Matrix generation method and device, image classification method and device, electronic equipment and medium

By generating and processing random noise signals and image probability matrix, the problem of insufficient true randomness of the random pulse matrix in the image classification model is solved, the performance and accuracy of the image classification model are improved, and the waste of computing resources is reduced.

CN119068203AActive Publication Date: 2024-12-03SHANGHAI ZHONGXING WEIXINZHUANG ARTIFICIAL INTELLIGENCE CHIP CO LTD +1
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202411079488.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-07
Publication Date
2024-12-03
Estimated Expiration
2044-08-07

AI Technical Summary

Technical Problem

In the prior art, the true randomness of the random pulse matrix obtained by directly encoding the pixel values ​​in the image is low, resulting in poor performance of the image classification model, slow processing speed, low accuracy, and wasted computing resources.

Method used

By obtaining the set of electronic noise signals, a random noise signal is generated, and an image probability matrix is ​​generated based on the image pixel information, combining the noise signal and the activation signal, cyclic processing is performed to generate a random pulse matrix, which enhances its true randomness and improves the performance of the image classification model.

Benefits of technology

It improves the true randomness of the random pulse matrix, improves the processing speed and accuracy of the image classification model, and reduces the waste of computing resources.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119068203A_ABST
    Figure CN119068203A_ABST
Patent Text Reader

Abstract

The embodiment of the invention discloses a matrix generation method and device, an image classification method and device, electronic equipment and a medium. A specific embodiment of the method comprises the following steps: acquiring an electronic noise signal set; generating each random noise signal; executing the following first cycle steps: updating the pulse matrix generation times; inputting a preset number of random noise signals in the random noise signals into a noise signal collector; generating an image probability matrix; each activation signal is generated; generating each probability signal group; generating a random pulse sub-matrix; adding the random pulse sub-matrix to a random pulse sub-matrix set; executing the first circulation step again; generating a random pulse matrix; and storing the random pulse matrix and the image category information in a preset database. According to the embodiment, the performance of the image classification model generated by using the random pulse matrix can be effectively improved, so that the processing speed and accuracy of image classification processing by using the model are improved, and computing resources are saved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of computer technology, and in particular to a matrix generation method, an image classification method, a device, an electronic device, and a medium. Background Art

[0002] Image classification is an image processing method that distinguishes different categories of image objects based on different features reflected in the image information. At present, when classifying images, the method usually adopted is: first, the pixel values ​​in the image are encoded to obtain a random pulse matrix. Then, the obtained random pulse matrix is ​​subjected to feature extraction processing, and the extracted features are fused. Then, the classifier is trained using the fused features. Finally, the image is classified using the trained classifier.

[0003] However, when using the above method to classify images, the following technical problems often occur:

[0004] First, the true randomness of the random pulse matrix obtained by directly encoding the pixel values ​​in the image is low, resulting in poor performance of the image classification model generated using the random pulse matrix, which in turn leads to slower processing speed and lower accuracy when using the model for image classification, resulting in a waste of computing resources.

[0005] Second, the random pulse matrix is ​​generated only based on the pixel values ​​in the image, which results in low true randomness of the generated random pulse matrix, which in turn leads to low accuracy and reliability of image classification processing, resulting in a waste of computing resources.

[0006] Third, the random pulse matrix obtained by directly encoding the pixel values ​​in the image has poor ability to capture image information, resulting in loss of image information or repeated capture of image information, which in turn leads to poor performance of the generated image classification model and low accuracy when using the model for image classification, resulting in a waste of computing resources.

[0007] The above information disclosed in this Background section is only for enhancement of understanding of the background of the inventive concept and therefore it may contain information that does not form the prior art that is already known in this country to a person of ordinary skill in the art. Summary of the invention

[0008] The content of this disclosure is used to introduce concepts in a brief form, which will be described in detail in the detailed implementation section below. The content of this disclosure is not intended to identify the key features or essential features of the technical solution claimed for protection, nor is it intended to limit the scope of the technical solution claimed for protection.

[0009] Some embodiments of the present disclosure propose a random pulse matrix generation method for image classification model training, an image classification method, a random pulse matrix generation device for image classification model training, an image classification device, an electronic device and a computer-readable medium to solve one or more of the technical problems mentioned in the above background technology section.

[0010] In a first aspect, some embodiments of the present disclosure provide a method for generating a random pulse matrix for image classification model training, the method comprising: obtaining an electronic noise signal set; generating each random noise signal based on at least one electronic noise signal in the electronic noise signal set; generating an image probability matrix corresponding to the above image pixel information based on pre-stored image pixel information, wherein the above image pixel information corresponds to image category information; based on each random noise signal, executing the following first loop step: updating a preset number of pulse matrix generation times based on a preset value; inputting a preset number of random noise signals in each random noise signal into a noise signal collector to obtain each noise signal corresponding to the above preset number of random noise signals; Based on the above-mentioned image probability matrix, each activation signal is generated; based on the above-mentioned each noise signal and the above-mentioned each activation signal, each probability signal group is generated; based on the above-mentioned each probability signal group, a random pulse sub-matrix is ​​generated; in response to determining that the updated pulse matrix generation number does not meet the preset loop condition, the random pulse sub-matrix is ​​added to the random pulse sub-matrix set to update the random pulse sub-matrix set; in response to determining that the updated pulse matrix generation number meets the preset loop condition, the above-mentioned first loop step is executed again; based on the updated random pulse sub-matrix set, a random pulse matrix is ​​generated; the image category information corresponding to the above-mentioned random pulse matrix and the above-mentioned image pixel information is stored in a preset database for training an image classification model.

[0011] In a second aspect, some embodiments of the present disclosure provide an image classification method, which includes: acquiring an image data set; performing segmentation processing on the foreground target image of each image data in the above-mentioned image data set to obtain a foreground target image set; for each foreground target image in the above-mentioned foreground target image set, performing the following steps: determining each pixel in the above-mentioned foreground target image as a corresponding foreground target image pixel matrix; based on the above-mentioned foreground target image pixel matrix, generating a random pulse matrix corresponding to the above-mentioned foreground target image pixel matrix; inputting the generated random pulse matrix into a pre-trained image classification model to obtain image category information corresponding to the above-mentioned random pulse matrix, wherein the above-mentioned image classification model is trained based on the random pulse matrix stored in a preset database, and the random pulse matrix stored in the preset database is generated by the method described in any implementation method of the above-mentioned first aspect; based on the above-mentioned image category information, determining the category corresponding to the above-mentioned foreground target image.

[0012] In a third aspect, some embodiments of the present disclosure provide a random pulse matrix generation device for image classification model training, the device comprising: an acquisition unit, configured to acquire an electronic noise signal set; a first generation unit, configured to generate each random noise signal based on at least one electronic noise signal in the electronic noise signal set; a second generation unit, configured to generate an image probability matrix corresponding to the above image pixel information based on pre-stored image pixel information, wherein the above image pixel information corresponds to image category information; an execution unit, configured to execute the following first loop steps based on each random noise signal: based on a preset value, updating a preset number of pulse matrix generation times; inputting a preset number of random noise signals in each random noise signal into a noise signal collector to obtain the above preset number of random noise signals. Corresponding noise signals; based on the above-mentioned image probability matrix, generating various activation signals; based on the above-mentioned noise signals and the above-mentioned activation signals, generating various probability signal groups; based on the above-mentioned probability signal groups, generating random pulse sub-matrices; in response to determining that the updated pulse matrix generation times do not meet the preset loop condition, based on adding the above-mentioned random pulse sub-matrix to the random pulse sub-matrix set, to update the random pulse sub-matrix set; in response to determining that the updated pulse matrix generation times meet the preset loop condition, executing the above-mentioned first loop step again; a third generating unit is configured to generate a random pulse matrix based on the updated random pulse sub-matrix set; a storage unit is configured to store the image category information corresponding to the above-mentioned random pulse matrix and the above-mentioned image pixel information in a preset database for training an image classification model.

[0013] In a fourth aspect, some embodiments of the present disclosure provide an image classification device, comprising: an acquisition unit, configured to acquire an image data set; a segmentation unit, configured to perform segmentation processing on the foreground target image of each image data in the above-mentioned image data set to obtain a foreground target image set; an execution unit, configured to perform the following steps for each foreground target image in the above-mentioned foreground target image set: determining each pixel in the above-mentioned foreground target image as a corresponding foreground target image pixel matrix; based on the above-mentioned foreground target image pixel matrix, generating a random pulse matrix corresponding to the above-mentioned foreground target image pixel matrix; inputting the generated random pulse matrix into a pre-trained image classification model to obtain image category information corresponding to the above-mentioned random pulse matrix, wherein the above-mentioned image classification model is trained based on the random pulse matrix stored in a preset database, and the random pulse matrix stored in the preset database is generated by the method described in any implementation method of the above-mentioned first aspect; based on the above-mentioned image category information, determining the category corresponding to the above-mentioned foreground target image.

