To solve the problems of poor quality and yield of camellia oil owing to the lack of basis for determining the maturity of camellia seeds during the process of harvesting, a method for detecting the maturity of camellia seeds based on mid- and far-infrared spectral data fusion was proposed herein. A Fourier transform infrared spectrometer was used to test the mid- and far-infrared spectroscopy data of camellia seeds with different oil contents at different maturity stages. Various feature extraction methods (principal component analysis, successive projection algorithm, and noninformation variable elimination method) were used to extract the original spectral data, and the methods were combined with a support vector machine algorithm (SVM) to develop a model for identifying the maturity of camellia seeds. The results show that the best discrimination accuracy in the mid-infrared band is 93.33% when using the successive projection algorithm combined with the genetic algorithm to optimize the SVM model. In the far-infrared band, nine variables extracted using the principal component analysis are used as input variables, and when combined with the SVM model optimized by the genetic algorithm, the identification accuracy of 96.67% is attained. The identification model of camellia seed maturity is established using the SVM algorithm after parameter optimization. The experimental results show that the accuracy of intermediate data fusion combined with the optimized SVM algorithm can reach 100%. The results of this study show that when combined with an improved SVM model, the infrared spectroscopy can accurately determine the oil content of camellia seeds. Data fusion can effectively increase the spectral information and remove redundant information from a single spectrum. The results can provide a reference to determine the best picking time of camellia and can be extended to determine the maturity of other agricultural and forestry products.