The accurate sensing of the size and distribution of wear debris in lubricating oil is an important method for evaluating service condition and remaining using life prediction of mechanical equipment. However, in practical application, the output of inductive debris detection sensor is often contaminated by a variety of noise and interference, which makes a challenge to identify the characteristics of debris signals. Therefore, an adaptive method for induced voltage feature identification is proposed in this article. Firstly, the detection signal is multi-filtered by low-pass filter with different cut-off frequencies. Based on the significant difference between multidimensional filtered data, the target signals are located and segmented. Finally, according to the established mathematical model, the signal numerical features are extracted to realize the identification, counting, as well as quantitative analysis of wear debris. Experimental results show that the proposed strategy successfully extract the induced voltage generated by a 70 μm ferromagnetic debris with little distortion of morphological characteristics, which provides a basis for improving the detection performance of the sensor and accurately evaluating the wear state.