A method for precise localization of seismic sources in rock mechanics was proposed. The method combines accurate picking of P-wave arrival times with an analysis-based filtering of anomalous arrivals and a joint localization approach using genetic algorithms (GA). The method was applied to locate fracture surfaces in rock materials and in mining engineering. To mitigate the impact of sensor timing accuracy, traditional iterative methods, and the selection of initial iterative values on localization, an improved STA/LTA method based on time-frequency analysis was used to pick the P-wave arrival times. A logic probability density function of the analysis-based solutions for different sensor combinations was employed to filter out anomalous arrivals. Genetic algorithms were then used to locate the clear arrivals from filtered data. The method was validated through tests on pencil lead break and coal-interlayer-rock compression, as well as controlled blasting experiments. The results show that combining precise filtering using variational mode decomposition (VMD) after pre-bandpass filtering with the improved STA/LTA method significantly enhances the accuracy of P-wave arrival time picking. The combination of the analysis method for filtering the impact of anomalous arrivals on positioning accuracy and the genetic algorithm, which is globally applicable and not influenced by initial values, allows for more accurate localization of seismic sources in rock material testing and mining engineering. This method has greater practical engineering application value compared to traditional iterative methods and linear positioning methods.