Aiming at solving the problem of the times of measurements, a scheme of sequential compressed sensing based on wavelet packet was proposed. The wavelet packet transform was firstly employed and the image was decomposed into wavelet packet blocks with the same size. The length of the initial sampling signal was then estimated using the relationship between the mathematical expectation and the sparsity of wavelet packet blocks. Meanwhile, the lengths of sampling signals were set dynamically in order to reduce the reconstruction and communication overhead. The problem of uncertain times of measurements in the traditional compressed sensing was thus overcome. Experimental results show that our proposed scheme has less reconstruction times and lower reconstruction error than the existed ones.