High-frequency surface wave radar (HFSWR) systems are usually located on the coast and measure electromagnetic backscatter from the sea in the 3-30 MHz band. Based on the theory of ground wave propagation, the detection distance of a low-frequency HFSWR system could be longer that of a high-frequency one. Therefore, a low-frequency HFSWR system is more suitable to extract the significant wave height at a distance. The Barrick's algorithm-based significant wave height extraction method is widely used, and the accuracy of the method has been validated by a large number of experiments. However, it is important to note that under the low sea state, the wave conditions of the data recorded by a low-frequency HFSWR system may not satisfy the wave height condition required in Barrick's algorithm. Hence, the algorithm may produce unreliable results. To address this issue, a method based on Barrick's algorithm, a new non-linear model, and the radial basis function neural network is proposed. Simulations on real-world data collected on the coast of the Tianjin from 1 May 2015 to 25 May 2015 support the method developed.