Abstract: Increasing the resolution of the radar reduces the amount of energy per cell backscattered by the distributed clutter. When the clutter-to-noise ratio (CNR) is lower than 10dB, the effect on the detection performance which is put by the thermal noise is not neglectful. On the condition of the low CNR, distributed targets detection embedded in compound-Gaussian clutter plus thermal noise is studied. Firstly, we assume that the thermal noise is statistically independent of compound-Gaussian clutter which is modeled as a spherically invariant random vector (SIRV). Given a specific value of τ which is usually named texture, the total interference which is composed by the superposition of compound-Gaussian clutter and white Gaussian thermal noise is approximatively equivalent to a new compound-Gaussian clutter, whose parameters are suitably adjusted for the actually condition. And then, based on the Rao test, the new N-Rao detection is derived to implement the distributed target. The distributed target, which is modeled as a subspace random signal, may be distributed both in range and also in Doppler frequency axes. By calculating the probability of false alarm, it is shown that the probability of false alarm is only a function of the number of pulses N, the number of target range resolution cells H, the number of scatterers in each range cell Nt. So the N-RAO detection is a constant false alarm rate (CFAR) detection. In the end, performances of the proposed detector are assessed through Monte Carlo simulations. The experimental results show that a decrease of the shape parameter v makes the N-Rao detection performance improved. This case that makes the performance improved can be made by increasing the CNR. And in low CNR, the N-Rao detection has better detection performance.