Stretchable, ultralow detection limit and anti-interference hydrogel strain sensor for intelligent throat speech recognition using Resnet50 neural network

Abstract

Flexible strain sensors have important applications in wearable electronics, human motion monitoring, and healthcare. Soft, compatible, and conductive hydrogels are considered promising stretchable materials. However, hydrogel-based sensors still face challenges with the combined advantages of good electrical conductivity, low strain limit (less than 5%), mechanical strength, and anti-interference. As a result, most hydrogel-based sensors cannot stably recognize human tiny movements (such as pulse, throat, etc.). Herein, we constructed a double network (DN) hydrogel consisting of SA–Zn and P(AA-AM) to boost the mechanical performance and prepare the graphene oxide (GO) modified by polydopamine@Ag nanoparticles (PDA@Ag) via a one-step redox reaction method to enhance the sensitivity. The prepared GO/PDA@Ag/SA–Zn hydrogel strain sensor has an ultralow strain detection limit (0.1%), good gauge factor (GF = 8.29), and high stretchability (600%). It is proven that the hydrogel strain sensor can work as a wearable device to continuously monitor full-scale human motions. Furthermore, it works as a speech interface to detect subtle throat speech movements and which is further classified by a transfer learning algorithm based on the Resnet50 neural network, demonstrating that our sensor can recognize ten kinds of common words expressing physical demands with a precise accuracy of 100%. It provides the possibility to help patients or weak people with voice disorders in the hospital express their requirements. This work provides new insights into fabricating flexible hydrogel strain sensors with promising applications in next-generation intelligent wearable devices for healthcare, medical treatments, diagnosis, therapy, and rehabilitation.