[0014] In a fifth aspect, some embodiments of the present disclosure provide an electronic device, comprising: one or more processors; a storage device on which one or more programs are stored, and when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation of the first or second aspect above.

[0015] In a sixth aspect, some embodiments of the present disclosure provide a computer-readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the method described in any implementation of the first aspect or the second aspect is implemented.

[0016] The above-mentioned embodiments of the present disclosure have the following beneficial effects: through the random pulse matrix generation method for image classification model training of some embodiments of the present disclosure, the performance of the image classification model generated by the random pulse matrix can be improved, the processing speed when the model is used to process image data can be improved, and the accuracy of image classification can be improved. Specifically, the reason for the poor performance of the image classification model, which in turn leads to a slow processing speed when the model is used to process image data, and the low accuracy of image classification is that the true randomness of the random pulse matrix obtained by directly encoding the pixel values ​​in the image is low, resulting in poor performance of the image classification model generated by the random pulse matrix, which in turn leads to a slow processing speed and low accuracy when the model is used to process image classification, resulting in a waste of computing resources. Based on this, the random pulse matrix generation method for image classification model training of some embodiments of the present disclosure, first, obtains an electronic noise signal set. Thus, a data basis for generating a random noise signal can be obtained. Then, based on at least one electronic noise signal in the electronic noise signal set, each random noise signal is generated. Thus, the complexity and diversity of the signal can be increased by performing a variety of processing on the electronic noise signal, and the true randomness of the random pulse matrix can be enhanced. Then, based on the pre-stored image pixel information, an image probability matrix corresponding to the above image pixel information is generated, wherein the above image pixel information corresponds to image category information. Thus, the image data can be converted into a probability matrix, the pixel information of the image can be better captured, and the true randomness of the generated random pulse matrix can be improved, and the performance of the image classification model generated by the random pulse matrix can be improved. Based on each random noise signal, the following first loop step is performed: first, based on a preset value, the preset pulse matrix generation times are updated. Thus, the number of cycles can be controlled by adjusting the pulse matrix generation times. Then, a preset number of random noise signals in each random noise signal is input into a noise signal collector to obtain each noise signal corresponding to the preset number of random noise signals. Thus, a preset number of noise signals that need to be processed can be obtained from each random noise signal. Then, based on the above image probability matrix, each activation signal is generated. Thus, the image data can be converted into a signal form, and each probability signal group is generated by the activation signal converted from the image and the above noise signal, thereby generating a random pulse sub-matrix. Then, based on the above noise signals and the above activation signals, each probability signal group is generated. Thus, by randomly processing each noise signal and each activation signal, each random probability signal can be obtained, which can effectively improve the true randomness of the subsequently generated random pulse matrix, improve the performance of the image classification model generated using the random pulse matrix, and thus improve the processing speed and accuracy when the model is used for image classification processing. Then, based on the above-mentioned probability signal groups, a random pulse sub-matrix is ​​generated.Thus, the final random pulse matrix can be obtained by randomly combining the obtained random pulse sub-matrices, and the true randomness of the generated random pulse matrix can be improved. Then, in response to determining that the updated pulse matrix generation times do not meet the preset loop condition, the random pulse sub-matrix is ​​added to the random pulse sub-matrix set to update the random pulse sub-matrix set. Thus, the random pulse sub-matrix set can reflect the updated noise signal and activation signal, thereby improving the performance of the image classification model generated by the random pulse matrix, and improving the processing speed and accuracy of the image classification model generated subsequently for the image data. Then, in response to determining that the updated pulse matrix generation times meet the preset loop condition, the first loop step is performed again. Based on the updated random pulse sub-matrix set, a random pulse matrix is ​​generated. Thus, a random pulse matrix with higher true randomness can be obtained, thereby improving the performance of the image classification model generated by subsequent training, improving the processing speed and accuracy during image classification processing, and reducing the waste of computing resources. Finally, the image category information corresponding to the random pulse matrix and the image pixel information is stored in a preset database for training the image classification model. Thus, the model training can be performed by the obtained random pulse matrix with higher true randomness, which can improve the processing speed and accuracy of the image classification model generated by the random pulse matrix, thereby reducing the waste of computing resources. Also because the noise signal used to generate the random pulse matrix is ​​obtained by performing multiple optimization processes on the electronic noise signal, and the activation signal used to generate the random pulse matrix is ​​obtained by performing random processing on the image pixel information, the random pulse matrix generated by the above noise signal and activation signal has a higher true randomness. Also because when generating the random pulse matrix, it is obtained by performing random processing on each noise signal and each activation signal, so the true randomness of the generated random pulse matrix is ​​higher, thereby making the performance of the trained image classification model higher, thereby improving the processing speed when processing image data is slow, and improving the accuracy of image classification. In summary, by processing the image pixel information and the electronic noise signal, the true randomness of the random pulse matrix can be improved, and the performance of the image classification model generated by the random pulse matrix can be improved, thereby improving the processing speed and accuracy when the model is used for image classification processing, thereby reducing the waste of computing resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.

[0018] Figure 1is a flow chart of some embodiments of a random pulse matrix generation method for image classification model training according to the present disclosure;

[0019] Figure 2 is a flow chart of some embodiments of the image classification method according to the present disclosure;

[0020] Figure 3 is a schematic structural diagram of some embodiments of a random pulse matrix generating device for image classification model training according to the present disclosure;

[0021] Figure 4 is a schematic diagram of the structure of some embodiments of the image classification device according to the present disclosure;

[0022] Figure 5 It is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION

[0023] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0024] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure can be combined with each other.

[0025] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0026] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0027] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0028] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0029] Figure 1The flowchart 100 of some embodiments of the random pulse matrix generation method for image classification model training according to the present disclosure is shown. The random pulse matrix generation method for image classification model training includes the following steps:

[0030] Step 101: Acquire an electronic noise signal set.

[0031] In some embodiments, the execution subject (e.g., a computing device) of the random pulse matrix generation method for image classification model training can obtain an electronic noise signal set. The execution subject may be a server that generates a random pulse matrix. The electronic noise signal in the electronic noise signal set may be an analog signal corresponding to the noise generated by the resistor in the circuit. In practice, the execution subject may obtain the electronic noise signal set from a data storage server. The data storage server may be a server for managing and storing electronic noise signals.

[0032] Step 102: Generate respective random noise signals based on at least one electronic noise signal in the electronic noise signal set.

[0033] In some embodiments, the execution subject may generate each random noise signal based on at least one electronic noise signal in the electronic noise signal set, wherein the random noise signal may be a digital signal obtained by processing the electronic noise signal.

[0034] In some optional implementations of some embodiments, the execution subject may generate each random noise signal based on at least one electronic noise signal in the electronic noise signal set by the following steps:

[0035] In the first step, based on the electronic noise signal set, the following second loop steps are performed:

[0036] The first sub-step is to perform current increase processing on the electronic noise signal in the electronic noise signal set to obtain a current noise signal. The current noise signal may be an analog signal obtained by amplifying the current in the electronic noise signal. In practice, first, the execution subject may input the electronic noise signal into a common collector amplifier circuit. Then, the current in the electronic noise signal may be amplified by the common collector amplifier circuit to obtain a current noise signal.

[0037] The second sub-step is to perform voltage reduction processing on the above-mentioned current noise signal to obtain a voltage-reduced noise signal. The voltage-reduced noise signal can be an analog signal obtained after processing the above-mentioned current noise signal. In practice, first, the above-mentioned execution subject can input the current noise signal output by the above-mentioned common-collector amplifier circuit into the diode. Then, the reverse breakdown property of the diode can be utilized to forward conduct the signal corresponding to the voltage lower than the voltage regulation value in the above-mentioned current noise signal, and reverse conduct the signal corresponding to the voltage higher than the voltage regulation value, so as to achieve the purpose of voltage limiting the above-mentioned current noise signal and obtain a voltage-reduced noise signal. The above-mentioned voltage regulation value can be a pre-set value. Here, there is no limitation on the specific setting of the above-mentioned voltage regulation value.

[0038] The third sub-step is to perform signal amplification processing on the above-mentioned buck noise signal to obtain an amplified noise signal. The above-mentioned amplified noise signal can be an analog signal obtained after amplifying the above-mentioned buck noise signal. In practice, first, the above-mentioned execution subject can input the above-mentioned buck noise signal into a transistor amplifier circuit. Then, the amplitude of the above-mentioned buck noise signal can be amplified by the transistor amplifier circuit, so as to achieve the purpose of amplifying the above-mentioned buck noise signal and obtain an amplified noise signal.

