A feature extraction and data compression method and device applied to an intelligent Internet of Things

Abstract

The present application relates to the technical field of intelligent internet of things signal processing, more particularly, to a feature extraction and data compression method and device applied to intelligent internet of things, which screens out local extreme points from original collected signals and ignores non-extreme point sampling points, so that the length of processed data can be further shortened and data redundancy can be reduced when high-frequency or long signals are processed. The present application packs the information of local extreme points into asynchronous pulses with time information and space information through asynchronous pulse sequence conversion, without re-encoding, thereby avoiding time delay and energy consumption caused by additional encoding process. When the asynchronous pulse sequence is input into a pulse neural network, the asynchronous effect can be realized, thereby avoiding the problem of multiple neuron waiting between pulses; the asynchronous pulse sequence can reduce the amount of calculation and realize pulse sparseness. The present application solves the problems of redundant data and high energy consumption in the prior art.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent Internet of Things signal processing, and more particularly to a feature extraction and data compression method and device applied to an intelligent Internet of Things. BACKGROUND

[0002] An intelligent Internet of Things collects various types of information in real time through various information sensors and analyzes the data intelligently through machine learning at terminal devices and edge domains. In order to efficiently apply the collected information, the machine learning more often adopts a manner based on a spiking neural network.

[0003] An existing event-driven chip gradually becomes a mainstream of Internet of Things chip design because of its low power consumption and time delay, so processing the original collected signal into a pulse form can better interface with this type of chip.

[0004] However, in the existing methods, one method is to filter the signal through a filter, take the signal energy output by each band-pass filter as the basic feature of the signal, and encode the feature into a pulse signal. This method requires high-precision Nyquist sampling and additional time delay and energy consumption for the filter itself and the encoding process, which not only reduces the reaction speed of the chip to external information, but also increases the overall power consumption. Another method is to use event-driven sampling, the basic principle of which is that when the input signal changes greatly, it is considered that the sampling time occurs and the position of the sampling point is recorded. While saving power consumption, the number of sampling points can be reduced, but when inputting high-frequency signals, the obtained sampling points are still too many and redundant, which puts great pressure on data processing and transmission. SUMMARY

[0005] Therefore, it is necessary to provide a feature extraction and data compression method and device applied to an intelligent Internet of Things to solve the problems of redundant data and high energy consumption in the existing methods.

[0006] The present application adopts the following technical solutions:

[0007] In a first aspect, the present application discloses a feature extraction and data compression method applied to an intelligent Internet of Things, which is used to convert an original collected signal into a pulse signal.

[0008] The feature extraction and data compression method applied to an intelligent Internet of Things includes the following steps:

[0009] Step one, normalizing the original collected signal to obtain a normalized data set; wherein the original collected signal includes M' original sampling points, the normalized data set includes M' normalized sampling points, and the original sampling points and the normalized sampling points correspond one by one;

[0010] Step two, screening the normalized data set to obtain a local extreme data set; wherein the local extreme data set includes M local extreme points;

[0011] Step three, time characteristic extraction and space characteristic extraction are performed on the M local extreme points.

[0012] Step four, for a single local extreme point, the extracted time characteristic and space characteristic are fused, and packaged into an asynchronous pulse with time information and space information; the asynchronous pulses corresponding to the M local extreme points are combined into an asynchronous pulse sequence to form a pulse signal.

[0013] The feature extraction and data compression method applied to the intelligent Internet of Things realizes the method or process according to the embodiments of the present disclosure.

[0014] In a second aspect, the present disclosure discloses a feature extraction and data compression device applied to the intelligent Internet of Things, which uses the feature extraction and data compression method applied to the intelligent Internet of Things disclosed in the first aspect.

[0015] A feature extraction and data compression device applied to the intelligent Internet of Things includes a normalization module, an extreme value screening module, a time characteristic extraction module, and a space characteristic extraction module.

