In this paper, in order to make fiber Bragg grating spectra easier to overlap, it is proposed to use ultra-short fiber Bragg grating to build a sensor network, and for serial wavelength division multiplexing (WDM) fiber Bragg grating (FBG) sensor networks using charge-coupled device (CCD) interrogator as data acquisition devices, an efficient method for measuring strain sensor signals is presented, which combines a deep belief network (DBN) with the sparrow search algorithm (SSA). The FBG sensor network uses serial WDM connectivity, negating the need for optical switches and reducing latency of the whole sensor system. The application of a low-precision, low-resolution CCD interrogator as the data acquisition device enhances the model's generalizability and facilitates its implementation in real-world projects. DBN, a generative graphical model in machine learning, for learning features from overlapping spectra of FBGs and build the center wavelength detection model. SSA is a swarm intelligence algorithm, for optimizing the hyperparameters of the DBN model. Experimental results show that even using spectral data collected by a CCD interrogator, the DBN-SSA model can achieve good demodulation accuracy and speed, with an optimal root mean square error of 1.68pm and a single inference time of 1.4 ms. In summary, the demodulation system offers a dependable and effective solution for FBG sensor networks with limited data precision.