|
With the development of Internet and IoT (Internet of Things), the specificity and sensitivity of traditional sensors cannot meet the multi-task requirements in complex environments. The sensor array can solve the limitation that a single sensor can only detect one target to a certain extent. However, the sensor array still cannot solve the recognition of cross-sensitive similar target. This work proposes a solution based on machine learning that can greatly improve the specificity of traditional sensors. It can potentially be used in non-contact breathing diagnosis. Firstly, carbon nanotube materials were chose as electrical carriers, taking advantage of its high specific surface area and high electron mobility characteristics. Secondly, a series of organic molecules are designed to modify carbon nanotubes. The ability of organic materials to capture amines enables response output to amine gases. In order to meet the needs of exhaled breath diagnosis, we need to further distinguish amines with high similarity (ammonia, n-propylamine, diethylamine, triethylamine). Therefore, we have adopted the method of machine learning. The data of resistance with time collected by the sensor array are input as eigenvalues into a pre-designed neural network model for training. The trained model can achieve over 80% recognition accuracy in testing data. The purpose of this work is to propose a new and better solution to the detection of targets with cross-sensitivity, thereby improving the selectivity of the sensor. It provides the possibility to use it under different requirements to achieve intelligent detection.
|