[0016] The normalization module is used to normalize the original collected signal to obtain a normalized data set. The original collected signal includes M' original sampling points, and the normalized data set includes M' normalized sampling points. The original sampling points and the normalized sampling points correspond one by one. The extreme value screening module is used to screen the normalized data set to obtain a local extreme data set. The local extreme data set includes M local extreme points. The time characteristic extraction module is used to extract the time characteristics of the M local extreme points. The space characteristic extraction module is used to extract the space characteristics of the M local extreme points. The asynchronous pulse sequence conversion module is used to fuse the extracted time characteristics and space characteristics for a single local extreme point, and package them into an asynchronous pulse with time information and space information; the asynchronous pulses corresponding to the M local extreme points are combined into an asynchronous pulse sequence to form a pulse signal.

[0017] The feature extraction and data compression device applied to the intelligent Internet of Things realizes the method or process according to the embodiments of the present disclosure.

[0018] In a third aspect, the present disclosure discloses a readable storage medium. The readable storage medium stores computer program instructions. When the computer program instructions are read and run by a processor, the steps of the feature extraction and data compression method applied to the intelligent Internet of Things disclosed in the first aspect are executed.

[0019] Compared with the prior art, the present disclosure has the following beneficial effects:

[0020] 1. The application integrates digital-to-analog conversion and feature extraction, avoiding high-precision ADC and complex digital processing, while reducing the subsequent pulse neural network calculation.

[0021] 2. The application filters out local extreme points from the original collected signal, ignoring non-extreme sampling points, which can further shorten the length of the processed data and reduce data redundancy when processing high-frequency or long signals.

[0022] 3. The application converts local extreme points into asynchronous pulses with time and space information by asynchronous pulse sequence conversion, without the need for re-encoding, avoiding the time delay and energy consumption caused by additional encoding process. The time feature contained in the asynchronous pulse sequence is represented by the time difference between adjacent extreme points, which can achieve asynchronous effect when inputting into the pulse neural network, avoiding the problem of multiple neurons waiting between pulses; the space feature contained in the asynchronous pulse sequence is converted into a 2 N row binary sequence, and only one row has a value of 1 for the real-time value of the space feature of each extreme point, which can reduce the amount of calculation and achieve pulse sparsity. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 The flowchart of the feature extraction and data compression method applied to the intelligent Internet of Things in embodiment 1 of the application;

[0024] Figure 2 The example graph of the original collected signal provided in embodiment 1;

[0025] Figure 3 The example graph of the normalized data set provided in embodiment 1;

[0026] Figure 4 The example graph of the local extreme data set provided in embodiment 1;

[0027] Figure 5 The example graph of the extracted time feature provided in embodiment 1;

[0028] Figure 6 The example graph of the extracted space feature provided in embodiment 1;

[0029] Figure 7 The schematic diagram of the asynchronous pulse conversion provided in embodiment 1;

[0030] Figure 8 The example graph of the asynchronous pulse conversion provided in embodiment 1. DETAILED DESCRIPTION

[0031] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments of the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative work belong to the scope of protection of the present application.

[0032] It should be noted that when a component is referred to as being "mounted on" another component, it can be directly on the other component or there can be a middle component. When a component is referred to as being "disposed on" another component, it can be directly disposed on the other component or there can be a middle component. When a component is referred to as being "fixed on" another component, it can be directly fixed on the other component or there can be a middle component.

[0033] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the description of the application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0034] Embodiment 1

[0035] Please refer to Figure 1 , Figure 1 The flow chart of the feature extraction and data compression method applied to the intelligent Internet of Things in the present application is used to convert the original collected signal into a pulse signal.

[0036] As Figure 1 shown, the feature extraction and data compression method applied to the intelligent Internet of Things includes the following steps:

[0037] Step 1: Normalizing the original collected signal to obtain a normalized data set.

[0038] The original collected signal includes M' original sampling points, and the normalized data set includes M' normalized sampling points. The original sampling points and the normalized sampling points correspond one by one.

[0039] It should be noted that normalization only changes the size of the sampling point value, but does not change the time information of each sampling point. That is, the M' original sampling points correspond to M' different collection times; the M' normalized sampling points also correspond to M' different collection times.

[0040] As Figure 2 , Figure 3The figure shows the original acquisition signal graph and the normalized data set graph provided by the embodiment 1 respectively. In the figure, the horizontal coordinate is the sampling point number, and the vertical coordinate is the sampling point amplitude. The original acquisition signal is generally obtained by using the collector, the sampling frequency is S, and the sampling time can be recorded by the collector. Of course, the sampling time can also be calculated according to the sampling frequency and the sampling point number.

