A bionic fruit fly intelligent identification method, system and computer storage medium

Abstract

The present application relates to the field of information retrieval, and in particular to an intelligent identification method, system and computer storage medium for bionic fruit flies. The intelligent identification method for bionic fruit flies comprises: acquiring multiple groups of sampling data sets at different sampling time points, wherein the sampling data sets include multiple sampling data collected by multiple acquisition units; performing response state data identification on the multiple groups of sampling data sets to obtain response state data identification results; based on the response state data identification results, determining multiple groups of steady-state data sets and at least one group of response state data sets in the multiple groups of sampling data sets; performing data identification processing based on the steady-state data sets and the response state data sets to obtain detection target data corresponding to the response state data sets; calling an intelligent identification model to determine a detection target result corresponding to the detection target data; by obtaining the detection target data corresponding to the response state data sets, not only the recognition calculation amount and data storage space are reduced, but also the recognition accuracy and recognition detection speed are improved.

Description

Technical Field

[0001] The present application relates to the field of information retrieval, and in particular to an intelligent identification method, system and computer storage medium for bionic fruit flies. Background Art

[0002] Existing detection technologies based solely on deep learning can only identify target classification information, but cannot precisely identify the specific target. To identify a specific target, the target must be further categorized into more and more detailed categories, which often results in complex search computations and requires a large amount of data storage space. Furthermore, the large amount of data required for retrieval results in long search times and low accuracy. Therefore, a bionic fruit fly-inspired intelligent recognition method is proposed to address these issues. Summary of the Invention

[0003] In response to the above-mentioned problems in the prior art, the purpose of this application is to provide an intelligent identification method for bionic fruit flies. By obtaining detection target data corresponding to a response state data set, the data dimension in the detection target data is higher than the data dimension in the corresponding response state data set, and the information in the detection target data is binary information, which not only reduces the recognition calculation amount and data storage space, but also improves the recognition accuracy and recognition detection speed.

[0004] In order to solve the above problems, the present application provides a bionic fruit fly intelligent recognition method, which is applied to an intelligent recognition system. The intelligent recognition system includes multiple acquisition units. The method includes:

[0005] Acquire multiple sets of sampling data sets at different sampling time points, wherein the sampling data sets include multiple sampling data collected by the multiple acquisition units;

[0006] Performing response state data recognition on the multiple groups of sample data sets to obtain response state data recognition results;

[0007] Based on the response state data identification result, determining a plurality of steady-state data sets and at least one response state data set from the plurality of sampling data sets;

[0008] performing data recognition processing based on the steady-state data set and the response-state data set to obtain detection target data corresponding to the response-state data set, wherein a data dimension in the detection target data is higher than a data dimension in the corresponding response-state data set, and information in the detection target data is binary information;

[0009] The intelligent recognition model is called to determine the detection target result corresponding to the detection target data.

[0010] On the other hand, the present application also provides an intelligent recognition system, the intelligent recognition system comprising:

[0011] An acquisition module, configured to acquire multiple sets of sampling data sets at different sampling time points, wherein the sampling data sets include multiple sampling data collected by the multiple acquisition units;

[0012] a response state data identification module, configured to perform response state data identification on the plurality of sampling data sets to obtain response state data identification results;

[0013] a classification module, configured to divide the plurality of sampling data sets into a plurality of steady-state data sets and at least one response-state data set based on the response-state data identification result;

[0014] an identification data processing module, configured to perform identification data processing based on the steady-state data set and the response-state data set to obtain detection target data corresponding to the response-state data set, wherein the information capacity of the detection target data is higher than the information capacity of the corresponding response-state data set, and the information in the detection target data is binary information;

[0015] The detection target result determination module is used to call the intelligent recognition model to determine the detection target result corresponding to the detection target data.

[0016] On the other hand, the present application also provides an intelligent identification device, which includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement the intelligent identification method of the bionic fruit fly as mentioned above.

[0017] On the other hand, the present application also provides a computer storage medium, which stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by a processor to implement the above-mentioned intelligent recognition method of bionic fruit flies.

[0018] Due to the above technical solution, the intelligent recognition method of a bionic fruit fly described in this application has the following beneficial effects:

[0019] The intelligent recognition method of the bionic fruit fly of the present application obtains the detection target data corresponding to the response state data set. The data dimension in the detection target data is higher than the data dimension in the corresponding response state data set, and the information in the detection target data is binary information. It not only reduces the recognition calculation amount and data storage space, but also improves the recognition accuracy and recognition detection speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] To more clearly illustrate the technical solution of this application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are merely some embodiments of this application, and those skilled in the art can derive other drawings based on these drawings without inventive effort.

[0021] Figure 1 This is a schematic diagram of an application environment provided by an embodiment of the present application;

[0022] Figure 2 This is a flow chart of an intelligent identification method for a bionic fruit fly provided in an embodiment of the present application;

[0023] Figure 3 This is a schematic diagram showing a display device for a gas identification method provided in an embodiment of the present application;

[0024] Figure 4 This is a schematic diagram of a display device showing a gas identification method provided by another embodiment of the present application;

[0025] Figure 5 This is a structural diagram of an intelligent recognition system provided in an embodiment of the present application.

