<p>The seismic first-arrival picking is the primary premise of tasks such as earthquake event identification, source location, and parameter inversion. Its real-time performance and accuracy directly affect the efficiency of seismic data processing. However, due to the increasing environmental noise, traditional seismic first-arrival picking algorithms are difficult to balance real-time performance and accuracy. To address this issue, this paper analyzes and improves the STA/LTA (Short-Term Average/Long-Term Average) algorithm, which is known for its high processing efficiency and strong real-time performance, and employs it as the initial detection method for first-arrival picking. Meanwhile, combined with the Delaunay triangulation algorithm, the spatial relationship of the first-arrival picking results is analyzed, and an automatic seismic first-arrival picking algorithm based on the improved STA/LTA and Delaunay triangulation is proposed. Firstly, by constructing the anti-interference (AR) characteristic function, the interference from baseline drift and low-frequency noise is effectively suppressed, significantly enhancing the anti-interference ability of the STA/LTA algorithm in seismic first-arrival picking. Secondly, the “delay long window” method is proposed to improve the computational efficiency and picking accuracy of the STA/LTA algorithm. Finally, by constructing the Delaunay triangular spatial relationship between stations and using the travel-time law of seismic waves, a Delaunay triangulation trigger criterion that integrates a travel-time threshold is proposed to test the spatiotemporal consistency of the first-arrival picking results of the STA/LTA algorithm. Validation with actual data shows that the proposed joint algorithm can effectively eliminate erroneous picks from single stations and significantly improve the accuracy of seismic first-arrival picking.</p>