[0041] In short, normalization is used to limit all original acquisition signals in a certain range of data space according to the maximum and minimum values of the original acquisition signal amplitude and through an algorithm, which is suitable for comparison.

[0042] Specifically, the method for normalization includes:

[0043] According to the maximum and minimum values of the M' original sampling points, a normalization equation is constructed to limit the M' original sampling points in the data space [-a, 1-a] to form M' normalized sampling points.

[0044] The normalization equation is:

[0045]

[0046] In the formula, x represents the value of the original sampling point, y represents the value of the normalized sampling point, min(x) represents the minimum value of the original sampling point, max(x) represents the maximum value of the original sampling point, and a represents a linear adjustment value.

[0047] This is because, if the normalization equation does not contain the -a term, it is to limit the original sampling points in [0, 1]; after adding the -a term, [0, 1] is offset. Generally, a is recommended to be 0.5. That is, it is limited in the data space [-0.5, 0.5] (which can also be represented as [-a, a]), which is symmetrical relative to the 0 coordinate, which is convenient for display.

[0048] Step two, screening the normalized data set to obtain a local extreme data set. The local extreme data set includes M local extreme points.

[0049] This step two is to compress the length of the data and eliminate redundant data. It should be noted that the M local extreme points correspond to M different acquisition times. The acquisition time of the mth local extreme point is t m .

[0050] Specifically, the method for screening in step two includes:

[0051] Determine whether the current normalized sampling point is greater than or less than the two adjacent normalized sampling points at the same time; if so, the current normalized sampling point is a local extreme point; otherwise, the current normalized sampling point is not a local extreme point.

[0052] By ignoring sampling points between adjacent local extrema and retaining local extrema, data redundancy can be reduced, the final asynchronous pulse sequence length can be shortened, energy consumption can be reduced, and the device response speed can be improved.

[0053] See Figure 4 ,right Figure 3 After performing step two, we obtain... Figure 4 The dataset contains six local extrema (named P1, P2, P3, P4, P5, and P6 in sequence along the positive x-axis), corresponding to six different acquisition times.

[0054] Step 3: Extract the temporal and spatial characteristics of the M local extreme points.

[0055] Temporal feature extraction aims to preserve the temporal information of local extrema. Spatial feature extraction aims to preserve the spatial information of local extrema.

[0056] Temporal and spatial feature extraction are not strictly ordered and can be performed simultaneously.

[0057] (1) Extraction of time characteristics:

[0058] For the m-th local extremum point, obtain the number of normalized sampling points Δs between it and the (m-1)-th local extremum point. m-1 Divide by the sampling frequency S to obtain the time difference Δt. m-1 , and serve as the time characteristic of the m-th local extremum point; 1 < m ≤ M.

[0059] For Δs m-1 This can be obtained by directly counting the number of normalized sampling points between the (m-1)th local extremum. Alternatively, the number of sampling points s between the m-th local extremum and the initial point of the normalized dataset can be counted first. m Then count the number of sampling points s between the (m-1)th local extremum point and the initial point of the normalized dataset. m-1 Find the difference to obtain Δs m-1 .

[0060] This can be expressed as a formula:

[0061]

[0062] In the formula, S represents the sampling frequency of the original acquired signal.

[0063] It should be noted that for the first local extremum, the number of normalized sampling points Δs0 between it and the initial point of the normalized dataset is obtained, and then divided by the sampling frequency S to obtain the time difference Δt0, which is used as the time feature of the first local extremum.

[0064] Referring to Figure 5 , the local extreme data set is subjected to time characteristic extraction, and six time characteristics of Δt0-Δt5 are obtained. Figure 4

[0065] (2) Spatial characteristic extraction:

[0066] The data space [-a, 1-a] is evenly divided into two N windows, each window is provided with a corresponding index and has upper and lower thresholds. Here, the division is according to the N bit specification, that is, the span of each window is 1 / 2 N .

[0067] In this embodiment 1, referring to Figure 6 , if the windows are indexed in order from bottom to top, the index is from 0 to 2 N -1. Among them, the window index below the abscissa axis is 0 to 2 N-1 -1, and the window index above the abscissa axis is 2 N-1 to 2 N -1.

