The work aims to monitor the agricultural products always under such conditions as low temperature and appropriate humidity in real time to achieve the timely feedback and early warning of the information on the agricultural product quality. With respect to the problem of quality and safety of agricultural products during the cold-chain logistics, first, the factors that affected the quality and safety were analyzed, and the retroactive and monitoring information on the supply chain was integrated to construct the early-warning indicator system for the quality and safety of the agricultural products. Then, an adaptive hybrid particle swarm optimization (AHPSO) algorithm combined with the crossover and mutation operator was designed to optimize support vector machine (SVM) parameters so that the AHPSO-SVM early-warning model of agricultural products' quality and safety during cold-chain logistics was built. With the early-warning indicator system of apples as an example, after the model training and prediction, the predicted output curve and the expected output curve could be better fitted, and the error value was small. Compared with the traditional BP neural network and SVM, in solving practical problems by the proposed method, the prediction accuracy is higher and the quality and safety early-warning accuracy during the cold-chain logistics of agricultural products can be effectively improved.