A Touchscreen Force Detection Scheme Based on Transfer Learning

Integrating force detection functionality into touchscreens can enhance the user experience. However, achieving high accuracy in force detection on a touchscreen is still challenging. Although the existing linear-model-based algorithm can differentiate between light and heavy touches, it cannot provide more finely-grained force levels. To overcome this limitation, we propose a neural network model for force detection. However, collecting data for each touchscreen and training every network from scratch can seriously affect production efficiency. Therefore, we introduce transfer learning methods in model training, which require only a small amount of data to complete training and improve production speed. Experimental results demonstrate that the proposed neural network model with transfer learning for force detection has an error of 0.027N, which is close to 1/10 of the error of a linear model. Therefore, we conclude that using neural network models with transfer learning can significantly enhance the accuracy of force detection and the production efficiency of touchscreens.