Piezoelectric films were used as sensor sensitive materials. A grain cleaning loss rate monitoring device suitable for corn harvesting was designed. The impact sensor, signal processing circuit and mounting device were designed, and the corresponding processing algorithm was used to collect the impact signal. Firstly, under different experimental conditions of impact angle and height, the impact signal acquisition experiments of different sizes of corn kernels and impurities were carried out. The main characteristics of impact signal were also extracted. Secondly, the support vector machine was used for multi-classification. The model was trained by the support vector machine multi-classification algorithm and real-time classification was implemented on the monitoring device. And the classification model was validated by using different corn varieties and moisture content. Then, under the conditions of different fan speeds and cleaning screen opening degrees, the relationship between the number of grains detected by the sensor and the total loss during the test time was obtained. And the real-time cleaning loss rate was calculated according to the grain flow value. Finally, the monitoring device was mounted on a 4YL-8 combine harvester and field trials were conducted. The results showed that the average relative error of the monitoring device compared with manual detection was 12.98%, which can provide feedback for the control of the harvester.