[0039] The fourth sub-step is to perform signal conversion processing on the amplified noise signal to obtain a binary noise signal. The binary noise signal may be a digital signal obtained by signal conversion of the amplified noise signal. In practice, the execution subject may convert the amplified noise signal into the binary noise signal through a signal converter. The signal converter may be various electronic components capable of converting analog signals into digital signals. For example, the signal converter may be an analog-to-digital converter.

[0040] The fifth sub-step is to perform digital extraction processing on the above binary noise signal to obtain a binary digital number. The above binary digital number can be a digital number in the above binary noise signal. For example, the above binary digital number can be 0 or 1. In practice, the above execution subject can perform digital extraction on the last bit of each binary signal information through the bit operation instruction in the program language to obtain a binary signal. For example, when the above binary noise signal is '10010010', the extracted binary signal is the 0 at the last bit of the binary noise signal. When the above binary noise signal is '10001111', the extracted binary signal is the 1 at the last bit of the binary noise signal.

[0041] The sixth sub-step is to add the above binary digital numbers to the binary digital sequence to update the binary digital sequence. In practice, the above execution subject can switch back and forth between the functions of storing data and sending data in the register through a clock signal of a preset frequency. Thus, the extracted binary digital numbers can be stored through the register and the binary digital numbers can be sent to the binary digital sequence. The above preset frequency can be a pre-set value. Here, the specific setting of the above preset frequency is not limited.

[0042] The seventh sub-step is, in response to determining that the updated binary digital sequence satisfies the preset sequence condition, determining the updated binary digital sequence as a random noise signal, and clearing the updated binary signal sequence. The preset sequence condition may be that the number of binary digits in the updated binary digital sequence is equal to the preset number of digits. The preset number of digits may be a pre-set value. Here, the specific setting of the preset number of digits is not limited.

[0043] In an eighth sub-step, in response to determining that the number of executions of the second loop step does not meet the preset number condition, an unused electronic noise signal is used to form an electronic noise signal set, and the second loop step is executed again. The preset number condition may be that the number of executions of the second loop step is greater than or equal to the preset number. The preset number may be a pre-set value. Here, the specific setting of the preset number is not limited.

[0044] Step 103: Generate an image probability matrix corresponding to the image pixel information based on the pre-stored image pixel information.

[0045] In some embodiments, the execution subject may generate an image probability matrix corresponding to the image pixel information based on pre-stored image pixel information. The image pixel information may correspond to image category information. The image pixel information may be information corresponding to each pixel in the image. The image pixel information may include, but is not limited to: image size, each pixel in the image, pixel values ​​corresponding to each pixel in the image, and coordinates of each pixel in the image. The length and width corresponding to the image size may be obtained in pixels. The coordinates of each pixel in the image may be obtained by taking the first pixel in the upper left corner of the image as the origin of the coordinates, the horizontal right direction of the image as the horizontal axis, and the vertical downward direction of the image as the vertical axis. The image probability matrix may be a matrix obtained by normalizing the image pixel matrix corresponding to the image pixel information. The image pixel matrix may be a matrix corresponding to the pixel points and pixel values ​​in the image pixel information. The image category information may be a label used to characterize the category of the image. For example, the image category information may be "fruit".

[0046] In some optional implementations of some embodiments, the execution subject may generate an image probability matrix corresponding to the image pixel information based on pre-stored image pixel information through the following steps:

[0047] The first step is to generate an image pixel matrix based on the pre-stored image pixel information. In practice, first, the execution subject can determine the number of rows and columns corresponding to the image pixel matrix through the image size included in the image pixel information. For example, when the image size is 20*30, the image pixel matrix can be a matrix of 30 rows and 20 columns. Then, for each pixel point in each pixel point included in the image pixel information, the execution subject can determine the position of the pixel point in the matrix through the coordinates corresponding to the pixel point. Then, the pixel value corresponding to the pixel point can be determined as the element value of the corresponding position. Finally, the obtained matrix can be determined as the image pixel matrix. As an example: when the coordinates of a pixel point are (0, 1) and the corresponding pixel value is 255, the corresponding element value at the position of the first column and the second row in the matrix is ​​255. When the coordinates of a pixel point are (3, 6) and the corresponding pixel value is 100, the corresponding element value at the position of the fourth column and the seventh row in the matrix is ​​100.

[0048] In the second step, based on the preset normalization value, each matrix element in the above-mentioned image pixel matrix is ​​normalized to obtain an image probability matrix. The above-mentioned preset normalization value can be a preset value. Here, there is no limitation on the specific setting of the above-mentioned preset normalization value. For example, the above-mentioned preset normalization value can be 255. The above-mentioned image probability matrix can be a matrix obtained by processing each element in the above-mentioned image pixel matrix with the same size as the above-mentioned image pixel matrix. In practice, the above-mentioned execution subject can divide the value corresponding to each element in the above-mentioned image pixel matrix by the preset normalization value to obtain the image probability matrix. For example, when the value corresponding to the matrix element in the 3rd row and 5th column of the above-mentioned image probability matrix is ​​204, and the above-mentioned preset normalization value is 255, 204 is divided by 255, and the obtained 0.8 is the matrix element corresponding to the 3rd row and 5th column of the above-mentioned image probability matrix.

[0049] Step 104, based on each random noise signal, execute the following first cycle steps:

[0050] Step 1041, based on a preset value, updating the preset pulse matrix generation times.

[0051] In some embodiments, the execution subject may update the preset pulse matrix generation times based on a preset value. The preset value may be a pre-set value. Here, there is no limitation on the specific setting of the preset value. The preset pulse matrix generation times may be a pre-set value for characterizing the number of cycles. Here, there is no limitation on the specific setting of the preset pulse matrix generation times. In practice, the execution subject may determine the sum of the pulse matrix generation times and the preset value as the updated pulse matrix generation times. For example, when the pulse matrix generation times is 2 and the preset value is 1, 2 plus 1 may be used to obtain 3 as the updated pulse matrix generation times.

[0052] Step 1042: input a preset number of random noise signals from each random noise signal into a noise signal collector to obtain each noise signal corresponding to the preset number of random noise signals.

[0053] In some embodiments, the execution subject may input a preset number of random noise signals from each random noise signal into a noise signal collector to obtain each noise signal corresponding to the preset number of random noise signals. The preset number may be a pre-set value. Here, there is no limitation on the specific setting of the preset number. The noise signal collector may be an electronic device for receiving noise signals. For example, the noise signal collector may be a microcontroller. The noise signal may be a random noise signal input into the noise signal collector.

[0054] Step 1043, generating various activation signals based on the image probability matrix.

[0055] In some embodiments, the execution entity may generate various activation signals based on the image probability matrix, wherein the activation signals may be binary data obtained through the image probability matrix.

[0056] In some optional implementations of some embodiments, the execution subject may generate various activation signals based on the image probability matrix through the following steps:

[0057] The first step is to determine a preset selection number of probability matrix elements in the above-mentioned image probability matrix that meet the preset selection conditions as each target probability matrix element. The above-mentioned preset selection condition may be that the probability matrix element has not been selected. The above-mentioned preset number may be a preset value. Here, there is no limitation on the specific setting of the above-mentioned preset number. The above-mentioned preset selection number may be a preset value. Here, there is no limitation on the specific setting of the above-mentioned preset selection number. The above-mentioned probability matrix element may be a matrix element in the above-mentioned image probability matrix.

[0058] In the second step, for each target probability matrix element in the above target probability matrix elements, perform the following steps:

[0059] In the first sub-step, the product of the target probability matrix element and the preset activation probability value is determined as the activation signal probability value. The preset activation probability value may be a pre-set value. Here, the specific setting of the preset activation probability value is not limited.

[0060] The second sub-step is to perform signal conversion processing on the activation probability value to obtain an activation signal. In practice, the execution subject may convert the activation probability value into binary data, and use the converted binary data as the activation signal.

