Screening of pork freshness feature wavelengths and optimisation of nondestructive intelligent detection device

Food safety and quality of pork are closely related to the physical well-being and life quality in modern agriculture. However, traditional detection can often be confined to low efficiency, long cycles, and high operating costs. The purpose of this study was to extract the characteristic wavelengths for each index of pork freshness using the experimental platform of fresh pork supernatant. Six characteristic wavelengths were selected from the highest correlation with pork freshness for further investigation. The freshness indexes were then analyzed, including the TVB-N concentration, pH value, and color (L*, a*, b*) of fresh pork as a function of storage duration, as well as the spectrum information of the supernatant. The characteristic wavelengths were also identified from the spectrum of fresh pork supernatant using CARS and SPA models. The results show that the performance of the PLS model was significantly improved using unique wavelengths. Furthermore, after SNV-CARS-SPA processing, the R_p of the TVB-N content, pH value, L*, a*, and b* were 0.996, 0.961, 0.731, 0.922, and 0.839, respectively; the RMSEP values were 0.388 mg/100g, 0.067, 2.377, 0.832, and 0.567, respectively. Among them, the six feature wavelengths of 540, 580, 680, 730, 760, and 860 nm were achieved in the best connection with the freshness of pork. According to the characteristic wavelength, the software, and the hardware in the laboratory, the miniature spectrometer of the handheld device was replaced with a multi-spectral sensor. Additionally, the light source and control modules were redesigned for the structure of the device, including the cooling, power supply, acquisition probe, and shell. The software of the gadget was improved using the Arduino system. The stability of the gadget was superior to that of the enhanced device. The prediction models were then established, according to the freshness indices of fresh pork, including the TVB-N content, pH value, L*, a*, and b*. The optimal models were obtained after SNV preprocessing. Nevertheless, the prediction models shared the higher TVB-N content, pH value, a* and b* indexes. The optimal model was obtained under six distinctive wavelengths, the R_p of the TVB-N content, pH value, L*, a*, and b* were 0.970, 0.937, 0.805, 0.908, and 0.915, respectively, and the RMSEP values were 0.874 mg/100g, 0.097, 1.972, 1.514, and 0.729, respectively. Therefore, the portable detection device of fresh pork freshness performed better on the rest indexes before enhancement. Except for the index brightness L*, there was some impact to differentiate between freshness and other factors. As such, the upgraded device can be expected to determine whether the fresh pork was fresh or not. There was the freshness index TVB-N with a higher degree of precision than before.