Determining cold resistance physiological indicators is an important way to evaluate the cold resistance of tea plants. Traditionally, methods of evaluating the cold tolerance of tea trees are mainly through the determination of physicochemical parameters of tea trees under low-temperature stress. However, these methods are not only time-consuming and labor-intensive but also destructive. This study established a prediction model for REC, SPAD, and MDA of tea tree cold resistance based on multispectral and hyperspectral imaging techniques. Firstly, multispectral and hyperspectral images of 32 breeding materials under low-temperature stress were collected, and the REC, SPAD, MDA, SP, and SS contents of the corresponding tea tree leaves were determined. Secondly, the hyperspectral image data among them were spectrally pre-processed using five methods, namely, MSC, SNV, S-G, 1-D, and 2-D, and the characteristic bands were screened using two methods, UVE and SPA. Finally, the REC, SPAD, and MDA prediction models of tea tree cold resistance were established using SVM, RF, and PLS algorithms for multispectral and hyperspectral data. The results showed that (1) the spectral curves were more stable, the peaks and valleys were more prominent, and the accuracy and reliability of the models were higher after the joint preprocessing of MSC, SNV, S-G, 1-D and 2-D; (2) the UVE algorithm screened the largest number of characteristic bands, while the SPA algorithm screened the smallest number of characteristic bands, which was more suitable for establishing regression models with hyperspectral data; (3) The RF model has the highest accuracy in predicting leaf REC (R_p=0.735 2，RMSEP=0.077 1), SPAD (R_p=0.502 9，RMSEP=6.681 8), and MDA (R_p=0.784 6，RMSEP=8.885 3) content under multispectral imaging techniques; the SPA-SVM model has the highest accuracy in predicting leaf SPAD (R_p=0.734 9，RMSEP=4.154 6) and MDA (R_p=0.685 8，RMSEP=8.548 8) under hyperspectral imaging techniques, and the SPA-PLS model has the highest accuracy in predicting REC (R_p=0.629 8，RMSEP=0.066 9). Therefore, the REC, SPAD, and MDA prediction models based on multispectral and hyperspectral imaging and machine learning algorithms provide an accurate, non-destructive, and efficient method, which is of great significance for evaluating tea tree cold resistance.