[0061] The above technical solution and its related contents, as an inventive point of an embodiment of the present disclosure, solve the technical problem three mentioned in the background technology, "the random pulse matrix obtained by directly encoding the pixel values ​​in the image has a poor ability to capture image information, resulting in loss of image information or repeated capture of image information, which in turn leads to poor performance of the generated image classification model, low accuracy when using the model for image classification, and waste of computing resources". The factors that lead to low accuracy of image classification processing and waste of computing resources are often as follows: the random pulse matrix obtained by directly encoding the pixel values ​​in the image has a poor ability to capture image information, resulting in loss of image information or repeated capture of image information, which in turn leads to poor performance of the generated image classification model, low accuracy when using the model for image classification, and waste of computing resources. If the above factors are solved, the performance of the model can be improved, the accuracy of image classification using the model can be improved, and the waste of computing resources can be reduced. In order to achieve this effect, the present disclosure determines the preset selection number of probability matrix elements that meet the preset selection conditions in the above image probability matrix as each target probability matrix element. As a result, probability matrix elements that have not been selected can be selected in the image probability matrix, repeated selection of probability matrix elements can be avoided, and the problem of repeated acquisition of image information can be avoided. Then, for each target probability matrix element in the above-mentioned target probability matrix elements, the following steps are performed: First, the product of the above-mentioned target probability matrix element and the preset activation probability value is determined as the activation signal probability value. Thus, the strength of the activation signal generated subsequently can be controlled by introducing the preset activation probability value, and the effectiveness of the activation signal generated subsequently can be improved. The above-mentioned activation probability value is subjected to signal conversion processing to obtain an activation signal. Thus, the corresponding activation signal can be generated by the probability matrix element in the image probability matrix. Also, because the probability matrix element that has not been selected can be selected first, and then the selected probability matrix element is processed, the problem of repeated acquisition of image information can be avoided, and the performance of the generated image classification model can be improved. Also, because the above steps can be executed cyclically until all the probability matrix elements are processed and then the random pulse matrix is ​​generated, the problem of image information loss can be avoided, the performance of the generated image classification model can be improved, the accuracy of image classification using the model can be improved, and the waste of computing resources can be reduced.

[0062] Step 1044: Generate respective probability signal groups based on respective noise signals and respective activation signals.

[0063] In some embodiments, the execution subject may generate each probability signal group based on each noise signal and each activation signal. The probability signal in each probability signal group may be a level signal representing a high or low level in a circuit. The probability signal may be a high level signal or a low level signal.

[0064] In some optional implementations of some embodiments, the execution subject may generate each probability signal group based on each noise signal and each activation signal through the following steps:

[0065] The first step is to perform bit separation processing on each of the above-mentioned noise signals to obtain each bit separation signal. The above-mentioned bit separation signal can be a digital signal obtained by performing bit separation processing on the above-mentioned noise signal. In practice, the above-mentioned execution subject can perform bit separation processing on the high and low bits of the above-mentioned noise signal through a splitter to obtain each bit separation signal. For example, when a noise signal is '1010110111010101', the above-mentioned execution subject can perform bit separation processing on the high eight bits and the low eight bits of the noise signal through a splitter, and the obtained '10101101' and '11010101' are two bit separation signals.

[0066] In the second step, for each activation signal in each activation signal, perform the following steps:

[0067] In the first sub-step, the activation signal is subjected to bit separation processing to obtain the high-order digit of the activation signal and the tail of the activation signal. The high-order digit of the activation signal may be a digital number in the activation signal. The tail of the activation signal may be binary data obtained after the activation signal is subjected to bit separation processing. In practice, the execution subject may perform bit separation processing on the activation signal through a wire splitter to obtain the high-order digit of the activation signal and the tail of the activation signal. For example, when an activation signal is '111010101', first, the execution subject may perform bit separation processing on the high-order digit and the low-order digit through a wire splitter. Then, the '1' obtained after separation may be determined as the high-order digit of the activation signal. The '11010101' obtained after separation may be determined as the tail of the activation signal.

[0068] In the second sub-step, for every two bit-separated signals in each bit-separated signal, the following steps are performed:

[0069] Sub-step 1, determining the sum of the above two bit-separated signals as the summed noise signal. In practice, the above-mentioned execution subject can perform sum processing on the above two bit-separated signals by using binary addition through an arithmetic logic unit. Then, the obtained sum can be determined as the summed noise signal. As an example: when one bit-separated signal is "10101101" and the other bit-separated signal is "11010101", the "110000010" obtained by binary addition is the above-mentioned summed noise signal.

[0070] Sub-step 2, performing bit separation processing on the above-mentioned summed noise signal to obtain the noise signal carry number and the noise signal mantissa corresponding to the above-mentioned summed noise signal. Among them, the noise signal carry number can be a binary signal obtained after the above-mentioned summed noise signal is subjected to bit separation processing. The above-mentioned noise signal mantissa can be a binary signal obtained after the above-mentioned summed noise signal is subjected to bit separation processing. In practice, the above-mentioned execution subject can perform bit separation processing on the above-mentioned summed noise signal through a wire splitter to obtain the noise signal carry number and the noise signal mantissa. For example, when the above-mentioned summed noise signal is '110010101', the above-mentioned execution subject can determine the separated high-bit digit '1' as the noise signal carry number through the wire splitter. Then, the separated '10010101' can be determined as the noise signal mantissa.

[0071] Sub-step three, compare the carry number of the noise signal and the high digit number of the activation signal to obtain a first comparison result. The first comparison result may be label information used to characterize the size relationship between the carry number of the noise signal and the high digit number of the activation signal. For example, the first comparison result may be "the carry number of the noise signal is greater than the high digit number of the activation signal".

[0072] Sub-step 4: Based on the first comparison result, a level signal corresponding to the first comparison result is generated as a first level signal. The first level signal may be a level signal used to characterize the first comparison result. For example, when the first comparison result is "the number of noise signal carries is greater than the number of high bits of the activation signal" or "the number of noise signal carries is equal to the number of high bits of the activation signal", the output first level signal is a low level signal. When the first comparison result is "the number of noise signal carries is less than the number of high bits of the activation signal", the output first level signal is a high level signal.

[0073] Sub-step 5: Compare the mantissa of the noise signal and the mantissa of the activation signal to obtain a second comparison result. The second comparison result may be label information used to characterize the magnitude relationship between the mantissa of the noise signal and the mantissa of the activation signal. For example, the second comparison result may be "the mantissa of the noise signal is greater than the mantissa of the activation signal".

[0074] Sub-step six, based on the second comparison result, generating a level signal corresponding to the second comparison result as a second level signal. The second level signal may be a level signal used to characterize the second comparison result. For example, when the second comparison result is "the mantissa of the noise signal is greater than the mantissa of the activation signal", a high level signal is output. When the second comparison result is "the mantissa of the noise signal is less than the mantissa of the activation signal" or "the mantissa of the noise signal is equal to the mantissa of the activation signal", a low level signal is output.

[0075] Sub-step seven, based on the first level signal and the second level signal, generate a probability signal. The probability signal may be a level signal used to characterize the logical relationship between the first level signal and the second level signal. In practice, the execution subject may perform an AND operation on the first level signal and the second level signal to obtain the probability signal.

[0076] The third sub-step is to determine the generated probability signals into a probability signal group.

[0077] The above technical solution and its related contents, as an inventive point of the embodiment of the present disclosure, solve the technical problem 2 mentioned in the background technology, "generating a random pulse matrix only according to the pixel values ​​in the image, resulting in low true randomness of the generated random pulse matrix, which in turn leads to low accuracy and reliability of image classification processing, resulting in waste of computing resources". The factors that lead to low accuracy and reliability of image classification processing and waste of computing resources are often as follows: generating a random pulse matrix only according to the pixel values ​​in the image, resulting in low true randomness of the generated random pulse matrix, which in turn leads to low accuracy and reliability of image classification processing, resulting in waste of computing resources. If the above factors are solved, the generalization ability of the model can be improved and the accuracy of model prediction can be improved. In order to achieve this effect, the present disclosure performs the following steps for each sample user-item interaction feature information in each sample user-item interaction feature information: First, each of the above noise signals is subjected to bit separation processing to obtain each bit separation signal. Therefore, by introducing a noise signal with high true randomness and performing bit separation processing on the noise signal, the true randomness of the subsequently generated probability signal can be improved, and then the true randomness of the random pulse matrix generated by the probability signal can be improved, and the accuracy and reliability of the image classification processing can be improved. Then, for each activation signal in each activation signal, the following steps are performed: first, the above-mentioned activation signal is subjected to bit separation processing to obtain the activation signal high-order digit and the activation signal tail number. Thus, the activation signal high-order digit and the activation signal tail number can be obtained, which can provide a data basis for subsequent processing. Then, for each two bit separation signals in each bit separation signal, the following steps are performed: first, the sum of the above-mentioned two bit separation signals is determined as the sum noise signal. Thus, the diversity and complexity of the noise signal can be improved, so that the generated activation signal has a higher true randomness. Then, the above-mentioned sum noise signal is subjected to bit separation processing to obtain the noise signal carry number and the noise signal tail number corresponding to the above-mentioned sum noise signal. Thus, the noise signal carry number and the noise signal tail number can be obtained, which can provide a data basis for subsequent processing. Then, the above-mentioned noise signal carry number and the above-mentioned activation signal high-order digit are compared and processed to obtain a first comparison result. Thus, by comparing the above-mentioned noise signal carry number and the above-mentioned activation signal high-order digit, the signal can be better controlled, the true randomness of the generated activation signal can be improved, and the true randomness of the random pulse matrix can be improved. Then, based on the first comparison result, a level signal corresponding to the first comparison result is generated as a first level signal. Thus, a level signal corresponding to the first comparison result can be obtained. Then, the mantissa of the noise signal and the mantissa of the activation signal are compared and processed to obtain a second comparison result. Thus, by comparing the mantissa of the noise signal and the mantissa of the activation signal, the subtle differences of the signal can be better captured, and the true randomness of the random pulse matrix can be further improved.Then, based on the second comparison result, a level signal corresponding to the second comparison result is generated as a second level signal. Thus, a level signal corresponding to the second comparison result can be obtained. Then, based on the first level signal and the second level signal, a probability signal is generated. Thus, a probability signal with higher true randomness can be obtained through the first level signal and the second level signal, and then a random pulse matrix with higher true randomness can be obtained, thereby improving the accuracy and reliability of image classification processing and reducing the waste of computing resources. Finally, each generated probability signal is determined as a probability signal group. Also because the noise signal is obtained by continuously optimizing the analog signal corresponding to the noise generated by the resistor in the circuit, the noise signal has a higher true randomness, which can improve the true randomness of the generated random pulse matrix, thereby effectively improving the accuracy and reliability of the image classification processing. Also because each noise signal and each activation signal are processed in a variety of ways to obtain each probability signal group, and the random pulse matrix is ​​obtained through each probability signal group, therefore, the obtained random pulse matrix has a higher true randomness, which can effectively improve the accuracy and reliability of the image classification processing and save computing resources.