[0026] Figure 6 This is a hardware structure block diagram of a bionic fruit fly intelligent identification method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0027] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0028] References to "one embodiment" or "embodiment" herein refer to specific features, structures, or characteristics that may be included in at least one implementation of the present application. Throughout the description of this application, it should be understood that the terms "upper," "lower," "left," "right," "top," and "bottom," etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely for ease of description and simplification. They do not indicate or imply that the devices or components referred to must have, be constructed, or operate in a specific orientation, and are therefore not to be construed as limiting the present application. Furthermore, the terms "first" and "second" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly specifying the number of the technical features referred to. Thus, a feature designated "first" or "second" may explicitly or implicitly include one or more of the features. Furthermore, the terms "first," "second," etc., are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential sequence. It should be understood that such terms are interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0029] Artificial intelligence (AI) refers to the theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. AI software technologies primarily encompass computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0030] In recent years, with the research and progress of artificial intelligence technology, artificial intelligence technology has been widely used in many fields. The solutions provided in the embodiments of this application involve artificial intelligence machine learning / deep learning and other technologies, which are specifically described through the following embodiments:

[0031] In conjunction with the instructions Figure 1 , Figure 1 This is a schematic diagram of an application environment provided by an embodiment of the present application, such as Figure 1 As shown, the application environment includes at least a detection perception server 01 and an intelligent recognition server 02.

[0032] In the embodiment of this specification, the detection perception server 01 can be used to perceive the data of the object to be detected and transmit the detected data to the intelligent recognition server. Specifically, the intelligent recognition server 03 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), as well as big data and artificial intelligence platforms.

[0033] In the embodiment of this specification, the intelligent recognition server 02 can be used to classify and identify the detected data. Specifically, the intelligent recognition server 02 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), as well as big data and artificial intelligence platforms.

[0034] In the embodiments of this specification, the above-mentioned detection perception server 01 and the intelligent recognition server 02 can be directly or indirectly connected through wired or wireless communication, and this application again does not impose any restrictions.

[0035] In the embodiments of this specification, in the intelligent recognition process, in order to improve the recognition accuracy of the sampled data set, a deep learning algorithm can be combined to train an intelligent recognition model that can convert the sampled data set into detection target data and recognize the detection target data.

[0036] Specifically, the training data used to train the intelligent recognition model can be collected from a large number of network nodes. Accordingly, the system formed by these network nodes and the equipment used to train the intelligent recognition model can be a distributed system connected through network communication. The distributed system can be a blockchain system.

[0037] In addition, it should be noted that the intelligent recognition model provided based on the embodiment of the present application can provide artificial intelligence cloud services, which are generally also referred to as AIaaS (AIasa Service, Chinese for "AI as a Service"). This is a mainstream service mode of artificial intelligence platforms. Specifically, the AIaaS platform will split several common AI services and provide independent or packaged services in the cloud. This service model is similar to opening an AI theme mall: all developers can access and use one or more artificial intelligence services provided by the platform through the API interface. Some senior developers can also use the AI ​​framework and AI infrastructure provided by the platform to deploy and operate their own exclusive cloud artificial intelligence services.

[0038] Through research, it was found that the dendrites of Kenyon cells (KCs) in the calyx of the Drosophila mushroom body far exceed the number of projection neurons (PNs), and GABA (a non-protein amino acid) can provide global feedback inhibition for neurons, and the anterior contralateral neurons (APL) can also participate in local lateral inhibition; comparing and integrating the Drosophila labeling and classification method with the local sensitive hashing algorithm in computational science can solve the basic computational problems faced by large-scale information retrieval systems, and thus obtain a recognition algorithm that combines artificial intelligence with the bionic Drosophila principle.

[0039] refer to Figure 2 The following describes an intelligent recognition method for a bionic fruit fly provided by an embodiment of the present application, which is applied to an intelligent recognition system. The intelligent recognition system includes multiple acquisition units. The method includes:

[0040] S1. Acquire multiple sets of sampling data sets at different sampling time points, wherein the sampling data sets include multiple sampling data collected by multiple acquisition units.

[0041] In some embodiments, the sampling data may be information data of the target to be detected, such as odor information data of the target to be detected and / or tactile information data of the target to be detected.

[0042] In an embodiment of the present application, adjacent sampling time points may have the same interval time; the sampling data set may be a collection of sampling data acquired by multiple acquisition units at the same time point; and the sampling data may be data collected by a single acquisition unit at any sampling time point.

[0043] In some embodiments, the intelligent recognition system has 24 acquisition units, and a sampling time point is set every 1 second. Then, 5 sets of sampling data sets can be acquired within 5 seconds, and each set of sampling data sets contains sampling data acquired by 24 acquisition units at the same sampling time point.

[0044] In a specific embodiment, when the sampling data is odor information data of the target to be detected, the acquisition unit may be a channel sensor; the channel sensor has a gas collection channel, and when the gas flows through the gas collection channel, the channel sensor will collect the odor information of the gas, the sampling data includes the odor information, and the sampling data set includes the odor information collected by different channel sensors at the same time point; in another specific embodiment, when the sampling data is tactile information data of the target to be detected, the acquisition unit may be a tactile sensor, the sampling data includes tactile information data, and the sampling data set includes tactile information data collected by different tactile sensors at the same time point.

[0045] In the embodiment of the present application, S1 includes:

[0046] S101 , acquiring multiple detection data sets obtained by multiple acquisition units through data collection on targets to be detected.

[0047] In an embodiment of the present application, the detection target data set can be a collection of sampling data acquired by the same acquisition unit at different sampling time points. The detection target data set can be a complete data set or a data set in which the number of sampling data increases in real time as time changes.

[0048] In a specific embodiment, when the sampling data is odor information data of the target to be detected, the multiple sets of detection data sets can be a collection of sampling data collected by multiple channel sensors at different sampling time points. For example, 24 channel sensors can obtain 24 sets of detection data sets; in another specific embodiment, when the sampling data is tactile information data of the target to be detected, the multiple sets of detection data sets are a collection of sampling data collected by multiple tactile sensors at different times. Specifically, 24 tactile sensors can obtain 24 sets of detection data sets.

[0049] S102 , sampling multiple detection data sets at multiple sampling time points to obtain sampling data sets corresponding to each of the multiple sampling time points, wherein a single sampling data set includes sampling data belonging to different detection data sets collected at the same sampling time point.

