Based on microwave resonance technology, a microwave amplitude change signal acquisition device of an activated carbon filter rod is created, and noise reduction and baseline deduction of the microwave amplitude change signal is combined with Gaussian filtering and the penalty least squares algorithm to improve the efficiency and accuracy of the detected activated carbon content. At first, the filtering effects of different Gaussian window lengths are compared. Following that, four processing methods were used to correct the baseline of the signal spectrum, including asymmetric least squares, adaptive iteratively reweighted penalized least squares, asymmetric reweighted penalized least squares, and multi-constrained reweighted penalized least squares, the peak height, peak area, and half-peak width of the microwave amplitude change signals after baseline correction were obtained. The models developed using the partial least squares algorithm, neural networks, and support vector regression were then compared. Finally, the repeatability, accuracy, and sensitivity of the detection device were assessed. The results revealed that the best model for activated carbon weight was “peak area-activated carbon weight” with a model determination coefficient of 0.9924, an average absolute error of 0.7979 mg, and a relative standard deviation of 1.4962%. The maximum standard deviation of activated carbon weight repeatability was 1.85 mg, the minimum absolute deviation for activated carbon weight testing was 0.03 mg, and the minimum relative deviation of activated carbon weight detection was 0.05%, resulting in a quick and effective method for quantifying analysis of activated carbon in an activated carbonfilter rod for tobacco.