[0078] Step 1045: Generate a random pulse sub-matrix based on each probability signal group.

[0079] In some embodiments, the execution subject may generate a random pulse sub-matrix based on the above-mentioned probability signal groups. The random pulse sub-matrix may be a three-dimensional matrix obtained by the above-mentioned probability signal groups. In practice, first, the execution subject may determine any one of the above-mentioned probability signal groups as the target probability signal group. Then, each probability signal group other than the above-mentioned target probability signal group may be determined as each probability signal group to be compared. Then, for each target probability signal in the above-mentioned target probability signal group, the above-mentioned target probability signal may be determined as a probability signal to be added. Then, for each probability signal group to be compared in the above-mentioned probability signal groups to be compared, the probability signal to be compared corresponding to the above-mentioned probability signal to be added in the above-mentioned probability signal group may be determined as a probability signal to be combined. Then, the above-mentioned probability signal to be added and the determined probability signals to be combined may be randomly combined into a matrix with a corresponding number of rows as one row and a corresponding number of columns as the sum of the number of the above-mentioned probability signals to be combined and the above-mentioned probability signals to be added as a signal matrix. Then, for any one of the obtained signal matrices, the execution subject can determine the direction perpendicular to the rows and columns of the signal matrix as the superposition direction through the library function. Then, the execution subject can superimpose another signal matrix onto the signal matrix along the superposition direction through the library function to obtain a two-layer signal matrix. Then, the execution subject can superimpose the third signal matrix onto the two-layer signal matrix along the superposition direction to obtain a three-layer signal matrix. And so on, until the superposition of each signal matrix is ​​completed, the obtained matrix is ​​determined as a random pulse submatrix. Among them, the library function can be the np.stack function.

[0080] Step 1046 , in response to determining that the updated pulse matrix generation times do not satisfy a preset loop condition, the random pulse sub-matrix set is updated based on adding the random pulse sub-matrix to the random pulse sub-matrix set.

[0081] In some embodiments, the execution subject may update the random pulse submatrix set based on adding the random pulse submatrix to the random pulse submatrix set in response to determining that the updated pulse matrix generation number does not meet the preset cycle condition. The preset cycle condition may be that the updated pulse matrix generation number is less than or equal to the preset cycle number. The preset cycle number may be a pre-set value. Here, the specific setting of the preset cycle number is not limited.

[0082] Step 1047, in response to determining that the updated pulse matrix generation times meets the preset loop condition, the first loop step is executed again.

[0083] In some embodiments, the execution subject may execute the first loop step again in response to determining that the updated pulse matrix generation times meets a preset loop condition.

[0084] Step 105 : generating a random pulse matrix based on the updated random pulse sub-matrix set.

[0085] In some embodiments, the execution subject may generate a random pulse matrix based on the updated random pulse submatrix set. The random pulse matrix may be a three-dimensional matrix obtained by processing each random pulse submatrix in the random pulse submatrix set. In practice, for each random pulse submatrix, the execution subject may determine any one of the random pulse submatrices as the initial random pulse submatrix. Then, the execution subject may determine the horizontal rightward direction corresponding to each row of elements in the initial random submatrix as the first combination direction. Then, the execution subject may randomly select a random pulse submatrix from the remaining random pulse submatrices as the second random pulse submatrix. Then, the second random pulse submatrix may be spliced ​​to the initial random submatrix along the first combination direction through a library function. Then, the random pulse submatrix that has not been spliced ​​may continue to be spliced ​​to the second random pulse submatrix along the first combination direction. By analogy, until the number of columns corresponding to the matrix obtained after splicing is equal to the preset number of columns, the vertical downward direction corresponding to each row of elements in the initial random submatrix may be determined as the second combination direction through the library function. Then, a random pulse submatrix can be taken out from each random pulse submatrix that has not been spliced ​​as the third random pulse submatrix. Then, the third random pulse submatrix can be spliced ​​along the second combination direction. Then, the horizontal right direction corresponding to each row of elements in the third random pulse submatrix can be determined as the third combination direction. Then, each random pulse submatrix that has not been spliced ​​can be spliced ​​along the third combination direction. When the number of columns corresponding to the spliced ​​matrix is ​​equal to the preset number of columns again, the vertical downward direction corresponding to each row of elements in the third random pulse submatrix can be determined as the fourth combination direction. Then, a random pulse submatrix that has not been spliced ​​can be determined as the fourth random pulse submatrix. Then, the fourth random pulse submatrix can be spliced ​​to the third random pulse submatrix along the fourth combination direction through the library function. Then, the horizontal right direction corresponding to each row of elements in the fourth random pulse submatrix can be determined as the fifth combination direction. Then, each random pulse submatrix that has not been spliced ​​can be spliced ​​along the fifth combination direction. And so on, until the last random pulse submatrix is ​​spliced. Finally, the matrix obtained after the splicing is completed is determined as a random pulse matrix. The above-mentioned preset number of columns can be a preset value. Here, the specific setting of the above-mentioned preset number of columns is not limited. The above-mentioned library function can be the np.stack function.

[0086] Step 106, storing the image category information corresponding to the random pulse matrix and the image pixel information into a preset database for use in training an image classification model.

[0087] In some embodiments, the execution subject may store the random pulse matrix in a preset database for training an image classification model. The preset database may be a pre-set database for storing random pulse matrices. The image classification model may be various neural networks that take the random pulse matrix corresponding to the image as input and the feature vector corresponding to the image as output. For example, the neural network may be a pulse neural network.

[0088] In some optional implementations of some embodiments, the image classification model may be obtained by training the execution subject through the following steps:

[0089] The first step is to obtain a sample set from the preset database, wherein the samples in the sample set include images, sample random pulse matrices corresponding to image pixel information in the images, and sample image category information. The sample random pulse matrix may be a random pulse matrix corresponding to the image in the sample. The sample image category information may be image category information corresponding to the image in the sample.

[0090] In the second step, the following training steps are performed based on the sample set:

[0091] In the first sub-step, the sample random pulse matrices corresponding to the image of at least one sample in the sample set are respectively input into the initial neural network to obtain the image category information corresponding to each sample in the at least one sample. The initial neural network includes a feature extraction network and an output layer. The feature extraction network can be various neural networks that can extract features from the random pulse matrix. For example, the feature extraction network can be a pulse neural network. The output layer can be an activation function that can map the extracted features to the probability that the image belongs to each image label information. For example, the activation function can be a normalized exponential function (softmax).

[0092] The second sub-step is to compare the image category information corresponding to each sample in the at least one sample with the corresponding sample image category information. In practice, first, the execution subject may generate the probability distribution of the image category information corresponding to each sample in the at least one sample by a normalized exponential function. Then, the execution subject may generate the probability distribution of the sample image category information corresponding to each sample in the at least one sample by a normalized exponential function. Finally, the execution subject may generate the loss value between the probability distribution of the image category information corresponding to each sample in the at least one sample and the probability distribution of the corresponding sample image category information by a cross entropy loss function.

