Abstract: To address the issues of low measurement accuracy and susceptibility to temperature effects in traditional non-dispersive infrared absorption CO₂ sensors, the temperature compensation algorithm of the sensor was investigated by combining it with the sparrow search algorithm–back propagation (SSA-BP) neural network model. Using a low-cost light source and pyroelectric detector and employing Teensy 4.0 open source hardware and software as the development platform, non-dispersive infrared absorption CO₂ sensors were developed based on dual-channel differential technology and a digital phase-locked method. To compensate for the nonlinear fitting error in the full-scale (0-10000 μmol·mol^-1) detection range, the high and low concentration segmented fitting method was adopted. The fitting results showed that the correlation coefficients for high and low concentrations were 0.99983 and 0.99979, and the corresponding maximum detection errors were 0.86% and 4.3%, respectively. A stability experiment was carried out on a 1100 μmol·mol^-1 CO₂ standard gas for 1 h, yielding a relative error of 1.2%. Finally, based on the back propagation neural network and SSA-BP hybrid neural network model, a comparative verification of the two temperature compensation models was carried out. The results showed that the maximum error of the temperature compensation model based on the SSA-BP hybrid neural network was 2.3% over the full range, which satisfied the requirements of vegetation carbon fixation detection.