[0068] In order from bottom to top, the first window is the window with index 0, the lower threshold is -a, and the upper threshold is -a+1 / 2 N ; the second window is the window with index 1, the lower threshold is -a+1 / 2 N , and the upper threshold is -a+2 / 2 N ; the third window is the window with index 2, the lower threshold is -a+2 / 2 N , and the upper threshold is -a+3 / 2 N ; and so on.

[0069] If the mth local extreme point is greater than the lower threshold of the m'th window and less than or equal to the upper threshold of the m'th window, it is determined that the mth local extreme point is in the m'th window, and at this time the index of the m'th window is taken as the spatial characteristic initial value of the m'th local extreme point; m'∈[1,2 N ].

[0070] As Figure 6 shown in this embodiment 1, the six local extreme points are respectively in different windows and have different sampling times. Specifically, along the positive direction of the abscissa, the indexes of the windows where the six local extreme points are located are: 2 N-1 -3, 2 N-1 +2, 1, 2 N -2, 2 N-1 -2, 2 N-1 -1.

[0071] ​It should be noted that different local extreme points may be in the same window, but different local extreme points must have different collection times.

[0072] It is obvious that the more the number of windows, the more fine the spatial feature of the local extreme point is represented. After the above processing, the time characteristics and spatial characteristics of the M local extreme points are saved. For the i-th local extreme point (denoted as P i ), its signal can be represented as (Δt i-1 ,p i ), wherein p i indicates the index of the window where the i-th local extreme point is located, i∈[1,M].

[0073] It should be noted that 2 N windows are used to change the spatial feature in real time according to the collection time of the m-th local extreme point when the asynchronous pulse is performed subsequently.

[0074] Step four, for a single local extreme point, fuse the extracted time characteristics and spatial characteristics, and pack into an asynchronous pulse with time information and spatial information; the asynchronous pulses corresponding to the M local extreme points form an asynchronous pulse sequence, forming a pulse signal.

[0075] This step four is to pack each extreme point into an asynchronous pulse to generate an asynchronous pulse sequence with time and space characteristics. That is, the asynchronous pulse sequence is composed of asynchronous pulses corresponding to all extreme points.

[0076] It should be emphasized that the length value of the asynchronous pulse sequence is the same as the number value of the local extreme points, that is, the two are consistent in the time dimension.

[0077] Specifically, the asynchronous pulse sequence includes 1 row of time sequence, 2 N rows of space sequence.

[0078] Among them, 1 row of time sequence is used to reflect the time characteristics of the M local extreme points. 2 N rows of space sequence are used to reflect the real-time value of the spatial characteristics of the M local extreme points.

[0079] 2 N rows of space sequence correspond to the index of 2 N windows one by one; wherein, if the m-th local extreme point is in the m'-th window, between t m and t m-1 , the index of the m'-th window corresponds to the space feature real-time value of the space sequence, which is placed at 1, and the rest 2 N -1 rows of space sequence space feature real-time value is placed at 0. Wherein, t m indicates the collection time of the m-th local extreme point, tm-1 the collection time of the m-1th local extreme point.

[0080] Referring to Figure 7 , that is, if the mth local extreme point is in the m'th window, the m'th window continuously sends out high level between t m and t m-1 , and the other 2 N -1 windows remain low level, thus forming a pulse signal.

[0081] Taking Figure 6 as an example, referring to Figure 8 , based on the 6 local extreme points P1-P6 of Figure 6 , six groups of data are generated, constituting a group of time-continuous data stream:

[0082] For P1, it forms a group of data between the collection time t0-t1; the group of data includes a row of time difference Δt0 and 2 N rows of binary sequence. Wherein, Δt0=t1-t0. The 2 N row binary sequence in the row corresponding to the window with index 2 N-1 -3 is 1, and the remaining 2 N -1 rows are 0.

[0083] For P2, it forms a group of data between the collection time t1-t2; the group of data includes a row of time difference Δt1 and 2 N rows of binary sequence. Wherein, Δt1=t2-t1. The 2 N row binary sequence in the row corresponding to the window with index 2 N-1 +2 is 1, and the remaining 2 N -1 rows are 0.

