Study On Industrial Gas Recognition Based On Deep Learning

Because of the rapid development of information technology, machine learning has emerged in many fields, such as security inspection, industrial production and so on. A great deal of technology has come into being, which has brought convenience to human life.Electronic nose can be used to detect and analyse the gases with any of various of smell. As a typical application of machine smell, Electronic nose has already penetrated into all walks of life. In the field of safety inspection, industrial production, environmental protection, medical diagnosis, gas recognition systems has been widely used.The gas recognition system uses a more complex time series signal, which is generated by vibrations of sensitive material due to adsorption by molecules of gas. Sensitive material,environmental factors, the type and density of gases all affect the production of this signal.Therefore, people usually design features by hand, and combine support vector machines and other methods to solve the problem of complicated waveform of gas data. In the identification of the sensor, the response value of the sensor will gradually decline, and the overall distribution of the data will change, which makes the analysis of the gas data become more difficult and the accuracy will be reduced. This makes gas data analysis difficult and the rate of accuracy drops, this phenomenon is called sensor drift. In order to eliminate this phenomenon, many methods have been adopted in the field of signal processing.In this thesis, to get started, deep learning method is introduced to compensate for the shortcomings of feature extraction methods in traditional gas recognition. The features which can reflect the essential of the raw gas data are automatically obtained by using the characteristics of deep learning networks and the characteristics of unsupervised learning.This can solve the shortcomings of traditional methods of feature extraction, and the unlabeled gas recognition problem can be effectively solved. Secondly, in the case of obvious drift of gas data, the accuracy of gas recognition can be improved obviously by using deep learning technology. This is because the method can get a better and more accurate input data than at first. It provides more reference for solving the drift data and provides more feasible solution in the aspect of drift compensation. This provides a reference for the detection of toxic gases and unknown gases in industrial production.