It is an important measure to ensure food quality and safety that real-time monitoring of the key quality indicators of fresh meat after packaging in the process of storage and transportation. The feasibility of combining hyperspectral imaging (HSI) technology with chemometrics and deep learning to detect the quality deterioration of polyethylene (PE)-packaged raw beef, especially for a key lipid oxidation indicator of malondialdehyde (MDA) content, was explored in this study. The feasibility of filtering to overcome the interference of packaging film on the spectral data was further investigated. Near-infrared HSI (400-1000 nm) was used to collect spectral and spatial data of beef samples during short-term storage. A least squares regression (PLSR) and echo-neural network optimized by vulture optimization algorithms (BES-ESN) models were developed by multivariate data processing methods. The results showed that the performance of models established by PE-packed beef samples was usually inferior to that established by unpacked beef samples. The changes of MDA content in beef were visualized according to the optimal model. In addition, Gaussian filtering was applied to reduce the interference effect caused by the packaging film. The results have demonstrated that HSI combined with Gaussian filter preprocessing and multivariate data processing provided an efficient and reliable tool for detecting the freshness of beef in PE packaging. The best model had a coefficient of determination (