Abstract: Abstract: Adulteration of beef with spoiled beef seriously endangers the health and economic interest of consumers. Therefore, it is of great significance to implement nondestructive detection of beef adulteration. Traditional methods for detecting meat adulteration are destructive and laborious. Biological speckle (bio-speckle) technique is a non-invasive and rapid detection method. Therefore, in this paper, bio-speckle technique was used for quantitative detection of beef adulteration by detecting bioactivity variance among samples. In the experiment, a total of 62 adulterated beef samples with different adulteration concentrations of 0, 1%, 3%, 5%-60% (5% increment) and 100% (w/w) were prepared by mixing fresh and spoiled beef at different ratios, and the samples were divided into calibration set and test set by sample set partitioning based on joint x-y distance (SPXY). The samples were illuminated by a 10 mw laser at 632 nm with 60° incident angle. Five hundred biological speckle images (640×480 pixels) of each sample were collected by a CCD (charge-coupled device) camera at 20 fps. To solve the problem of poor stability of traditional single row/column inertia moment (IM) method in characterizing bio-speckle activities of samples, a new method named inertia moment spectrum (IM spectrum) was developed. Specifically, the temporal history speckle patterns (THSPs) for individual columns of the bio-speckle images were generated, based on which IM was calculated for each column. By splicing IM of each column, a spatially continuous spectrum, i.e., IM spectrum is established. IM spectrum represents the spatial biological activity information of the sample and thus has strong anti-interference ability. By comparing the IM spectrum of different adulteration samples, it was found that IM spectrum offered a good alternative for visualizing sample variance due to adulteration levels. The width of a broad peak corresponding to the illumination center of laser decreased with the increase of adulteration levels, which may be contributed to physical and chemical component difference that resulted in scatter coefficient variations of different samples. Moreover, the IM spectrum was normalized and utilized in the development of support vector regression machine (SVR) model for beef adulteration detection. Model parameters including penalty parameter and kernel function parameter were optimized by particle swarm algorithm. The results showed that the SVR model based on IM spectrum was feasible to predict the levels of adulteration in beef, in which the coefficients of determination and the root mean squared errors were 0.85 and 0.12 as well as 0.81 and 0.11 for calibration set and test set, respectively. The penalty parameter and kernel function parameter of the model were 1.96 and 0.01, respectively. The coefficients of determination of the model were greater than 0.8, and the root mean squared errors for calibration and test were close, indicating that the model has high stability and good precision. Besides, from the scatter plot of the predicted and true values of the model, it was found that the main errors of the model originated from 100% adulteration samples and one unadulterated sample. However, since all spoiled beef samples were predicted as high adulteration levels, it was therefore considered that such SVR model was still of practical use. This study demonstrates that it is feasible to use bio-speckle imaging and IM spectrum analysis for detecting beef adulteration with spoiled beef. Nevertheless, more studies are needed to further improve model performance and explore the usefulness of IM spectrum in bio-speckle analysis.