Our submission to the track 1 of the CN-CELEB Speaker Recognition Challenge 2022 (CNSRC 2022) is described by this report. The track 1 task only uses the CN-Celeb training set for training/tuning the system. The objective of this task is to improve performance on the standard CN-Celeb evaluation set. Based on the state-of-the-art SEResnet speaker embedding network, we explore a novel network architecture with split-attention, called ResNeSt, and novel hybrid statistics pooling methods. Based on these techniques, we achieve significant improvement over the SEResnet baselines. Furthermore, in-domain data finetuning, attention back-end methods, speaker-wise adaptive score normalization (AS-Norm) and score calibration on duration efficiently improve the robustness. Finally, our system is a fusion of 23 models and achieves tenth place in the track 1 of CNSRC 2022. The minDCF of our submission is 0.4159, and the corresponding EER is 7.333%.