In this letter, a novel microwave sensor is proposed to classify different microfluidic liquids and simultaneously detect their concentrations. The proposed three-layer microwave sensor consists of a reoriented complementary split-ring resonator (RCSRR) rotor on top and a slot in the middle. In particular, the slot connects the ports and acts as the microfluidic channel. In this circumstance, the liquid's dielectric property will directly change the electrical parameters of the RCSRR structure and slot. We also implement the RCSRR structure into a rotor, extending the sensing paradigm from 1-D frequency responses into 2-D spectrograms. Finally, a convolutional neural network (CNN) is deployed to extract features from the spectrograms to classify liquids and detect concentration simultaneously. Alcohol, glucose, and saline solutions are tested. With less than 50~\mu \text{L} of liquid, the classification accuracy is 100%, and the standard derivation of the relative error for concentration sensing is 2.81%. These results demonstrate the remarkable performance of the proposed microwave microfluidic sensor.