Copy number variations (CNVs) significantly influence the diversity of the human genome and the occurrence of many complex diseases. The next-generation sequencing (NGS) technology provides rich data for detecting CNVs, and the read depth (RD)-based approach is widely used. However, low CN (copy number of 3-4) duplication events are challenging to identify with existing methods, especially when the size of CNVs is small. In addition, the RD-based approach can only obtain rough breakpoints. We propose a new method, CNV-PCC (detection of CNVs based on Principal Component Classifier), to identify CNVs in whole genome sequencing data. CNV-PPC first uses the split read signal to search for potential breakpoints. A two-stage segmentation strategy is then implemented to enhance the identification capabilities of low CN duplications and small CNVs. Next, the outlier scores are calculated for each segment by PCC (Principal Component Classifier). Finally, the OTSU algorithm calculates the threshold to determine the CNVs regions. The analysis of simulated data results indicates that CNV-PCC outperforms the other methods for sensitivity and F1-score and improves breakpoint accuracy. Furthermore, CNV-PCC shows high consistency on real sequencing samples with other methods. This study demonstrates that CNV-PCC is an effective method for detecting CNVs, even for low CN duplications and small CNVs.