Currently, the pharmaceutical manufacturing industry faces problems such as low accuracy in detecting metal foreign objects due to product effects. To address this issue, this paper proposes a metal detection algorithm based on phase rotation and time-frequency analysis. Phase rotation suppresses product effect interference, and smoothed pseudo Wigner-Ville distribution (SPWVD) is used to acquire time-frequency images, identifying significant differences representing metal foreign objects and achieving metal detection under strong product effect interference. To ensure computational efficiency meets industrial real-time requirements, an embedded software system with a dual-core CPU and CLA working in tandem is employed, improving algorithm efficiency through hardware improvements. Test results show that the system achieves a detection accuracy exceeding 98% for 0.8 mm ferromagnetic metals and 1.2 mm non-ferromagnetic metals, with a single detection cycle completed within 10 ms.