[0084] For P3, it forms a group of data between the collection time t2-t3; the group of data includes a row of time difference Δt2 and 2 N rows of binary sequence. Wherein, Δt2=t3-t2. The 2 N row binary sequence in the row corresponding to the window with index 1 is 1, and the remaining 2 N -1 rows are 0.

[0085] For P4, P5, P6, similar to the above, no longer described.

[0086] In general, each group of data contains a row of time difference information and 2 N rows of binary sequence. The window corresponding to each extreme point will continuously send out pulses in the corresponding time interval, and the whole is continuous in time without interval and idle time.

[0087] Thus, since there is only one local extreme point at the same acquisition time, the local extreme point has only one pulse after the above processing. In other words, 2 N The spatial sequence is converted into a 2 N row binary sequence, and only one row has a value of 1 for the real-time value of the spatial feature of each extreme point. This makes the pulse signal sparse.

[0088] Through the above operation, the 2 N channel signals can directly generate a pulse signal, without the need for re-encoding of the time feature and the spatial feature using a filter or the like, thereby avoiding time delay and energy consumption caused by the additional encoding process. Meanwhile, since there is a time sequence, the 2 N channel signals can be asynchronous when input into a pulse neural network, thereby avoiding the problem of multiple neurons waiting between pulses and achieving a low-power effect.

[0089] Embodiment 2

[0090] The embodiment 2 discloses a feature extraction and data compression device applied to an intelligent Internet of Things, which uses the feature extraction and data compression method applied to an intelligent Internet of Things of the embodiment 1.

[0091] A feature extraction and data compression device applied to an intelligent Internet of Things includes a normalization module, an extreme value screening module, a time characteristic extraction module, and a spatial characteristic extraction module.

[0092] The normalization module is configured to normalize an original acquisition signal to obtain a normalized data set. The original acquisition signal includes M' original sampling points, and the normalized data set includes M' normalized sampling points. The original sampling points correspond to the normalized sampling points in a one-to-one manner. The extreme value screening module is configured to screen the normalized data set to obtain a local extreme data set. The local extreme data set includes M local extreme points. The time characteristic extraction module is configured to extract time characteristics of the M local extreme points. The spatial characteristic extraction module is configured to extract spatial characteristics of the M local extreme points. The asynchronous pulse sequence conversion module is configured to fuse the extracted time characteristics and spatial characteristics of a single local extreme point, and pack the single local extreme point into an asynchronous pulse with time information and spatial information. The asynchronous pulse sequence conversion module is further configured to group asynchronous pulses corresponding to the M local extreme points to form an asynchronous pulse sequence, thereby forming a pulse signal.

[0093] Embodiment 3

[0094] The embodiment 3 discloses a readable storage medium, and the readable storage medium stores computer program instructions. When the computer program instructions are read and executed by a processor, the steps of the feature extraction and data compression method applied to an intelligent Internet of Things of the embodiment 1 are executed.

[0095] The method of the embodiment 1 can be applied in the form of software, such as a program designed to be independently run on a computer readable storage medium, which can be a U disk, and the program is designed to start the whole method through external triggering.

[0096] The technical features of the above embodiments can be combined in any manner. For the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combinations of the technical features do not contradict each other, they should be considered as falling within the scope of the present disclosure.

