Currently, shrimp fry vitality detection mainly relies on manual observation of shrimp fry swimming, but the evaluation results are susceptible to subjective influence. To address this problem, this study proposes a shrimp fry vitality detection method that combines the dense optical flow method and statistics. Firstly, shrimp fry with different vitality were obtained by rearing them in different water temperatures, and shrimp fry videos were collected; then the inter-frame motion features of shrimp fry in the videos were extracted using the Gunner Farneback optical flow method, which represents the instantaneous swimming state of the shrimp fry, and contains two dimensions, namely, magnitude and angle, and statistical analyses were performed at the same time; finally, we extracted the motion features between all the frames of the fixed-length shrimp video and applied the improved information entropy and mutual information to detect the vitality of the shrimp fry. After testing, the precision and recall of the two detection methods exceeded 98%, which can accurately and objectively realize the shrimp fry vitality detection. This study provides ideas and methods for automated shrimp fry quality detection, which has application value in aquaculture industry.