Research And Application Of Intelligent Sensory System Based On Deep Learning

Intelligent sensory system is a new type of modern intelligent analysis and detection system by mimicking human visual sense,taste and other perceptual mechanisms.It is usually composed of sensor,signal transmission module and pattern recognition algorithm.The latter plays a decisive role in the accuracy of system detection,but its algorithm for intelligent sensory system is relatively single,which limits its detection efficiency.Meanwhile,the single intelligent sensory system can only obtain one-sided effective information of the sample.Nevertheless,the multi-sensory fusion is mainly implemented on account of the conventional superficial machine learning algorithms,which affects the data fusion effect.With the electronic tongue system and the electronic eye system,this paper proposes an intelligent sensory detection and data fusion method based on deep learning.Moreover,this method is applied to the detection of drugs,agricultural products and food samples in practice.The specific research contents are as follows:(1)Deep learning algorithm is introduced into the field of pattern recognition of electronic tongue.As the electronic tongue signal is a one-dimensional discrete signal,the LeNet-5 is modified to one-dimensional convolutional neural network.What's more,the parameters are optimized to improve its recognition performance.For the purpose of settling the problems of many training parameters,slow training speed and poor generalization of the fully-connected layer of CNNs,the extreme learning machine was employed to replace the fully-connected layer.Then,the optimized model was applied to the traceability detection of wolfberry.The results show that the location of wolfberry could be accurately identified by using LeNet-5 combined with extreme learning machine.(2)In this section,a small sample learning model based on transfer learning is proposed for analyzing the electronic tongue signals,which can effectively solve the overfitting problem of deep learning model caused by the insufficiency of electronic tongue signals.A one-dimensional convolutional neural network is designed,then,the voice signal is set as the source domain and the electronic tongue signals as the target domain.The optimized model is fine-tuned in order to implement transfer learning.This model is introduced to the application of discrimination of wheat storage time.The results show that the convolutional neural network model based on transfer learning can realize more accurate and quick identification for the electronic tongue signals of wheat from different storage time.(3)A fusion model of electronic tongue and electronic eye based on deep learning is proposed.The one-dimensional convolutional and two-dimensional convolutional neural network are respectively designed for fitting the characteristics of electronic tongue signal and electronic eye signal,which are optimized automatically by Bayesian optimization.Moreover,the feature-level fusion is utilized for the fusion of different data.The results illustrate that the detection system using the data fusion method can identify the storage time of Pu-erh tea more quickly and accurately than the single detection system.The above researches can provide a new thought for the application of deep learning in the field of intelligent sensory detection and multi-sensory fusion,and enhance the accuracy and speed of detection of drugs,food and agricultural products.That can provide an effective approach for the detection of the quality and safety of food,drugs and agricultural products.