In order to address the challenge of contact-based detection of multiphase liquid interfaces in chemical production processes, an intelligent interface detection system based on machine vision was investigated. The system enables external non-contact monitoring of multiphase liquid interfaces and investigates improvements to interface detection algorithms along with the construction of a dedicated dataset. The results demonstrate that the system boasts a detection error rate of only 5.3%, with model accuracy reaching 86.3%, satisfying the precision and real-time detection requirements for industrial applications. Additionally, the model's weight file is a mere 13.6 MB in size, allowing for deployment on edge computing devices. It is concluded that the system exhibits real-time, precise, and efficient characteristics, providing an innovative solution to the problem of liquid separation in multiphase systems within the chemical production industry.