Ultrasound imaging is one of the most important medical imaging technologies in recent years and is widely used in clinical diagnosis, because of its advantages of non-invasiveness, no radiation, and real-time capability. The actual sound speed in the human body varies from organ to organ, despite the fact that the majority of ultrasound imaging equipment operate under the assumption that it is constant at 1540 m/s. Local sound speed estimation can further provide local sound speed distribution images with quantitative information, forming a new imaging modality to assist traditional ultrasound imaging which is of great significance for improving the diagnostic effect of ultrasound imaging. Therefore, we propose an Ultrasound Channel Attention Full Resolution Residual Network (UCA-FRRN). UCA-FRRN integrates the three-angle input data using strided convolution and divides the extracted features into two processing streams. The UCA-FRRN method uses the ultrasound channel attention module to improve the feature extraction effect of the down-sampling stream. FRRN is used to achieve high precision pixel positioning, and UCA module is used to improve the accuracy of sound speed estimation. For the purpose of evaluating the UCA-FRRN, a plane-wave simulation dataset is built by numerical simulation. In terms of average absolute error (4.79 m/s), standard deviation of error (13.93 m/s), root mean square error (13.93 m/s), and mean structural similarity index measure (0.91), the UCA-FRRN technique performs better than the other examined approaches on the simulated dataset.