All-fiber current sensors exhibit excellent insulation performance, strong immunity to electromagnetic interference, and good dynamic response characteristics, making them suitable for various current measurement scenarios in high-voltage power systems. However, optical fibers and optical components are susceptible to temperature variations, which may introduce additional birefringence. In addition, the relatively complex optical path structures used in existing schemes further aggravate the temperature coupling of the sensor. Therefore, temperature drift and structural complexity remain key bottlenecks that limit the measurement stability of all-fiber current sensors. In this article, an all-fiber current sensing method based on single-polarization modulation and an improved sparrow search algorithm is proposed, which enables temperature drift compensation with a simplified optical path structure. First, a single-polarization circulator and a Faraday rotator mirror are connected to the two ends of the sensing fiber coil, respectively. The circulator performs single-polarization modulation while outputting the incident light and receiving the reflected light, thereby realizing current sensing. Based on Jones matrix theory, the expression of the sensing output is derived, and the influence mechanism of temperature on the sensing parameters is analyzed. Secondly, the sensing parameters are classified into current-dependent and current-independent parameters. The former are adaptively adjusted according to the variation of the latter, so that temperature drift compensation can be achieved during current demodulation. Since the relationship between these two types of parameters is highly nonlinear, a nonlinear support vector machine is employed for fitting. Meanwhile, an improved sparrow search algorithm is used to determine the optimal hyperparameters, thereby improving the compensation performance. Finally, temperature experiments from -20~66℃ are carried out for validation. The results show that the maximum ratio error of the proposed method is reduced from 1.30% to 0.49% after compensation. In addition, the proposed structure requires only two optical modulation devices, outperforming existing polarimetric all-fiber sensing methods.