[0093] The third sub-step is to determine whether the initial neural network has achieved the preset optimization goal based on the comparison result. As an example, when the loss value between the image category information corresponding to a sample and the corresponding sample image category information is less than the preset loss threshold, the above sample is determined as a positive sample. Among them, the above positive sample can be used to characterize samples that meet the loss value condition. The above loss value condition can be that the loss value between the image category information corresponding to a sample and the corresponding sample image category information is less than the preset loss threshold. The above optimization goal can be that the proportion of positive samples in all samples is greater than the preset accuracy threshold. The above loss threshold can be a preset value. Here, there is no limitation on the specific setting of the above loss threshold. The above accuracy threshold can be a preset value. Here, there is no limitation on the specific setting of the above accuracy threshold.

[0094] In a fourth sub-step, in response to determining that the initial neural network achieves the above-mentioned optimization goal, a feature extraction network included in the initial neural network is determined as an image classification model.

[0095] The fifth sub-step is, in response to determining that the initial neural network does not achieve the above optimization goal, adjusting the network parameters of the initial neural network, using unused samples to form a sample set, using the adjusted initial neural network as the initial neural network, and performing the above training step again. In practice, the above execution subject can use the back propagation algorithm (BackPropagation Algorithm, BP algorithm) and the gradient descent method (such as the mini-batch gradient descent algorithm) to adjust the network parameters of the above initial neural network.

[0096] The above-mentioned embodiments of the present disclosure have the following beneficial effects: through the random pulse matrix generation method for image classification model training of some embodiments of the present disclosure, the performance of the image classification model generated by the random pulse matrix can be improved, the processing speed when the model is used to process image data can be improved, and the accuracy of image classification can be improved. Specifically, the reason for the poor performance of the image classification model, which in turn leads to a slow processing speed when the model is used to process image data, and the low accuracy of image classification is that the true randomness of the random pulse matrix obtained by directly encoding the pixel values ​​in the image is low, resulting in poor performance of the image classification model generated by the random pulse matrix, which in turn leads to a slow processing speed and low accuracy when the model is used to process image classification, resulting in a waste of computing resources. Based on this, the random pulse matrix generation method for image classification model training of some embodiments of the present disclosure, first, obtains an electronic noise signal set. Thus, a data basis for generating a random noise signal can be obtained. Then, based on at least one electronic noise signal in the electronic noise signal set, each random noise signal is generated. Thus, the complexity and diversity of the signal can be increased by performing a variety of processing on the electronic noise signal, and the true randomness of the random pulse matrix can be enhanced. Then, based on the pre-stored image pixel information, an image probability matrix corresponding to the above image pixel information is generated, wherein the above image pixel information corresponds to image category information. Thus, the image data can be converted into a probability matrix, the pixel information of the image can be better captured, and the true randomness of the generated random pulse matrix can be improved, and the performance of the image classification model generated by the random pulse matrix can be improved. Based on each random noise signal, the following first loop step is performed: first, based on a preset value, the preset pulse matrix generation times are updated. Thus, the number of cycles can be controlled by adjusting the pulse matrix generation times. Then, a preset number of random noise signals in each random noise signal is input into a noise signal collector to obtain each noise signal corresponding to the preset number of random noise signals. Thus, a preset number of noise signals that need to be processed can be obtained from each random noise signal. Then, based on the above image probability matrix, each activation signal is generated. Thus, the image data can be converted into a signal form, and each probability signal group is generated by the activation signal converted from the image and the above noise signal, thereby generating a random pulse sub-matrix. Then, based on the above noise signals and the above activation signals, each probability signal group is generated. Thus, by randomly processing each noise signal and each activation signal, each random probability signal can be obtained, which can effectively improve the true randomness of the subsequently generated random pulse matrix, improve the performance of the image classification model generated using the random pulse matrix, and thus improve the processing speed and accuracy when the model is used for image classification processing. Then, based on the above-mentioned probability signal groups, a random pulse sub-matrix is ​​generated.Thus, the final random pulse matrix can be obtained by randomly combining the obtained random pulse sub-matrices, and the true randomness of the generated random pulse matrix can be improved. Then, in response to determining that the updated pulse matrix generation times do not meet the preset loop condition, the random pulse sub-matrix is ​​added to the random pulse sub-matrix set to update the random pulse sub-matrix set. Thus, the random pulse sub-matrix set can reflect the updated noise signal and activation signal, thereby improving the performance of the image classification model generated by the random pulse matrix, and improving the processing speed and accuracy of the image classification model generated subsequently for the image data. Then, in response to determining that the updated pulse matrix generation times meet the preset loop condition, the first loop step is performed again. Based on the updated random pulse sub-matrix set, a random pulse matrix is ​​generated. Thus, a random pulse matrix with higher true randomness can be obtained, thereby improving the performance of the image classification model generated by subsequent training, improving the processing speed and accuracy during image classification processing, and reducing the waste of computing resources. Finally, the image category information corresponding to the random pulse matrix and the image pixel information is stored in a preset database for training the image classification model. Thus, the model training can be performed by the obtained random pulse matrix with higher true randomness, which can improve the processing speed and accuracy of the image classification model generated by the random pulse matrix, thereby reducing the waste of computing resources. Also because the noise signal used to generate the random pulse matrix is ​​obtained by performing multiple optimization processes on the electronic noise signal, and the activation signal used to generate the random pulse matrix is ​​obtained by performing random processing on the image pixel information, the random pulse matrix generated by the above noise signal and activation signal has a higher true randomness. Also because when generating the random pulse matrix, it is obtained by performing random processing on each noise signal and each activation signal, so the true randomness of the generated random pulse matrix is ​​higher, thereby making the performance of the trained image classification model higher, thereby improving the processing speed when processing image data is slow, and improving the accuracy of image classification. In summary, by processing the image pixel information and the electronic noise signal, the true randomness of the random pulse matrix can be improved, and the performance of the image classification model generated by the random pulse matrix can be improved, thereby improving the processing speed and accuracy when the model is used for image classification processing, thereby reducing the waste of computing resources.

[0097] Continue to refer Figure 2 , shows a process 200 of some embodiments of the image classification method according to the present disclosure. The image classification method comprises the following steps:

[0098] Step 201: Acquire an image dataset.

[0099] In some embodiments, an execution subject (such as a computing device) of the image classification method may obtain an image data set. The execution subject may be a terminal or a server for image classification. The image category information corresponding to each image data in the image data set is different.

[0100] Step 202 , segmenting the foreground target image of each image data in the image data set to obtain a foreground target image set.

[0101] In some embodiments, the execution subject may perform segmentation processing on the foreground target image of each image data in the image data set to obtain a foreground target image set. The foreground target image may be an image that is different from the background image. In practice, the execution subject may perform segmentation processing on the foreground target image of each image data in the image data set by a foreground target segmentation model to obtain a foreground target image set. The foreground target segmentation model may be a segmentation model.

[0102] Step 203, for each foreground object image in the foreground object image set, perform the following steps:

[0103] Step 2031, determining each pixel in the foreground target image as a corresponding foreground target image pixel matrix.

[0104] In some embodiments, the execution subject may determine each pixel in the foreground target image as a corresponding foreground target image pixel matrix. The foreground target image pixel matrix may be a matrix obtained by the foreground target image. In practice, first, the execution subject may determine the first pixel point in the upper left corner of the foreground target image as the coordinate origin. Then, the foreground target image may be determined horizontally to the right as the abscissa axis and the foreground target image may be determined vertically downward as the ordinate axis. Thus, a coordinate system may be determined. Then, the execution subject may determine the coordinates corresponding to each pixel point in the foreground target image through the coordinate system. Then, for each pixel point in each pixel point of the foreground target image, the execution subject may determine the position of the pixel point in the matrix through the coordinates corresponding to the pixel point. Then, the pixel value corresponding to the pixel point may be determined as the element value of the corresponding position. Finally, the obtained matrix may be determined as the foreground target image pixel matrix. As an example: when the coordinates of a pixel point are (0, 1) and the corresponding pixel value is 255, the corresponding element value at the position of the first column and the second row in the matrix is ​​255. When the coordinates of a pixel point are (3, 6) and the corresponding pixel value is 100, the corresponding element value at the position of the seventh row in the fourth column of the matrix is ​​100.

[0105] Step 2032: Based on the foreground target image pixel matrix, generate a random pulse matrix corresponding to the foreground target image pixel matrix.

[0106] In some embodiments, the execution subject may generate a random pulse matrix corresponding to the foreground target image pixel matrix based on the foreground target image pixel matrix. The method of generating the random pulse matrix corresponding to the foreground target image pixel matrix may refer to Figure 1 The corresponding steps 103 to 105 in the embodiments are not described in detail here.

[0107] Step 2033, input the generated random pulse matrix into a pre-trained image classification model to obtain image category information corresponding to the random pulse matrix.

