According to the analysis of the shortcomings of the joint inspection of HPV-DNA viral load testing and TCT liquid thin-layer cytology technical, this paper applies computer aided diagnosis (CAD) technology to improve cervical cancer screening accuracy, and proposes a cervical lesion recognition based on random forest (RF) model. Characteristic value of cervical cell image are extracted to form the cervical cells characteristic data sets, and then to do the Bootstrap sampling, using a random sample to build a decision tree. The results of cervical cell classification are voted out, and the model is applied to cervical cancer screening. According to theoretical analysis and experimental data, compared with traditional combined examination, this method improves the performance of cervical lesion identification, and its results have certain competitiveness and advantages.