[0050] In an embodiment of the present application, the sampling data has corresponding acquisition unit label information and corresponding sampling time point information. The sampling data with corresponding acquisition unit label information and corresponding sampling time point information are recombined, and the sampling data with consistent sampling time point information but inconsistent acquisition unit label information are combined to form a sampling data set.

[0051] In one embodiment, 24 acquisition units respectively collect 24 groups of detection data sets within 5 seconds, and each group of detection data sets has sampling data at 5 different sampling time points; each sampling data is separated to obtain corresponding sampling data, and the sampling data has corresponding sampling time point information and corresponding acquisition unit label information; for example, one of the sampling data is the data collected by the first acquisition unit at the first sampling time point, and so on, 120 sampling data with different labels can be obtained; the sampling data are recombined according to the corresponding labels to form a sampling data set; for example, the first sampling data set is the sampling data collected at the first sampling time point, and the first sampling data set has data collected by 24 acquisition units at the first sampling time point; and so on, 5 groups of sampling data sets can be obtained, each of which has sampling data collected by 24 different acquisition units at the same sampling time point.

[0052] In a specific embodiment, when the sampling data is the odor information data of the target to be detected, the sampling data set obtained by the 24 channel sensors at a certain sampling time point is X=[x1x2…x 24 ].

[0053] S2. Perform response state data recognition on multiple groups of sampling data sets to obtain response state data recognition results.

[0054] In the embodiment of the present application, the response state data identification may be a process of dynamically identifying the response state data, and the response state data identification result may be a starting point for determining the response state data.

[0055] In the embodiment of the present application, S2 includes:

[0056] Denoising is performed on multiple sets of sampling data sets to obtain denoised sampling data sets corresponding to each of the multiple sets of sampling data sets. By performing denoising on the sampling data sets, the influence of certain factors on the sampling data can be reduced, thereby improving the accuracy of intelligent recognition. Certain factors may be transmission loss and / or the influence of the acquisition unit itself.

[0057] In the embodiment of the present application, the noise reduction process includes:

[0058] 1) For a single set of detection data, performing de-extreme value weighted averaging processing based on a first number of sampling data to obtain noise reduction sampling data corresponding to one of the first number of sampling data.

[0059] In an embodiment of the present application, a single acquisition unit can acquire a set of detection data sets, and the de-extreme value weighted averaging processing is to exclude the maximum and minimum values ​​of the first number of sampling data, and then perform weighted averaging calculation on the remaining sampling data. The value obtained by the de-extreme value weighted averaging processing is the noise reduction sampling data corresponding to one of the sampling data of the first number of sampling data; through the de-extreme value weighted averaging processing, the excessive fluctuation range of the sampling data caused by certain factors is significantly reduced, thereby significantly improving the accuracy of intelligent recognition.

[0060] In some embodiments, the noise reduction sampling data may be sampling data corresponding to the latest sampling time point in the first number of sampling data, or may be sampling data corresponding to an intermediate sampling time point in the first number of sampling data.

[0061] 2) The denoised sampling data corresponding to different detection data sets at the same sampling time point are used as the denoised sampling data set corresponding to the sampling time point, to obtain the denoised sampling data sets corresponding to the multiple sampling time points, wherein the sampling data set at the same sampling time point corresponds to the denoised sampling data set.

[0062] In an embodiment of the present application, the calculated denoised sampling data also has corresponding acquisition unit label information and corresponding sampling time point information. The denoised sampling data with consistent sampling time point information but inconsistent acquisition unit label information are combined to form a denoised sampling data set.

[0063] In one embodiment, 24 acquisition units respectively collect 24 groups of detection data sets, each of which contains 9 sampling data. In one group of detection data sets, every 7 sampling data are subjected to de-extreme value weighted averaging processing to obtain the noise reduction sampling data corresponding to the sampling data at the latest sampling time point among the 7 sampling data. Therefore, 3 corresponding noise reduction sampling data can be obtained in each group of detection data sets. The noise reduction sampling data with consistent sampling time point information but inconsistent acquisition unit label information are combined to form a noise reduction sampling data set, that is, 3 groups of noise reduction sampling data sets are formed.

[0064] In a specific embodiment, taking the sampled data as the odor information data of the target to be detected as an example, p groups of sampled data sets X1, X2, ..., X q , based on the detection data set of each channel sensor, the following formula is calculated:

[0065]

[0066] in, is the sum of p sampling data collected continuously in a single channel sensor; x i,max is the maximum value of p consecutively collected sample data in a single channel sensor; x i,minis the minimum value of p consecutively collected sample data in a single channel sensor; n is equal to p, which is the number of sample data in a single channel sensor; x′ i is the noise reduction sampling data corresponding to the p-th sampling data in a single channel sensor.

[0067] In an embodiment of the present application, the intelligent recognition system is in communication with the display device, and the noise reduction process includes:

[0068] The noise reduction processed data is transmitted to a display device, and the corresponding noise reduction processed data is displayed on a display diagram on the display device, wherein the horizontal axis on the display diagram is the sampling time point, and different shape marks are used for the noise reduction processed data of different acquisition modules. For example, the mark shape of the first acquisition unit is a triangular point, and the mark shape of the second acquisition unit is a square point.

[0069] In the embodiment of the present application, S2 includes:

[0070] S201 : For a single set of detection data, perform standard deviation processing based on a second number of sampling data to obtain a real-time standard deviation corresponding to each acquisition unit.

[0071] In an embodiment of the present application, a single acquisition unit can acquire a set of detection data sets, and the standard deviation processing is to calculate the standard deviation of the second number of sampling data, and the real-time standard deviation is the standard deviation corresponding to the latest sampling time point in the second number of sampling data.

