Autonomous Sliding Window Detection Mechanism Based On Deep Learning And Expert System Diagnosis Model For Epileptic State Recognition

With the increase in the capacity and access speed of electronic devices and storage devices,the amount of data generated by human society is increasing,and the research and analysis of massive data has become an urgent need in various industries.Time-series data classification is one of the important techniques in big data analysis,which can be used to classify time-series data.This technique can extract the distribution pattern of data from the huge amount of data,so as to identify the different states of the data.Traditional classification methods for time-series data are usually based on some statistical indicators to classify the whole segment of data.However,this can lead to a large detection computation and low detection efficiency for long or high-dimensional sequences.Currently,some artificial intelligence-based classification methods are emerging,which usually use some machine learning or deep learning models to classify whole-segment time-series data,but such methods require a large amount of labeled data for training,and have poor interpretation for some complex models.Mutation point detection technique is another important technique in big data analysis,which can quickly locate the location of mutation points from the massive data and analyze the causes of them.This technique can effectively extract useful information from massive data,and is a hot direction in the field of big data analysis.In order to improve the efficiency of fast analysis and classification of time-series data,this thesis proposes a BP neural network-based classification method for time-series data segments based on the combination of TSTKS algorithm and sliding window multi-mutation point detection model;proposes a variety of autonomous sliding window detection models based on BP neural networks,which can adaptively adjust the size of the sliding window when detecting and analyzing time-series data;and initially an expert system model that can localize seizures and determine the specific status of seizure stages from the detected data is constructed.First,we repeatedly sliding windows of different fixed sizes to slice the temporal data stream into multiple segments,and use the TSTKS algorithm to detect mutation point on each sub-segment of data,and obtain the relevant feature values within each sub-window.After obtaining the feature values of all windows,the input sample matrix is used as training data to train a BP neural network with the corresponding window size,after which the trained neural network can be used to classify and recognize the temporal data with the specified data size.The experiments based on epileptic EEG and EMG signals demonstrate that the BP neural network-based temporal data segment classification method can accurately classify and identify data segments.Next,a window-alternative set for each category of data segments was constructed based on the distribution of data segments in different categories.Then,three autonomous sliding window detection models based on BP neural networks are proposed using the constructed windowalternative sets.Based on the epileptic EEG and EMG signals,the experimental results show that the three autonomous sliding window mechanisms proposed in this thesis can achieve a balance between detection time consumption and classification accuracy,and lay the foundation for the construction of an expert system.Finally,a prior knowledge judgment mechanism is introduced on the basis of the autonomous sliding window detection model under the regression mechanism,and a lesion recognition method based on neural network and prior knowledge is proposed.And based on the above method,an expert system model capable of seizure localization and recognition of lesion status is designed and implemented by combining three inference rules for lesion status recognition.Based on the clinical epileptic EEG signals,the experimental results show that the expert system model can provide auxiliary guidance for the identification of epileptic disease states.