The late-time signals of semi-airborne transient electromagnetic method are relatively weak and easily masked by environmental noise. To better reduce the impact of noise and ensure the accuracy of subsequent imaging analysis and related research, this paper proposes a Kalman Filtering (KF) improved by Sparrow Search Algorithm (SSA) for denoising semi-airborne transient electromagnetic signals to improve the Signal-to-Noise Ratio (SNR) of raw data. Since the performance of conventional Kalman filtering is affected by the setting of noise covariance matrix, fixed covariance matrix cannot accurately adapt to noise changes in complex noise interference scenarios. Based on this, this paper establishes a dynamic noise covariance matrix and then uses sparrow search algorithm to globally optimize the intensity parameters in the covariance matrix, and adopts the filter with optimal parameter combination for denoising. The denoising effect of the proposed SSA-KF denoising algorithm is compared with traditional methods such as Empirical Mode Decomposition (EMD), Wavelet Transform (WT) and Singular Value Decomposition (SVD) for different noises. Experimental results show that the SNR of the proposed algorithm is improved by 30.57 dB, 24.61 dB, 33.01 dB and 20.03 dB respectively for Gaussian white noise, atmospheric noise, power frequency noise and mixed noise, and the denoising effect is better than other algorithms. In the application of measured data of underground rivers in Lichuan City, the apparent resistivity imaging results of the data denoised by SSA-KF algorithm show clearer abnormal characteristics compared with direct imaging and imaging results after SVD denoising.