[0072] In some embodiments, the second number may be equal to the first number, for example, both equal to 7.

[0073] S202: If the number of acquisition units whose real-time standard deviation exceeds a preset standard deviation threshold satisfies a preset number condition, the latest sampling time point corresponding to the second number of sampling data is determined as the peak starting point.

[0074] In the embodiment of the present application, since there are multiple acquisition units, each acquisition unit corresponds to a set of detection data sets. Only when the real-time standard deviation in the detection data set of the acquisition unit exceeds a preset number, can the latest sampling time point corresponding to the second number of sampling data be determined as the peak point. In this way, the influence of the factors of a single acquisition unit itself on the overall detection result is reduced, thereby improving the accuracy of intelligent recognition.

[0075] S3. Based on the response state data identification result, determine multiple groups of steady-state data sets and at least one group of response state data sets in the multiple groups of sampling data sets.

[0076] Based on S202, S3 includes:

[0077] S301: Determine the sampling data at the peak point and after the peak point as response state data.

[0078] In this embodiment of the present application, the sampling data set before the peak onset data is the steady-state data set, and the response-state data set includes the sampling data set at the peak onset and the sampling data set after the peak onset. When the sampling data is acquired in real time, the peak onset is identified, and the next sampling time point has not yet been reached, the multiple sampling data sets include one and only one response-state data set, namely, the sampling data set at the peak onset.

[0079] In the embodiment of the present application, S3 further includes:

[0080] S302: Determine a denoised steady-state data set from the plurality of denoised sampling data sets based on the response state data identification result.

[0081] In an embodiment of the present application, a noise reduction response state data set can be determined based on the noise reduction steady-state data set, and the noise reduction steady-state data set and the noise reduction response state data set are transmitted to a display device. The corresponding noise reduction processing data is displayed on a display graph on the display device to facilitate intuitive monitoring of the change process of the noise reduction sampling data set.

[0082] S4. Perform data recognition processing based on the steady-state data set and the response-state data set to obtain detection target data corresponding to the response-state data set, wherein the data dimension of the detection target data is higher than the data dimension of the corresponding response-state data set, and the information in the detection target data is binary information; by obtaining the detection target data corresponding to the response-state data set, wherein the data dimension in the detection target data is higher than the data dimension in the corresponding response-state data set, and the information in the detection target data is binary information, not only the recognition calculation amount and data storage space are reduced, but also the recognition accuracy and recognition detection speed are improved.

[0083] In an embodiment of the present application, the steady-state data set may be a steady-state data set before any peak point, the response-state data set may be a response-state data set at the peak point and at any sampling time point after the peak point, and the data recognition processing may be a process of performing data conversion calculations based on the steady-state data set and the response-state data set.

[0084] In a specific embodiment, taking the sampled data as the odor information data of the target to be detected as an example, q groups of sampled data sets X1, X2, ..., X1 are continuously collected. q , and calculate the real-time standard deviation of the sampling data collected by each channel sensor. The real-time standard deviation corresponds to the channel sensor in X q For the sampled data within, when the number of real-time standard deviations meets the quantity condition, for example, when there are 24 channel sensors and 18 real-time standard deviations meet the quantity condition, determine X q The sampling time point is the peak point; then the sampling data sets X1, X2, ..., X before the peak pointq-1 is the peak data set, the sampling data set X at and after the peak point q ,…are response state data sets.

[0085] In the embodiment of the present application, S4 includes:

[0086] S401 , performing matrix transformation on the response state data set based on the denoised steady state data set to obtain an input vector matrix corresponding to the response state data set.

[0087] In an embodiment of the present application, matrix conversion can be a process of converting a response state data set into an input vector matrix based on a denoised steady-state data set; the number of columns of the input vector matrix is ​​equal to the number of acquisition units, and the number of rows of the input vector matrix is ​​always 1.

[0088] In a specific embodiment, taking the sampled data as the odor information data of the target to be detected as an example, the sampled data set includes sampled data collected by 24 channel sensors at the same sampling time point. Therefore, a sampled data set can be compared to a 1×24 matrix. The matrix conversion of the response state data set based on the denoised steady-state data set can be performed by running the following formula:

[0089] X in =(X1-X′0) / X′0

[0090] Among them, X1 is the response state data set; X′0 is the denoised steady state data set; X in is the input vector matrix.

[0091] S402. Standardize the input vector matrix based on the response state data set to obtain a standard input matrix. By standardizing the input vector matrix, the interference of abnormal sampling data of a certain acquisition unit can be reduced, thereby improving recognition accuracy.

[0092] In the embodiment of the present application, the normalization processing includes but is not limited to Z-score normalization, maximum-minimum normalization and decimal scaling normalization.

[0093] In a specific embodiment, taking the sampled data as the odor information data of the target to be detected as an example, Z-score normalization is applied, that is, the following formula:

[0094]

[0095] Among them, X in is the input vector matrix; μ is the mean of all elements within the input vector matrix; δ is the standard deviation of all elements within the input vector matrix; is the standard input matrix.

[0096] S403 , randomly generating a sparse projection matrix that meets preset conditions, wherein the number of rows of the sparse projection matrix is ​​equal to the number of columns of the standard input matrix, and the number of columns of the sparse projection matrix is ​​greater than the number of rows of the sparse projection matrix.

[0097] S404. Perform matrix mapping processing on the standard input matrix based on the sparse projection matrix to obtain an input variable matrix.

[0098] In an embodiment of the present application, a sparse projection matrix is ​​generated that meets preset conditions, the number of rows of the matrix is ​​equal to the number of columns of the standard input matrix, and the number of columns of the matrix is ​​greater than the number of rows of the matrix, and the sparse projection matrix and the standard input matrix are subjected to matrix mapping processing, thereby achieving dimensionality increase of the input data, so that the data dimension of the final detection target data is higher than the data dimension of the corresponding response state data set, thereby improving the accuracy of recognition and classification; it should be noted that the data dimension in the matrix is ​​equal to the number of columns of the matrix × 1, for example, the data dimension of a 1×24 matrix is ​​a 24-dimensional matrix.

