Abstract: In the risk monitoring and early warning system of tailings ponds in non-coal mines，to ad‐ dress the key issues，such as the low accuracy of surface displacement monitoring methods，the easy drift of data reference points，the high false alarm rate of dry beach length monitoring，and the lack of intelli‐ gent analysis of tailings dam stability，a tailings pond risk monitoring and early warning model based on machine vision measurement algorithms is constructed through a research method combining indoor and field tests. It is proposed to apply machine vision technology to the risk monitoring and early warning sys‐ tem of tailings ponds，adopting intelligent analysis methods for surface displacement，dry beach monitor‐ ing and stability，as well as countermeasures for machine vision monitoring under complex climatic condi‐ tions and data transmission in harsh environments，to ensure the measurement accuracy and real-time per‐ formance of machine vision. The research selected the representative Yangshan Tailings Pond of Jiangxi Jinshan Mining Co.，Ltd. as the on-site pilot project. The results were compared with the detection results of the monitoring equipment already installed in the tailings pond. The measurement accuracy of both hori‐ zontal and settlement dislocations was improved. Moreover，edge computing was adopted，and abnormal problems were promptly reported. The intelligent recognition rate of the dry beach length of the tailings pond by the AI analysis measurement method was high. The false alarm rate was 0%，and it had strong adaptability in adverse weather and environmental conditions，thus having great value for promotion and application.

Key words: