A Wavelet‐ALS‐Assisted CNN–BiGRU Method for Peak Recognition in CCD‐Based Capillary Electrophoresis Fluorescence Detection

Capillary electrophoresis (CE) is an efficient, rapid, and sensitive separation and analysis technique that has been widely applied in biochemistry, pharmaceutical research, and environmental monitoring. However, charge-coupled device (CCD)-based CE fluorescence signals are susceptible to baseline drift, random noise, and channel crosstalk, which reduce the accuracy of peak recognition. To meet the demand for high-precision signal analysis, this study proposes a convolutional neural network (CNN)-bidirectional gated recurrent unit (BiGRU) peak recognition algorithm on the basis of wavelet transform and one-dimensional asymmetric least squares (ALS). In the proposed system, an FPGA-based platform is used for real-time signal acquisition and data transmission, whereas the host computer performs subsequent processing of the received spectral data. Experimental results show that the proposed method achieves high-accuracy peak recognition on the spectral dataset, with Precision, Recall, and F1-score reaching 98.44%, 99.93%, and 99.18%, respectively. These results demonstrate its good robustness and practical applicability, providing technical support for the miniaturization and intelligent development of portable gene sequencers.