extract features. Additionally, other feature extraction methods such as statistical numerical features, fast Fourier transform, and discrete cosine transform were compared. Subsequently, partial least squares and visualization were used for separability analysis. Experimental results revealed that the fast Fourier transform, discrete cosine transform, and second-order one-dimensional Laplacian convolution kernels effectively improved separability relative to the original features. The R2 effect size of partial least squares was increased from 0.61 to 0.95, 0.96, and 1.00. The one-dimensional Laplacian convolution kernel feature extraction method was able to accurately distinguish between domestically produced and foreign (Australia) base powder. In a case study, it achieved the best discrimination effect, indicating that it can effectively extract time response features of electronic nose channel sequence signals. This method can facilitate the differentiation work of dehydrated milk powder samples in China and Australia, providing technical support for the rapid identification of dehydrated milk powder sources.