Traditional palm vein recognition is susceptible to temperature variations, resulting in low success rates. To address this issue, we propose a palm vein recognition method based on Long Short-Term Memory (LSTM). Initially, palm vein image data and temperature information are collected over a period. A Fully Convolutional Network (FCN) model is trained using manually annotated palm vein images. Features vectors of the annotated palm vein images are then computed through a Convolutional Neural Network (CNN), and an LSTM-CNN model for palm vein recognition is trained using LSTM. Finally, the to-be-recognized palm vein images and temperature information are obtained through sensors. These are matched with the corresponding temperature-based LSTM-CNN model and feature vector templates. If the matching score exceeds a certain threshold, the recognition is successful; otherwise, it fails. Experimental results show that under a temperature range of -20 to 40 degrees Celsius, and with approximately 12,000 palm vein image data and temperature information collected from 10 individuals for training, complete and accurate recognition of the palm vein information of the 10 individuals can be achieved. This method overcomes the impact of temperature variations on the accuracy of palm vein recognition, meeting the application needs of real-world scenarios.