[0099] In the embodiments of the present application, the data dimension of the target data is detected to be higher than the data dimension of the corresponding response state data set, thereby improving the accuracy of recognition. The principle of this approach is consistent with the principle that the dendrites of Kenyon cells (KCs) in the calyx of the Drosophila mushroom body far exceed the number of projection neurons (PNs), that is, highly accurate intelligent recognition is achieved through bionic Drosophila coding.

[0100] In a specific embodiment, taking the sampled data as the odor information data of the target to be detected as an example, the randomly generated sparse projection matrix is:

[0101]

[0102] Where m is the data dimension of the sparse projection matrix; d is the data dimension of the sparse projection matrix; A binary matrix is ​​a matrix that satisfies the requirement that there are only S 1s in each row, where s<<d and d<<m.

[0103] S405 , performing binary simplification processing on the input variable matrix to obtain detection target data corresponding to the response state data set.

[0104] In the embodiment of the present application, the binary simplification process is to perform binary simplification on the elements in the input variable matrix.

[0105] In the embodiment of the present application, S405 includes:

[0106] S4051. Sort the elements in the input variable matrix to obtain a sorting result.

[0107] S4052. Determine reserved elements and non-reserved elements in the input variable matrix based on the sorting result.

[0108] In an embodiment of the present application, the elements in the input variable matrix may be arranged in descending order, and a certain number of larger elements in the arrangement may be determined as reserved elements, and the remaining elements may be non-reserved elements.

[0109] S4053. Perform binary assignment on at least one of the reserved elements and the non-reserved elements to obtain detection target data.

[0110] In some embodiments, the values ​​of non-reserved elements can be set to 0 or -1; by setting the values ​​of non-reserved elements to 0 or -1, the values ​​of non-reserved elements can be ignored during the recognition process, thereby improving recognition efficiency and reducing data storage space.

[0111] In another embodiment, the value of the reserved element can be set to 1 to obtain the detection target data corresponding to the response state data set; by setting the reserved element to 1, the detection target data in binary form is formed, which improves the recognition efficiency and reduces the data storage space.

[0112] In another embodiment, the values ​​of the reserved elements may be retained to obtain detection target data corresponding to the response state data set; by retaining the values ​​of the reserved elements, the accuracy of intelligent recognition may be improved.

[0113] In summary, the overall encoding function of a specific embodiment of the present application can be obtained:

[0114] h:R d →{0, 1} m

[0115] Among them, R d is the input vector matrix with data dimension d; {0,1} m is a binary matrix with data dimension m; h is the detection target data, specifically:

[0116]

[0117] Where X is the input vector matrix; μ is the mean of all elements in the input vector matrix; δ is the standard deviation of all elements in the input vector matrix; To satisfy the binary matrix with only S 1s in each row, and the matrix data dimension (ie the number of matrix columns) is m; f k A conditional function that only retains elements in reserved positions or sets elements in reserved positions to 1 and elements in non-reserved positions to 0 or -1.

[0118] In an embodiment of the present application, by randomly generating a sparse projection matrix and setting the values ​​of non-reserved elements to 0 or -1, the interference of a single acquisition unit on the detection result is reduced, the importance of the response state is amplified in advance, and the recognition accuracy and recognition detection speed are improved.

[0119] S5. Call the intelligent recognition model to determine the detection target result corresponding to the detection target data.

[0120] In an embodiment of the present application, data classification is performed based on multiple detection target results within a preset time, and the detection target result with the largest number is determined as the final detection target result; by determining the final detection target result within the preset time, the accuracy of intelligent recognition is guaranteed, and the controllability of the duration of intelligent recognition is improved.

[0121] In an embodiment of the present application, the final detection target result is output to a display device.

[0122] In a specific embodiment, taking the sampling data as the odor information data of the target to be detected as an example, one sampling data is obtained every 1 second, and 5 detection target results can be obtained within 5 seconds. The final detection target result is determined based on the number of types of detection target results.

[0123] In an embodiment of the present application, when the number of response state data sets exceeds a preset number range, entering a training recognition mode, the method further includes:

[0124] S601: Use multiple detection target data as input to an intelligent recognition model to perform intelligent recognition classification and obtain intelligent recognition results corresponding to the multiple detection target data.

[0125] S602. Determine the detection target result of the detection target data based on the detection target data and the corresponding intelligent recognition result; wherein the intelligent recognition model includes a model obtained by performing intelligent recognition training on a preset neural network based on the sample detection data and the corresponding recognition result label.

[0126] In a specific embodiment, taking the sampled data as the odor information data of the target to be detected as an example, the intelligent recognition result can be the result of secondary hash recognition; specifically, two hash mappings are performed on multiple detection target data, and after the first hash mapping, candidate objects with the same numerical value as the first 100 digits of the target data to be detected are selected to preliminarily reduce the number of candidate objects, and then a second detailed hash mapping is performed on the remaining candidate objects, and classification recognition is performed based on the result after the secondary hash mapping, and the parameters involved in the encoding function are adjusted according to the result to obtain an optimized intelligent recognition model; correspondingly, in calling the intelligent recognition model, the detection target data and the sample detection data in the intelligent recognition model are hash mapped twice, and the corresponding target detection result is obtained from the intelligent recognition model based on the calculated result.

