In order to address the problem of detecting and tracking low signal-to-noise-ratio (SNR) targets in bistatic passive coherent location (PCL) system, an improved dynamic programming track-before-detect algorithm is proposed. First, the polar measurements are converted to Cartesian measurements. Second, the Cartesian measurements are clustered with the neural network, and the amplitudes of the valid measurements are scaled to improve the performance. Finally, the DP-TBD algorithm is used for amplitude accumulation and state backtracking. Simulation results verify the effectiveness of the proposed method.