[0108] In some embodiments, the execution subject may input the random pulse matrix into a pre-trained image classification model to obtain image category information corresponding to the random pulse matrix. The image classification model is trained based on the random pulse matrix stored in a preset database. The random pulse matrix stored in the preset database is obtained by Figure 1 The image classification model may be a pre-trained neural network model that takes a random pulse matrix corresponding to an image as input and image category information corresponding to the image as output. For example, the neural network model may be a pulse neural network.

[0109] Step 2034, based on the image category information, determine the category corresponding to the foreground target image.

[0110] In some embodiments, the execution subject may determine the category corresponding to the foreground target image based on the image category information. In practice, the execution subject may determine the category corresponding to the foreground target image through the image category information. For example, when the image category information is "fruit", the category corresponding to the foreground target image is fruit. When the image category information is "animal", the category corresponding to the foreground target image is animal.

[0111] The above-mentioned embodiments of the present disclosure have the following beneficial effects: through the image classification method of some embodiments of the present disclosure, the performance of the image classification model generated by the random pulse matrix can be improved, the processing speed when the model is used to process image data can be improved, and the accuracy of image classification can be improved. Specifically, the reason for the poor performance of the image classification model, which in turn leads to a slow processing speed when the model is used to process image data, and the low accuracy of image classification is that the true randomness of the random pulse matrix obtained by directly encoding the pixel values ​​in the image is low, resulting in the poor performance of the image classification model generated by the random pulse matrix, which in turn leads to a slow processing speed and low accuracy when the model is used to process image classification, resulting in a waste of computing resources. Based on this, the image classification method of some embodiments of the present disclosure, first, obtains an image data set. Thus, image data for model training can be obtained. Then, the foreground target image of each image data in the above-mentioned image data set is segmented to obtain a foreground target image set. Thus, the foreground target can be quickly located, the processing speed of subsequent image classification processing can be improved, and the accuracy of image classification can be improved. Then, for each foreground target image in the foreground target image set, the following steps are performed: First, each pixel in the foreground target image is determined as a corresponding foreground target image pixel matrix. In this way, the image data can be converted into an image pixel matrix, which can provide a data basis for generating a random pulse matrix corresponding to the image data. Then, based on the foreground target image pixel matrix, a random pulse matrix corresponding to the foreground target image pixel matrix is ​​generated. Thus, through a variety of processing, the foreground target image pixel matrix corresponding to the image data can be converted into a random pulse matrix with higher true randomness, which can improve the processing speed and accuracy during image classification processing and save computing resources. Then, the generated random pulse matrix is ​​input into a pre-trained image classification model to obtain the image category information corresponding to the random pulse matrix, wherein the image classification model is trained based on the random pulse matrix stored in a preset database, and the random pulse matrix stored in the preset database is obtained through Figure 1 The steps in the corresponding embodiments are generated. Thus, the corresponding image category information of the image can be obtained through the random pulse matrix corresponding to the image. Finally, based on the above image category information, the category corresponding to the above foreground target image is determined. Thus, the final image category information can be determined to complete the image classification task. Also, because the foreground target image of each image data in the above image data set can be segmented to obtain a foreground target image set, and then the images in the foreground target image set can be subjected to image classification processing. Therefore, when classifying images, the images to be classified can be quickly located, the processing speed and accuracy of the image classification processing can be improved, and computing resources can be saved.

[0112] Further references Figure 3 As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of a random pulse matrix generation device for image classification model training. These device embodiments are similar to Figure 1 Corresponding to the method embodiments shown, the device can be specifically applied to various electronic devices.

[0113] like Figure 3 As shown, some embodiments of the random pulse matrix generation device 300 for image classification model training include: an acquisition unit 301, a first generation unit 302, a second generation unit 303, an execution unit 304, a third generation unit 305 and a storage unit 306. The acquisition unit 301 is configured to acquire an electronic noise signal set; the first generation unit 302 is configured to generate each random noise signal based on at least one electronic noise signal in the electronic noise signal set; the second generation unit 303 is configured to generate an image probability matrix corresponding to the above image pixel information based on pre-stored image pixel information, wherein the above image pixel information corresponds to image category information; the execution unit 304 is configured to perform the following first loop steps based on each random noise signal: based on a preset value, update the preset pulse matrix generation times; input a preset number of random noise signals in each random noise signal into a noise signal collector to obtain each noise signal corresponding to the preset number of random noise signals; based on the above image probability matrix array, generating each activation signal; based on the above-mentioned each noise signal and the above-mentioned each activation signal, generating each probability signal group; based on the above-mentioned each probability signal group, generating a random pulse sub-matrix; in response to determining that the updated pulse matrix generation number does not meet the preset loop condition, adding the above-mentioned random pulse sub-matrix to the random pulse sub-matrix set to update the random pulse sub-matrix set; in response to determining that the updated pulse matrix generation number meets the preset loop condition, executing the above-mentioned first loop step again; the third generating unit 305 is configured to generate a random pulse matrix based on the updated random pulse sub-matrix set; the storage unit 306 is configured to store the image category information corresponding to the above-mentioned random pulse matrix and the above-mentioned image pixel information in a preset database for training the image classification model.

[0114] It is understood that the units described in the device 300 are similar to those described in the reference Figure 1 Therefore, the operations, features and beneficial effects described above for the method are also applicable to the device 300 and the units contained therein, and will not be described in detail here.

[0115] Further references Figure 4 As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of an image classification device. These device embodiments are similar to Figure 2 Corresponding to the method embodiments shown, the device can be specifically applied to various electronic devices.

[0116] like Figure 4 As shown, the image classification device 400 of some embodiments includes: an acquisition unit 401, a segmentation unit 402 and an execution unit 403. The acquisition unit 401 is configured to acquire an image data set; the segmentation unit 402 is configured to perform segmentation processing on the foreground target image of each image data in the above image data set to obtain a foreground target image set; the execution unit 403 is configured to perform the following steps for each foreground target image in the above foreground target image set: determine each pixel in the above foreground target image as a corresponding foreground target image pixel matrix; based on the above foreground target image pixel matrix, generate a random pulse matrix corresponding to the above foreground target image pixel matrix; input the generated random pulse matrix into a pre-trained image classification model to obtain image category information corresponding to the above random pulse matrix, wherein the above image classification model is trained based on the random pulse matrix stored in a preset database, and the random pulse matrix stored in the preset database is obtained by Figure 1 The steps in the corresponding embodiments are generated; based on the above image category information, the category corresponding to the above foreground target image is determined.

[0117] It is understood that the units described in the device 400 are similar to those described in the reference Figure 2 Therefore, the operations, features and beneficial effects described above for the method are also applicable to the device 400 and the units included therein, and will not be described in detail here.

[0118] Reference below Figure 5 , which shows a structural schematic diagram of an electronic device 500 suitable for implementing some embodiments of the present disclosure. Figure 5 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0119] like Figure 5 As shown, the electronic device 500 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage device 508 into a random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the electronic device 500 are also stored. The processing device 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0120] Typically, the following devices may be connected to the I / O interface 505: an input device 506 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 507 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 508 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 509. The communication device 509 may allow the electronic device 500 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 5 The electronic device 500 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead. Figure 5 Each block shown in the figure may represent one device, or may represent multiple devices as required.

[0121] In particular, according to some embodiments of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, some embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In some such embodiments, the computer program can be downloaded and installed from the network through the communication device 509, or installed from the storage device 508, or installed from the ROM 502. When the computer program is executed by the processing device 501, the above-mentioned functions defined in the method of some embodiments of the present disclosure are executed.

[0122] It should be noted that the computer-readable medium recorded in some embodiments of the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In some embodiments of the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In some embodiments of the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0123] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (HyperText Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0124] The above-mentioned computer-readable medium may be included in the above-mentioned electronic device; or it may exist independently without being assembled into the electronic device. The above-mentioned computer-readable medium carries one or more programs. When the above-mentioned one or more programs are executed by the electronic device, the electronic device: obtains an electronic noise signal set; generates each random noise signal based on at least one electronic noise signal in the electronic noise signal set; generates an image probability matrix corresponding to the above-mentioned image pixel information based on pre-stored image pixel information, wherein the above-mentioned image pixel information corresponds to image category information; based on each random noise signal, performs the following first loop step: based on a preset value, updates a preset number of pulse matrix generation times; inputs a preset number of random noise signals in each random noise signal into a noise signal collector, and obtains each noise signal corresponding to the above-mentioned preset number of random noise signals. signal; based on the above-mentioned image probability matrix, generate each activation signal; based on the above-mentioned each noise signal and the above-mentioned each activation signal, generate each probability signal group; based on the above-mentioned each probability signal group, generate a random pulse sub-matrix; in response to determining that the updated pulse matrix generation number does not meet the preset loop condition, based on adding the above-mentioned random pulse sub-matrix to the random pulse sub-matrix set, to update the random pulse sub-matrix set; in response to determining that the updated pulse matrix generation number meets the preset loop condition, execute the above-mentioned first loop step again; based on the updated random pulse sub-matrix set, generate a random pulse matrix; store the image category information corresponding to the above-mentioned random pulse matrix and the above-mentioned image pixel information in a preset database for training an image classification model.