[0127] In another specific embodiment, taking the sampled data as the tactile information data of the target to be detected as an example, the intelligent recognition result can be the result of nearest neighbor node recognition; specifically, Hamming distance or Euclidean distance calculation is performed on multiple detection target data, and the detection data is classified and identified based on the Hamming distance, and the parameters involved in the encoding function are adjusted according to the result to obtain an optimized intelligent recognition model; correspondingly, in calling the intelligent recognition model, after the Hamming distance or Euclidean distance is calculated between the detection target data and the sample detection data in the intelligent recognition model, the corresponding target detection result is obtained from the intelligent recognition model based on the calculated result.

[0128] The following describes a process of identifying gas types using the bionic fruit fly intelligent identification method of the present application embodiment in a specific embodiment of the present invention:

[0129] The denoised sampling data sets of the 24 channel sensors are displayed on the display device in the form of 6 groups of detection data sets, so four display sub-images can be obtained; the sampling data collected by sensors on different channels in each sub-image are marked with different shapes, for example, the first channel, the seventh channel, the thirteenth channel and the nineteenth channel are marked with small dots; after the peak point is determined, the sampling time point corresponding to the peak point is sent to the display device, and the display device marks the position of the peak sampling time point with a thick line to make it more intuitive to observe the changing time point of the sampling data.

[0130] In a specific gas identification process, please refer to Figure 3, the subgraph of the sampling data is shown in the figure. Data collection is performed every 1 second. Then, based on the response state data within 5 seconds of the peak point, 5 detection target results can be obtained by intelligent recognition. The detection target results obtained are: acetone, acetone, acetone, acetone, and acetone. Therefore, it can be determined that the final detection target result is acetone. And the detection target results after 5 seconds are all acetone, that is, the recognition is correct. Therefore, in this embodiment, fast and accurate gas recognition can be achieved within 5 seconds.

[0131] For another specific gas identification process, please refer to Figure 4 , intelligent recognition based on the response state data within 5s of the peak point can obtain 5 detection target results; the detection target results obtained are: carbon monoxide, methanol, ethanol, ethanol and ethanol, so the final detection target result is determined to be ethanol, and the detection target results after 5s are all ethanol, that is, the identification is correct; in this embodiment, gas recognition makes an incorrect identification at the initial intake stage, identifying it as carbon monoxide and methanol, but it is quickly adjusted to the correct identification result in the subsequent process; therefore, in this embodiment, fast and accurate gas identification can also be achieved within 5s.

[0132] In another specific gas identification process, three rounds of intelligent identification are performed on nine gases of different concentrations in turn, wherein each round of intelligent identification takes 60 to 180 seconds, and the nine gas categories are formaldehyde, ethanol, propane, methanol, methane, carbon monoxide, acetone, hydrogen sulfide and ammonia; and in the intelligent identification process, the corresponding gas results can be accurately identified within 5 seconds of each peak point, and the identification results after 5 seconds do not change; in the 5-second accurate identification process, 135 detection target data can be calculated, and the detection target results corresponding to the detection target data are compared with the actual collected gas categories. It can be concluded that the recognition rate of a single detection target result is 91.9%, and the recognition rate of the final detection target result within 5 seconds of determining each peak point is 100%. Therefore, in this embodiment, the recognition rate of a single detection target result is high, and the recognition rate of the final detection result within 5 seconds of the peak point is extremely high.

[0133] Combine Figure 5 , an intelligent recognition system provided in an embodiment of the present application is introduced, the intelligent recognition system comprising:

[0134] The acquisition module 101 is configured to acquire multiple sets of sampling data at different sampling time points, wherein the sampling data sets include multiple sampling data collected by multiple acquisition units.

[0135] The response state data identification module 201 is used to perform response state data identification on multiple groups of sample data sets to obtain response state data identification results.

[0136] The classification module 301 is configured to determine, based on the response state data identification result, a plurality of steady-state data sets and at least one response state data set among the plurality of sampling data sets.

[0137] The identification data processing module 401 is used to perform data identification processing based on the steady-state data set and the response-state data set to obtain the detection target data corresponding to the response-state data set, wherein the data dimension of the detection target data is higher than the data dimension of the corresponding response-state data set, and the information in the detection target data is binary information.

[0138] The detection target result determination module 501 is used to call the intelligent recognition model to determine the detection target result corresponding to the detection target data.

[0139] In some embodiments, the acquisition module includes:

[0140] The acquisition unit is used to acquire multiple detection data sets obtained by multiple acquisition units through data collection on the target to be detected.

[0141] The sampling data set determination unit is used to sample multiple groups of detection data sets at multiple sampling time points to obtain sampling data sets corresponding to each of the multiple sampling time points, wherein a single sampling data set includes sampling data collected at the same sampling time point and belonging to different detection data sets.

[0142] In some embodiments, the intelligent recognition system further comprises:

[0143] The noise reduction processing module is used to perform noise reduction processing on multiple groups of sampling data sets to obtain noise-reduced sampling data sets corresponding to each of the multiple groups of sampling data sets.

[0144] Correspondingly, the response state data identification module includes:

[0145] The denoising data classification unit is used to determine a denoising steady-state data set from a plurality of denoising sampling data sets based on a response state data identification result.

[0146] In some embodiments, the identification data processing module includes:

[0147] The matrix conversion unit is used to perform matrix conversion on the response state data set based on the denoised steady state data set to obtain an input vector matrix corresponding to the response state data set.

[0148] The standardization processing unit is used to perform standardization processing on the input vector matrix based on the response state data set to obtain a standard input matrix.

[0149] The sparse projection unit is used to randomly generate a sparse projection matrix that meets preset conditions of the sparse projection matrix, wherein the number of rows of the sparse projection matrix is ​​equal to the number of columns of the standard input matrix, and the number of columns of the sparse projection matrix is ​​greater than the number of rows of the sparse projection matrix.

[0150] The matrix mapping unit is used to perform matrix mapping processing on the standard input matrix based on the sparse projection matrix to obtain an input variable matrix.

