The traditional method of obtaining the trajectories of labeled particles requires a lot of manual operations, such as manually selecting a particle in each image and obtaining its center, and then obtaining its trajectory by connecting the lines, which not only can only obtain the trajectory of one labeled particle, but also needs a lot of time to get the center of the labeled particles. Aiming at the above problems, a labeled particle trajectory detection method based on Convolutional Long Short-Term Memory is proposed for the first time. First, a large number of consecutive frames are captured by an industrial camera and preprocessed using an algorithm, and a dataset is constructed using the preprocessed images. Then, a Convolutional Long Short-Term Memory network is constructed and trained on the dataset using this network. Finally, the trained model is tested using a test set and evaluated by metrics. The test results show that the PSNR of the model on the test set is 40.44, the SSIM is 0.95, and the LPIPS is 0.14, and all these figures indicate that Convolutional Long Short-Term Memory has achieved success in acquiring labeled particle trajectories.