[0125] Or the electronic device is configured to: obtain an image data set; segment the foreground target image of each image data in the image data set to obtain a foreground target image set; for each foreground target image in the foreground target image set, perform the following steps: determine each pixel in the foreground target image as a corresponding foreground target image pixel matrix; based on the foreground target image pixel matrix, generate a random pulse matrix corresponding to the foreground target image pixel matrix; input the generated random pulse matrix into a pre-trained image classification model to obtain image category information corresponding to the random pulse matrix, wherein the image classification model is trained based on the random pulse matrix stored in a preset database, and the random pulse matrix stored in the preset database is obtained by Figure 1 The steps in the corresponding embodiments are generated; based on the above image category information, the category corresponding to the above foreground target image is determined.

[0126] Computer program code for performing the operations of some embodiments of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0127] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0128] The units described in some embodiments of the present disclosure may be implemented by software or hardware. The units described may also be set in a processor, for example, it may be described as: a processor includes an acquisition unit, a first generation unit, a second generation unit, an execution unit, a third generation unit and a storage unit. The names of these units do not constitute a limitation on the units themselves in some cases, for example, the acquisition unit may also be described as a "unit for acquiring an electronic noise signal set".

[0129] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0130] The above descriptions are only some preferred embodiments of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above invention concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the embodiments of the present disclosure.

Claims

1. A random pulse matrix generation method for image classification model training, comprising: acquiring an electronic noise signal set; generating respective random noise signals based on at least one electronic noise signal in the electronic noise signal set; Based on pre-stored image pixel information, generating an image probability matrix corresponding to the image pixel information, wherein the image pixel information corresponds to image category information; Based on each random noise signal, the following first cycle steps are performed: Based on the preset value, the preset pulse matrix generation times are updated; Inputting a preset number of random noise signals from each random noise signal into a noise signal collector to obtain each noise signal corresponding to the preset number of random noise signals; Based on the image probability matrix, generating respective activation signals; generating respective probability signal groups based on the respective noise signals and the respective activation signals; Based on the respective probability signal groups, generating a random pulse sub-matrix; In response to determining that the updated pulse matrix generation number does not satisfy a preset cycle condition, the random pulse sub-matrix is ​​added to the random pulse sub-matrix set to update the random pulse sub-matrix set; In response to determining that the updated pulse matrix generation number satisfies a preset loop condition, executing the first loop step again; generating a random pulse matrix based on the updated set of random pulse sub-matrices; The image category information corresponding to the random pulse matrix and the image pixel information is stored in a preset database for use in training an image classification model.

2. The method according to claim 1, wherein: The generating of the random noise signals based on at least one electronic noise signal in the electronic noise signal set comprises: Based on the electronic noise signal set, the following second loop steps are performed: Performing current increase processing on the electronic noise signal in the electronic noise signal set to obtain a current noise signal; Performing voltage reduction processing on the current noise signal to obtain a voltage-reduced noise signal; Performing signal amplification processing on the voltage-reduced noise signal to obtain an amplified noise signal; Performing signal conversion processing on the amplified noise signal to obtain a binary noise signal; Performing digital extraction processing on the binary noise signal to obtain binary digital numbers; Adding the binary digit to the binary digit sequence to update the binary digit sequence; In response to determining that the updated binary digital sequence meets a preset sequence condition, determining the updated binary digital sequence as a random noise signal, and clearing the updated binary signal sequence; In response to determining that the number of executions of the second loop step does not satisfy a preset number condition, an unused electronic noise signal is used to form an electronic noise signal set, and the second loop step is executed again.

3. The method according to claim 1, wherein: The step of generating an image probability matrix corresponding to the image pixel information based on the pre-stored image pixel information includes: Generate an image pixel matrix based on pre-stored image pixel information; Based on a preset normalization value, each matrix element in the image pixel matrix is ​​normalized to obtain an image probability matrix.

4. The method according to claim 1, wherein: The image classification model is trained in the following way: Acquire a sample set from the preset database, wherein the samples in the sample set include an image, a sample random pulse matrix corresponding to image pixel information in the image, and sample image category information; Perform the following training steps based on the sample set: Inputting the sample random pulse matrices corresponding to the image of at least one sample in the sample set into the initial neural network respectively to obtain the image category information corresponding to each sample in the at least one sample, wherein the initial neural network includes a feature extraction network and an output layer; Comparing the image category information corresponding to each sample of the at least one sample with the corresponding sample image category information; Determine whether the initial neural network reaches the preset optimization goal according to the comparison result; In response to determining that the initial neural network achieves the optimization goal, determining a feature extraction network included in the initial neural network as an image classification model; In response to determining that the initial neural network does not achieve the optimization goal, the network parameters of the initial neural network are adjusted, and unused samples are used to form a sample set, and the adjusted initial neural network is used as the initial neural network to perform the training step again.

5. An image classification method, comprising: Get image dataset; Segmenting the foreground target image of each image data in the image data set to obtain a foreground target image set; For each foreground target image in the foreground target image set, the following steps are performed: Determine each pixel in the foreground target image as a corresponding foreground target image pixel matrix; Based on the foreground target image pixel matrix, generating a random pulse matrix corresponding to the foreground target image pixel matrix; Inputting the generated random pulse matrix into a pre-trained image classification model to obtain image category information corresponding to the random pulse matrix, wherein the image classification model is trained based on the random pulse matrix stored in a preset database, and the random pulse matrix stored in the preset database is generated by the method according to any one of claims 1 to 4; Based on the image category information, a category corresponding to the foreground target image is determined.

6. A random pulse matrix generation device for image classification model training, comprising: an acquisition unit configured to acquire an electronic noise signal set; A first generating unit configured to generate respective random noise signals based on at least one electronic noise signal in the electronic noise signal set; A second generating unit is configured to generate an image probability matrix corresponding to the image pixel information based on pre-stored image pixel information, wherein the image pixel information corresponds to image category information; The execution unit is configured to execute the following first loop step based on each random noise signal: based on a preset value, update the preset pulse matrix generation times; input a preset number of random noise signals in each random noise signal into a noise signal collector to obtain each noise signal corresponding to the preset number of random noise signals; based on the image probability matrix, generate each activation signal; based on the each noise signal and the each activation signal, generate each probability signal group; based on the each probability signal group, generate a random pulse sub-matrix; in response to determining that the updated pulse matrix generation times do not meet the preset loop condition, add the random pulse sub-matrix to the random pulse sub-matrix set to update the random pulse sub-matrix set; in response to determining that the updated pulse matrix generation times meet the preset loop condition, execute the first loop step again; A third generating unit is configured to generate a random pulse matrix based on the updated random pulse sub-matrix set; The storage unit is configured to store the image category information corresponding to the random pulse matrix and the image pixel information in a preset database for training an image classification model.

7. An image classification device, comprising: an acquisition unit configured to acquire an image dataset; a segmentation unit configured to perform segmentation processing on the foreground target image of each image data in the image data set to obtain a foreground target image set; The execution unit is configured to perform the following steps for each foreground target image in the foreground target image set: determining each pixel in the foreground target image as a corresponding foreground target image pixel matrix; Based on the foreground target image pixel matrix, generating a random pulse matrix corresponding to the foreground target image pixel matrix; The generated random pulse matrix is ​​input into a pre-trained image classification model to obtain image category information corresponding to the random pulse matrix, wherein the image classification model is trained based on the random pulse matrix stored in a preset database, and the random pulse matrix stored in the preset database is generated by the method described in one of claims 1-4; based on the image category information, the category corresponding to the foreground target image is determined.

8. An electronic device, comprising: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 4 or 5.

9. A computer readable medium having a computer program stored thereon, wherein: When the program is executed by a processor, the method according to any one of claims 1 to 4 or 5 is implemented.

Citation Information

Patent Citations

  • Deep learning object classification method based on multi-channel model fusion

    CN113505802A

  • Semi-supervised semantic segmentation method based on self-adaptive false label correction

    CN115761735A

  • Radiation source signal sorting method, device and equipment based on feature extraction and medium

    CN115963463A

  • Image classification model training method and device, image classification method and device and electronic equipment

    CN117934948A

  • Image processing method and device, neural network training method, image processing method of combined neural network model, construction method of combined neural network model, neural network processor and storage medium

    WO2021073493A1