[0151] The winner-takes-all unit is used to perform binary simplification processing on the input variable matrix to obtain the detection target data corresponding to the response state data set.

[0152] In some embodiments, the winner-takes-all unit includes:

[0153] The sorting subunit is used to sort the elements in the input variable matrix to obtain the sorting result;

[0154] A reserved element determination subunit, configured to determine reserved elements and non-reserved elements in the input variable matrix based on the sorting result;

[0155] The element processing subunit is used to perform binary assignment on at least one of the reserved elements and the non-reserved elements to obtain the detection target data.

[0156] In some embodiments, the noise reduction processing module includes:

[0157] a de-extreme value weighted average processing unit, configured to perform de-extreme value weighted average processing on a single set of detection data sets based on a first number of sample data to obtain noise-reduced sample data corresponding to one of the first number of sample data;

[0158] The correspondence determination unit is used to use the denoised sampling data corresponding to different detection data sets at the same sampling time point as the denoised sampling data set corresponding to the sampling time point, to obtain the denoised sampling data sets corresponding to each of the multiple sampling time points, wherein the sampling data set at the same sampling time point corresponds to the denoised sampling data set.

[0159] In some embodiments, the response state data identification module includes:

[0160] a standard deviation processing unit, configured to perform standard deviation processing on a single set of detection data sets based on a second number of sampled data, to obtain a real-time standard deviation corresponding to each acquisition unit;

[0161] a preset condition judgment unit, configured to determine the latest sampling time point corresponding to the second number of sampled data as the peak starting point if the number of acquisition units whose real-time standard deviation exceeds the preset standard deviation threshold satisfies a preset number condition;

[0162] Correspondingly, the response state data identification module includes:

[0163] The response state data determining unit is configured to determine the sampling data at and after the peak onset point as the response state data.

[0164] In some embodiments, the intelligent recognition system further comprises:

[0165] An intelligent recognition result learning module is used to take multiple detection target data as input of the intelligent recognition model to perform intelligent recognition classification and obtain intelligent recognition results corresponding to the multiple detection target data;

[0166] a detection target result determining unit, configured to determine a detection target result of the detection target data based on the detection target data and the corresponding intelligent recognition result;

[0167] Among them, the intelligent recognition model includes a model obtained by intelligent recognition training of a preset neural network based on sample detection data and corresponding recognition result labels.

[0168] An embodiment of the present application also provides an intelligent recognition device, which includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement the intelligent recognition method of the bionic fruit fly as described above.

[0169] The memory can be used to store software programs and modules. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory. The memory can mainly include a program storage area and a data storage area. The program storage area can store the operating system, application programs required for the functions, etc.; the data storage area can store data created based on the use of the device, etc. In addition, the memory can include high-speed random access memory and non-volatile memory, such as at least one hard disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory can also include a memory controller to provide the processor with access to the memory.

[0170] The method embodiments provided in the embodiments of the present application can be executed in electronic devices such as mobile terminals, computer terminals, servers or similar computing devices. Figure 6 This is a hardware structure block diagram of an electronic device for an image processing method provided in an embodiment of the present application. Figure 6As shown, the electronic device 900 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPUs) 910 (the processor 910 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 930 for storing data, and one or more storage media 920 (such as one or more mass storage devices) for storing application programs 923 or data 922. Among them, the memory 930 and the storage medium 920 can be temporary storage or permanent storage. The program stored in the storage medium 920 may include one or more modules, each module may include a series of instruction operations on the electronic device. Furthermore, the central processing unit 910 can be configured to communicate with the storage medium 920 and execute a series of instruction operations in the storage medium 920 on the electronic device 900. The electronic device 900 may also include one or more power supplies 960, one or more wired or wireless network interfaces 950, one or more input and output interfaces 940, and / or one or more operating systems 921, such as Windows Server TM , Mac OS X TM , Unix TM , LinuxTM, FreeBSDTM, etc.

[0171] The input / output interface 940 can be used to receive or send data via a network. Specific examples of the aforementioned network may include a wireless network provided by a communications provider of the electronic device 900. In one embodiment, the input / output interface 940 includes a network adapter (NIC), which can be connected to other network devices via a base station to communicate with the Internet. In one embodiment, the input / output interface 940 can be a radio frequency (RF) module for wirelessly communicating with the Internet.

[0172] It can be understood by those skilled in the art that Figure 6 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 6 More or fewer components than shown, or with Figure 6 Different configurations shown.

[0173] An embodiment of the present application further provides a storage medium storing at least one instruction or at least one program segment, which is loaded and executed by a processor to implement the above-mentioned intelligent identification method of bionic fruit flies.

[0174] The above description has fully disclosed the specific embodiments of this application. It should be noted that any changes made by those skilled in the art to the specific embodiments of this application do not depart from the scope of the claims of this application. Accordingly, the scope of the claims of this application is not limited to the above specific embodiments.

Claims

1. A bionic fruit fly intelligent recognition method, applied to an intelligent recognition system, wherein the intelligent recognition system includes multiple acquisition units, characterized in that: The method comprises: Acquire multiple sets of sampling data sets at different sampling time points, wherein the sampling data sets include multiple sampling data collected by the multiple acquisition units; Performing response state data identification on the plurality of sample data sets to obtain response state data identification results; the response state data identification results are used to determine the starting point of the response state data; Based on the response state data identification result, determining multiple steady-state data sets and at least one response state data set among the multiple sampling data sets; the sampling data set located before the peak point data is the steady-state data set, and the response state data set includes the sampling data set at the peak point and the sampling data set located after the peak point data; performing data recognition processing based on the steady-state data set and the response-state data set to obtain detection target data corresponding to the response-state data set, wherein the data dimension of the detection target data is higher than the data dimension of the corresponding response-state data set, and the information in the detection target data is binary information; the data recognition processing is a process of performing data conversion calculation based on the steady-state data set and the response-state data set; The intelligent recognition model is called to determine the detection target result corresponding to the detection target data.

