Abstract: China is the world's largest producer and consumer of fresh pork, but in the production and sales of pork, fresh pork is very prone to spoilage. In order to realize the real-time detection of fresh pork freshness during storage and transportation, this study developed a handheld fresh pork freshness non-destructive intelligent detection and grading device based on visible/near-infrared spectroscopy. The detection device took the visible/near-infrared spectrum acquisition unit as the core, built a hardware system, and developed a software system for simultaneous detection of fresh pork freshness and freshness grading. The detection device mainly consists of a power supply, central controller, display unit, heat dissipation unit, detection light source and spectral sensor. The spectral sensor and detection light source are integrated with a visible/near-infrared spectral acquisition unit by the detection probe. The spectral acquisition unit transmits the diffuse reflectance spectral signal of pork to the central controller for processing and analysis. Through the display unit and control buttons, human-computer interaction operation, real-time control and result display can be achieved. According to the selection of each component and the actual application scenario, the whole machine structure was designed, the size of the whole machine is only 165 mm×75 mm×115 mm, the weight is 745 g, and it can be held by one person and completed the detection with one button. Through the developed device, the diffuse reflection spectrum of pork in the wavelength range of 650-1 100 nm was collected. After the spectra were preprocessed by standard normal variable transformation (SNV), the continuous projection algorithm (SPA) and the competitive adaptive weighted sampling algorithm (CARS) algorithm were compared, and the general prediction model of pork freshness index in different parts was established. According to the predicted value of volatile salt-based nitrogen (TVB-N) content and pH, pork was divided into three grades: fresh, sub-fresh and spoilage. The experimental results show that the prediction model established by SNV-CARS-PLS has better performance. The prediction set correlation coefficients of TVB-N content, pH, L*, a* and b* were 0.942, 0.945, 0.940, 0.933 and 0.833, respectively. And the prediction root mean square errors were 1.131 mg/100 g, 0.136, 1.706, 1.217 and 0.717, respectively. The results of the general detection model were introduced into the device and verified by experiments, and the test results showed that the prediction results of root mean square errors of TVB-N content, pH, L*, a* and b* were 1.109 mg/100 g, 0.134, 1.140, 1.094 and 0.636, respectively. The correct grading rate of freshness was 92.86%. The detection time for a single sample was about 1 s. Research has shown that the detection device can meet the needs of multi-index on-site rapid detection and grading of pork freshness of different parts, with advantages such as low cost, portability, simple operation, and efficiency. It is suitable for various links in the supply chain of pork and providing guidance for production and consumption decision-making. The detection also plays an important role in timely grasping the freshness of fresh pork in the process of storage and transportation, assisting in decision-making storage, transportation and sales plans, and ensuring the quality and safety of fresh pork.