Abstract:  Long-time coherent integration is an effective way to improve high speed weak target detection performacne. In order to resolve the problems of large calculation load caused by across range units (ARU) and across Doppler unit (ADU), a parallel processing technique is proposed in this paper . At first, the uniformly accelerated rectilinear motion model is established based on the motion feature of high speed target in a coherent processing interval. Then, the echoes are reorganized into multiple groups in parallel by the number of range units that the interested targets may walk across. Subsequently, Fast Fourier transform (FFT) is executed for each groups along the slow time dimension and the first threshold is set to reduce the parameter searching range. Moreover, the particle swarm optimization (PSO) is utilized to search the optimum parameters of the targets along (range, velocity, acceration) dimensions and the second threshold is set to declare the existence of the target. By two-threshold and PSO methods, the proposed technique simplifies the search complexity, reduce the calculation load effectively, showing good hardware feasibility. Finally, the upper bound of integration loss is analyzed and the simulation results show its cauculation effectiveness.