Abstract: A huge amount of data will be generated per unit of time at high sampling rates. Some data will be discarded when the data cannot be processed in real-time, thus missing the episodic transient signals. Practical test scenarios are often faced with sparse signals in the frequency domain, and the concern is often narrow frequency bands. Therefore, a transient signal detection method for the target band is proposed based on a signal dictionary of the Slepian series, which is efficient by focusing only on target band energy information. First, a set of orthonormal Slepian sequences are selected to form a signal dictionary, which can characterize the signal features in the target band; then, the detection is achieved by judging the matching degree between the observed signal samples and the dictionary. The comparison experiment shows that the computational complexity of the proposed detection method is reduced by more than 92% compared with the short-time Fourier transform, more than 71% compared with the discrete wavelet transform, and more than 35% compared with the windowed Wigner-Ville distribution. Simulation experiments are performed using pulse-modulated signals with a sparse time domain for verification, and the results show the effectiveness of the proposed detection method. Based on the detection results, the sampling data that are not of interest can be discarded, which saves storage resources and reduces the amount of data for signal processing.