<p>A spiral concentrator is an efficient gravity separation device that utilizes a flowing film. It is recognized for its environmentally friendly operation, absence of power requirements, ease of large-scale deployment, and low capital and operational costs. It has been widely applied in the beneficiation of iron, tin, titanium, tantalum‑niobium, and other metals, as well as non-metallic ores such as sulfur and coal. Currently, the control of concentrate grades in spiral concentrators relies heavily on manual intervention. Operators must continuously observe the mineral zoning and physically adjust the splitter position to align with the boundary between concentrate and tailings. This manual process is subjective, prone to delays, and can lead to fluctuations in the beneficiation index. To address this limitation, we propose an automated machine vision system that eliminates the need for human intervention by intelligently monitoring and adjusting the concentrate cut-point. The system employs an enhanced YOLOv5 algorithm to determine the separation point of the concentrate and actuates an automated collection mechanism. The novelty of this work lies in the integration of attention mechanisms (CAM and SAM), a small‑target detection layer, and an improved loss function (CIoU) into YOLOv5, specifically tailored for the challenging conditions of spiral concentrators where ore bands are often indistinct and dynamically changing. Experimental results show that the optimized YOLOv5 algorithm achieves a detection accuracy of approximately 90% and a speed of 63 frames per second. The adaptive interception system responds reliably with control accuracy exceeding 90%, meeting industrial requirements for both precision and speed. This system significantly enhances the automation and intelligence of spiral concentrator operations, contributing to more consistent product quality and reduced labor costs.</p>