[0097] The above embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be pointed out that, for ordinary skilled persons in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which all fall within the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A feature extraction and data compression method applied to the Internet of Things, for converting original collected signals into pulse signals, characterized in that, The feature extraction and data compression method applied to the intelligent Internet of Things comprises the following steps: Step one, normalizing the original collected signal to obtain a normalized data set; wherein the original collected signal comprises M' original sampling points, and the normalized data set comprises M' normalized sampling points, and the original sampling points correspond to the normalized sampling points one by one; Step two, screening the normalized data set to obtain a local extreme value data set; wherein the local extreme value data set comprises M local extreme value points; Step three, extracting the time characteristics and the space characteristics of the M local extreme value points; The method for extracting the time characteristics comprises: For the mth local extreme point, the number of normalized sampling points existing between it and the (m-1)th local extreme point is obtained as Δ s m-1 , and divided by the sampling frequency again S , to obtain the time difference Δ t m-1 , and taken as the time feature of the mth local extreme point; 1 < m ≤ M; For the 1st local extreme point, get the number of normalized sampling points Δ between it and the beginning of the normalized data set s 1, divide by the sampling frequency again S , get the time difference Δ t 1, and take it as the time feature of the 1st local extreme point; The method for extracting the space characteristics comprises: Divide the data space [-a, 1-a] into 2 N windows averagely, each window is provided with a corresponding index and has upper and lower thresholds; [-a, 1-a] is a normalized defined data space; a represents a linear adjustment value; If the mth local extreme point is greater than the lower threshold of the m'th window and less than or equal to the upper threshold of the m'th window, it is determined that the mth local extreme point is in the m'th window, and the index of the m'th window is taken as the initial value of the spatial feature of the mth local extreme point; m'∈[1,2 N ]. Step four, for a single local extreme value point, fusing the extracted time characteristics and the space characteristics, and packing into an asynchronous pulse with time information and space information; and grouping the asynchronous pulses corresponding to the M local extreme value points into an asynchronous pulse sequence to form a pulse signal. 2.The feature extraction and data compression method applied to the intelligent Internet of Things according to claim 1, wherein, The method for normalizing in step one comprises: According to the maximum value and the minimum value of the M' original sampling points, a normalization equation is constructed to limit the M' original sampling points in the data space [-a, 1-a] to form M' normalized sampling points. 3.The feature extraction and data compression method applied to the intelligent Internet of Things according to claim 2, characterized in that, a=0.5。 4.The feature extraction and data compression method applied to the intelligent Internet of Things according to claim 1, wherein, The method for screening in step two comprises: Judging whether the current normalized sampling point is greater than or less than the two adjacent normalized sampling points at the same time; if yes, the current normalized sampling point is a local extreme value point; otherwise, the current normalized sampling point is not a local extreme value point. 5.The feature extraction and data compression method applied to the intelligent Internet of Things according to claim 1, characterized in that, The length value of the asynchronous pulse sequence is the same as the number value of the local extreme value points. 6.The feature extraction and data compression method applied to the intelligent Internet of Things according to claim 1, wherein, The asynchronous pulse sequence comprises a 1 row time sequence, 2 N row spatial sequence; Wherein, 1 row time sequence is used for reflecting time characteristics of M local extreme points; 2 N Row space sequence is used for reflecting real-time values of space characteristics of M local extreme points; 2 N The row space sequence is in one-to-one correspondence with the indexes of the 2 N windows; wherein, if the mth local extreme point is in the m'th window, the real-time value of the spatial feature of the spatial sequence corresponding to the index of the m'th window is set to 1, and the real-time values of the spatial features of the spatial sequences corresponding to the indexes of the other 2 t m -1 windows are set to 0, the m'th window continuously outputs a high level, and the other 2 t m-1 -1 windows maintain a low level, forming a pulse signal. N N -1 windows maintain a low level, forming a pulse signal.​ wherein, t m the acquisition time of the mth local extreme point, t m-1 the acquisition time of the m-1th local extreme point.

7. A feature extraction and data compression device applied to an intelligent Internet of Things, characterized by, The feature extraction and data compression method applied to the intelligent Internet of Things is used; The feature extraction and data compression device applied to the intelligent Internet of Things comprises: a normalization module for normalizing the original collected signal to obtain a normalized data set; wherein the original collected signal comprises M' original sampling points, and the normalized data set comprises M' normalized sampling points, and the original sampling points correspond to the normalized sampling points one by one; an extreme value screening module for screening the normalized data set to obtain a local extreme value data set; wherein the local extreme value data set comprises M local extreme value points; a time characteristic extraction module for extracting the time characteristics of the M local extreme value points; a space characteristic extraction module for extracting the space characteristics of the M local extreme value points; and an asynchronous pulse sequence conversion module for, for a single local extreme value point, fusing the extracted time characteristics and the space characteristics, and packing into an asynchronous pulse with time information and space information; and grouping the asynchronous pulses corresponding to the M local extreme value points into an asynchronous pulse sequence to form a pulse signal. The readable storage medium stores computer program instructions, and when the computer program instructions are read and run by a processor, the steps of the feature extraction and data compression method applied to the intelligent Internet of Things are executed.

8. A readable storage medium, characterized by, ​

Images