Aiming at the problems of time-consuming and labor-consuming analysis of water pollution by traditional physical and chemical methods, a water quality monitoring method based on fish abnormal behavior recognition was proposed. In this paper, red zebrafish was used as the research object. Through computer vision technology, the zebrafish images were pre-processed first and GLCM was used to obtain the texture features of the fish school. Then Lucas-Kanade optical flow method was used to calculate the motion information entropy of fish, and the obtained texture features and information entropy were normalized. Finally, the LightGBM was used to detect the abnormal behaviors of fish for comparison with the detection results of DNN and SVM. The results showed that the accuracy rate of the fish abnormal behavior detection with LightGBM was 98.5%, which was improved by 0.5% and 25.3% respectively compared with other models. Researches show that the LightGBM model-based fish abnormal behavior detection method can more accurately identify abnormal fish swimming than other models, and is suitable for automatic water quality monitoring.