2. The intelligent identification method of a bionic fruit fly according to claim 1, characterized in that: The obtaining of multiple sets of sampling data sets at different sampling time points includes: Acquire multiple groups of detection data sets obtained by the multiple acquisition units through data collection of the target to be detected; Data sampling is performed on the multiple detection data sets at multiple sampling time points to obtain sampling data sets corresponding to each of the multiple sampling time points, wherein a single sampling data set includes sampling data collected at the same sampling time point and belonging to different detection data sets.

3. The intelligent identification method of a bionic fruit fly according to claim 2, characterized in that: Before performing response state data identification on the multiple sets of sample data sets to obtain response state data identification results, the method further includes: Performing noise reduction processing on the multiple groups of sample data sets to obtain noise-reduced sample data sets corresponding to each of the multiple groups of sample data sets; The determining, based on the response state data identification result, a plurality of steady-state data sets and at least one response state data set in the plurality of sampling data sets comprises: Based on the response state data identification result, a denoised steady-state data set is determined from the multiple groups of denoised sampling data sets.

4. The intelligent identification method of a bionic fruit fly according to claim 3, characterized in that: The performing identification data processing based on the steady-state data set and the response-state data set to obtain detection target data corresponding to the response-state data set includes: Performing matrix transformation on the response state data set based on the denoised steady-state data set to obtain an input vector matrix corresponding to the response state data set; Performing standardization processing on the input vector matrix based on the response state data set to obtain a standard input matrix; Randomly generating a sparse projection matrix that meets preset conditions, wherein the number of rows of the sparse projection matrix is ​​equal to the number of columns of the standard input matrix, and the number of columns of the sparse projection matrix is ​​greater than the number of rows of the sparse projection matrix; Performing matrix mapping processing on the standard input matrix based on the sparse projection matrix to obtain an input variable matrix; Binary simplification processing is performed on the input variable matrix to obtain detection target data corresponding to the response state data set.

5. The intelligent identification method of a bionic fruit fly according to claim 4, characterized in that: The input variable matrix is ​​subjected to binary simplification processing to obtain the detection target data corresponding to the response state data set, including: Sorting the elements in the input variable matrix to obtain a sorting result; Determining reserved elements and non-reserved elements in the input variable matrix based on the sorting result; Perform binary assignment on at least one of the reserved elements and the non-reserved elements to obtain the detection target data.

6. The intelligent identification method of a bionic fruit fly according to claim 3, characterized in that: The performing denoising on the multiple groups of sampling data sets to obtain multiple groups of denoised sampling data sets corresponding to the multiple groups of sampling data sets includes: For a single set of the detection data set, performing de-extreme value weighted averaging processing based on a first number of sample data to obtain noise reduction sample data corresponding to one of the first number of sample data; The denoised sampling data corresponding to different detection data sets at the same sampling time point are used as the denoised sampling data set corresponding to the sampling time point, to obtain the denoised sampling data sets corresponding to each of the multiple sampling time points, wherein the sampling data sets at the same sampling time point correspond to the denoised sampling data set.

7. The intelligent identification method of a bionic fruit fly according to claim 2, characterized in that: The performing response state data identification on the plurality of sampling data sets to obtain response state data identification results includes: For a single set of the detection data set, performing standard deviation processing based on the second number of sampled data to obtain a real-time standard deviation corresponding to each acquisition unit; If the number of acquisition units whose real-time standard deviation exceeds the preset standard deviation threshold satisfies the quantity condition, the latest sampling time point corresponding to the second number of sampling data is determined as the peak starting point; The determining, based on the response state data identification result, a plurality of steady-state data sets and at least one response state data set in the plurality of sampling data sets comprises: The peak point and sampling data after the peak point are determined as response state data.

8. The intelligent identification method of a bionic fruit fly according to claim 1, characterized in that: In the case that the number of the response state data sets exceeds a preset number range, the method further includes: Using the plurality of detection target data as input of the intelligent recognition model to perform intelligent recognition classification and obtain intelligent recognition results corresponding to the plurality of detection target data; Determining a detection target result of the detection target data based on the detection target data and the corresponding intelligent recognition result; The intelligent recognition model includes a model obtained by performing intelligent recognition training on a preset neural network based on sample detection data and corresponding recognition result labels.

9. An intelligent recognition system, characterized in that: The intelligent recognition system includes: An acquisition module, configured to acquire multiple sets of sampling data sets at different sampling time points, wherein the sampling data sets include multiple sampling data collected by multiple acquisition units; A response state data identification module is used to perform response state data identification on the plurality of sample data sets to obtain response state data identification results; the response state data identification results are used to determine the starting point of the response state data; a classification module, configured to determine, based on the response state data identification result, a plurality of steady-state data sets and at least one response state data set among the plurality of sampling data sets; the sampling data sets located before the peak point data are the steady-state data sets, and the response state data sets include the sampling data sets at the peak point and the sampling data sets located after the peak point data; an identification data processing module, configured to perform data identification processing based on the steady-state data set and the response-state data set to obtain detection target data corresponding to the response-state data set, wherein the data dimension of the detection target data is higher than the data dimension of the corresponding response-state data set, and the information in the detection target data is binary information; the data identification processing is a process of performing data conversion calculation based on the steady-state data set and the response-state data set; The detection target result determination module is used to call the intelligent recognition model to determine the detection target result corresponding to the detection target data.

10. A computer storage medium, characterized in that The storage medium stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement the intelligent identification method of the bionic fruit fly as described in any one of claims 1-8.

Images