Abstract: Abstract: It's of great significance to rapidly, non-destructively and simultaneously detect quality parameters of pork. In order to solve this problem, a detection device for quality parameters of meat based on dual-band visible/near infrared (Vis/NIR) spectroscopy technology was proposed in this paper. Combined with hardware unit and software control program, the device could collect, real-time process, calculate, display and preserve the dual-band spectral information at the same time. Among them, the hardware consisted of three main units, including the spectrum acquisition unit, the light source unit and the control unit. The spectrum acquisition unit included a pick-up optical fiber and two spectrometers whose wavelength ranges were 350~1 100 nm and 1 000~2 500 nm, respectively. For the light source unit, we used tungsten halogen lamp that came to the surface of sample through ring light guide. The light source and pick-up optical fiber were designed as a whole to form a handheld probe. The control unit worked with the help of a microcontroller. The software was self-developed based on Visual Studio 2010 platform using C language. One could aim the handheld probe at samples when detecting and then pressed the button to complete the automatic acquisition. Based on the detection device, Vis/NIR spectral information which covered multiple quality parameters from 350~2 500 nm of pork was collected and pretreated by smoothing method and standard normal variate (SNV) to eliminate the noise in the spectrum and correct the spectral errors caused by the scattering of the samples. Then coefficients of variation at each wavelength for five acquisition results were analyzed to determine whether the system was stable and the data were reliable. At last, partial least square (PLS) prediction models based on single band and dual-band spectrum data between reflection information and pork color (L*, a*, b*), pH value, total volatile base nitrogen (TVB-N), water content, cooking loss and tenderness determined by standard methods were established, respectively, in order to build the most robust models. The best PLS models for L*, a* and b* were established using single band (350~1 100 nm). For other parameters, the best PLS models were based on dual-band spectroscopy with higher detection accuracy and lower prediction error. As real time detection device calls for fast detection speed and high accuracy, competitive adaptive reweighted sampling (CARS) was employed to select characteristic variables on this basis and new PLS prediction models were established between the chosen important variables and quality attributes values determined by standard methods. The results indicated that CARS method could select the characteristic variables effectively, simplify the established PLS prediction models, reduce the numbers of variables, and improve the running speed and model performance. The correlation coefficients in the prediction set were 0.962 5, 0.933 6, 0.938 9, 0.941 5 and 0.936 3, 0.912 3, 0.920 0 and 0.901 9, prediction errors were 0.628 7, 0.757 6, 0.547 1, 0.0782 , 2.835 4 mg/(100 g), 0.380 9%, 2.560 0% and 6.896 7N for L*, a*, b*, pH value, TVB-N, water content, cooking loss and tenderness, respectively. The results showed that the device had advantages of small size, fast detection speed and high detection accuracy which could realize the simultaneous detection of meat quality parameters and the use of dual band spectroscopy could provide the spectral information of multi-quality parameters of pork samples. In conclusion, the self-developed device could satisfy the demands for nondestructive detection device for multiple parameters in the market and had great potential in application.