This paper presents an automated high-speed die edge detection framework, based on improved K-Mean and landscape analysis, that is highly suitable to be implemented in precision engineering for optical non-destructive testing of die-level defects inspection in the semiconductor industry. This paper specifically aims to achieve high accuracy (or yield) of greater than 99.95% with stable performance within a short computation time (i.e. less than 20 ms). To demonstrate the applicability of the proposed framework, it is validated using three production units (i.e. Production unit A: 6000 units; Production unit B: 3500 units; Production unit C: 4000 units) and is benchmarked to two baseline edge detection methods, namely cross-correlation and normalised cross-correlation methods, as well as state-of-the-art vision libraries, recent works, and several conventional edge detection methods. The results obtained show that the proposed framework is capable of performing die edge detection with promising accuracy and stable performance by achieving 100.0% yield in all three production units, having outperformed the benchmarking methods. Also, the overall computation time (considering die edge detection and die rotation) of the proposed framework is short, at approximately 15 ms.