Abstract: The gravimetric (weight-based) method is widely used for detecting suspended particulate matter (SPM) in aquaculture water. However, it is labor-intensive and time-consuming. To enable rapid and efficient detection, this study focused on the SPM in the aquaculture environment of Scophthalmus maximus. By capturing video footage of suspended particulates in a tank, we developed an automatic detection method based on the Gaussian Mixture Model (GMM) for identifying SPM in water. The results demonstrated that dynamic grayscale processing combined with GMM-based background modeling enabled the extraction of recognizable images of SPM. An intelligent image screening and particle-counting approach was then established. The recognition algorithm was implemented and automated using Python, incorporating relevant image processing libraries. The GMM-based method achieved a detection limit as low as 0.6 mg/L in an industrial recirculating aquaculture system (RAS) for Scophthalmus maximus. Moreover, particle counts obtained through intelligent recognition showed strong correlation with gravimetric measurements (R² = 0.981). To further validate the method, 24-hour continuous monitoring of SPM was conducted, and the relative error between the intelligent detection and the traditional weight method remained below 5%. These results indicate that the GMM-based intelligent recognition approach can reliably and automatically quantify SPM concentration. This method offers advantages such as real-time monitoring, continuity, intuitive visualization, and operational simplicity, showing strong potential for practical application in aquaculture water monitoring.

Key words: recirculating aquaculture system, suspended particulate matter, intelligent recognition, Scophthalmus maximus

摘要： 重量法是工厂化水产养殖水中悬浮颗粒物浓度的常用检测方法，但该方法的检测过程繁琐耗时。本研究选择大菱鲆(Scophthalmus maximus)工厂化养殖水中的悬浮颗粒物为试验对象，通过采集水中悬浮颗粒物视频图像，构建了基于高斯混合模型(Gaussian Mixture Model, GMM)的悬浮颗粒物智能识别方法，以实现水中悬浮颗粒物浓度的自动快速检测。结果显示，通过对水中悬浮颗粒物视频的动态图像灰度化处理和GMM识别背景建模，提取出悬浮颗粒物的可识别图像，建立了识别图像的智能筛选和计数，并使用Python的相关库对智能识别代码程序进行封装，创建出一种水中悬浮颗粒物的智能识别方法。该方法对大菱鲆工厂化循环水中悬浮颗粒物质量浓度检测下限低至0.6 mg/L，且智能识别方法与重量法测定结果表现出良好相关性(R2 =0.981)。通过对大菱鲆工厂化养殖水中悬浮颗粒物24 h连续监测对比，可知智能识别测定值与重量法测定值的相对误差小于5%，表明本方法可实际应用于工厂化水产养殖水中悬浮颗粒物质量浓度的测定。

关键词: 循环水养殖系统, 悬浮颗粒物, 智能识别, 大菱鲆