Image feature transmission method, device and system

Abstract

The present disclosure relates to a method, device and system for transmitting image features, and to the field of communication technology. The transmission method includes: using a machine learning model to extract first feature information of an image to be processed, the first feature information including feature matrices of each channel; determining a redundant feature matrix according to the amount of information of each feature matrix; deleting the redundant feature matrix from the first feature information to generate second feature information; transmitting the encoded second feature information and related information of the redundant feature matrix to a decoding end.

Description

Technical Field

[0001] The present disclosure relates to the field of communication technology, and in particular to a method for transmitting image features, a device for transmitting image features, a system for transmitting image features, and a non-volatile computer-readable storage medium. Background Art

[0002] With the growth of machine learning applications, many intelligent platforms have been adopted in the fields of Internet of Vehicles, video surveillance, smart cities, etc. These platforms generate massive data communications with a large number of sensors.

[0003] In the related art, data in communication is encoded by a coding method oriented toward human vision.

[0004] The growth in data volume has directly led to previous inefficiencies, and it is difficult to meet reality in terms of latency and scale, so feature coding for intelligent machines has been put on the agenda. Summary of the invention

[0005] The inventors of the present disclosure have found that the above-mentioned related technologies have the following problems: the coding compression rate is low, and it is difficult to ensure the communication quality when the amount of communication data increases.

[0006] In view of this, the present disclosure proposes a technical solution for transmitting image features, which can improve the coding compression rate and thus ensure the communication quality.

[0007] According to some embodiments of the present disclosure, a method for transmitting image features is provided, including: using a machine learning model to extract first feature information of an image to be processed, the first feature information including a feature matrix of each channel; determining a redundant feature matrix based on the amount of information of each feature matrix; deleting the redundant feature matrix from the first feature information to generate second feature information; and transmitting the encoded second feature information and related information of the redundant feature matrix to a decoding end.

[0008] In some embodiments, determining the redundant feature matrix according to the information content of each feature matrix includes: calculating the sum of the eigenvalues ​​or the mean of the eigenvalues ​​in each feature matrix as the information content of each feature matrix; and determining the feature matrix with an information content of 0 as a redundant feature matrix.

[0009] In some embodiments, determining the redundant feature matrix based on the amount of information of each feature matrix includes: sorting the feature matrices whose information amount is not 0 in order from small to large; and determining the feature matrices whose sorted sequence numbers are less than or equal to the sequence number threshold and the feature matrices whose information amount is 0 as redundant feature matrices.

[0010] In some embodiments, determining the redundant feature matrix based on the information content of each feature matrix includes: calculating the sum of the eigenvalues ​​or the mean of the eigenvalues ​​in each feature matrix as the information content of each feature matrix; and determining the feature matrix whose information content is less than or equal to the information content threshold as a redundant feature matrix.

[0011] In some embodiments, the transmission method further includes: obtaining the second feature information and related information of the redundant feature matrix through decoding processing at the decoding end; generating third feature information at the decoding end based on the second feature information and related information of the redundant feature matrix for processing the image to be processed.

[0012] In some embodiments, the decoding end generates the third characteristic information based on the relevant information of the second characteristic information and the redundant characteristic matrix, including: generating a corresponding number of all-0 matrices based on the relevant information of the redundant characteristic matrix, and determining the channels corresponding to the redundant characteristic matrix; and inserting the all-0 matrix into the second characteristic information based on the corresponding channels to generate the third characteristic information.

[0013] In some embodiments, transmitting the encoded second feature information and related information of the redundant feature matrix to the decoding end includes: quantizing and encoding the second feature information and related information of the redundant feature matrix and transmitting them to the decoding end.

[0014] In some embodiments, the transmission method further includes performing decoding processing and inverse quantization processing at the decoding end to obtain relevant information of the second feature information and the redundant feature matrix.

[0015] According to some other embodiments of the present disclosure, there is provided an image feature transmission device, comprising: an extraction unit, used to extract first feature information of an image to be processed by using a machine learning model, the first feature information comprising a feature matrix of each channel; a determination unit, used to determine a redundant feature matrix according to the amount of information of each feature matrix; a first generation unit, used to delete the redundant feature matrix from the first feature information, and generate second feature information; and a transmission unit, used to transmit the encoded second feature information and related information of the redundant feature matrix to a decoding end.

[0016] In some embodiments, the determination unit calculates the sum of eigenvalues ​​or the mean of eigenvalues ​​in each feature matrix as the information content of each feature matrix, and determines a feature matrix with an information content of 0 as a redundant feature matrix.

[0017] In some embodiments, the determination unit sorts the feature matrices whose information amount is not 0 in order of small to large information amount, and determines the feature matrices whose sorted sequence numbers are less than or equal to the sequence number threshold and the feature matrices whose information amount is 0 as redundant feature matrices.

[0018] In some embodiments, the determination unit calculates the sum of the eigenvalues ​​or the mean of the eigenvalues ​​in each feature matrix as the information content of each feature matrix, and determines the feature matrix whose information content is less than or equal to the information content threshold as a redundant feature matrix.

[0019] In some embodiments, the transmission device also includes: an acquisition unit, used to acquire the second feature information and related information of the redundant feature matrix through decoding processing at the decoding end; a second generation unit, used to generate third feature information at the decoding end based on the second feature information and related information of the redundant feature matrix for processing the image to be processed.

[0020] In some embodiments, the second generating unit generates a corresponding number of all-0 matrices according to relevant information of the redundant feature matrix, determines the channels corresponding to the redundant feature matrix, and inserts the all-0 matrices into the second feature information according to the corresponding channels to generate third feature information.

[0021] In some embodiments, the transmission unit quantizes and encodes the second feature information and related information of the redundant feature matrix and transmits the result to the decoding end.

[0022] In some embodiments, the acquisition unit performs decoding processing and inverse quantization processing at the decoding end to acquire the second feature information and related information of the redundant feature matrix.

[0023] According to some other embodiments of the present disclosure, a system for transmitting image features is provided, including: an encoding end, used to extract first feature information of an image to be processed by using a machine learning model, the first feature information including a feature matrix of each channel, determining a redundant feature matrix according to the amount of information of each feature matrix, deleting the redundant feature matrix from the first feature information, generating second feature information, and transmitting the encoded second feature information and related information of the redundant feature matrix to a decoding end; a decoding end, used to obtain related information of the second feature information and the redundant feature matrix through decoding processing, and generating third feature information at the decoding end according to the second feature information and related information of the redundant feature matrix for processing the image to be processed.

[0024] According to some further embodiments of the present disclosure, there is provided a device for transmitting image features, comprising: a memory; and a processor coupled to the memory, wherein the processor is configured to execute the method for transmitting image features in any of the above embodiments based on instructions stored in the memory device.

[0025] According to some further embodiments of the present disclosure, a non-volatile computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the method for transmitting image features in any of the above embodiments is implemented.

[0026] In the above embodiment, redundant information in the information to be transmitted is eliminated according to the amount of information in each channel feature matrix of the machine learning model, thereby improving the coding compression rate and ensuring the communication quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings, which constitute a part of the specification, illustrate embodiments of the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0028] The present disclosure may be more clearly understood from the following detailed description with reference to the accompanying drawings:

[0029] Figure 1 A flowchart showing some embodiments of the image feature transmission method disclosed herein;

[0030] Figure 2 Flowcharts showing other embodiments of the image feature transmission method disclosed herein;

[0031] Figure 3 Flowcharts showing some further embodiments of the image feature transmission method disclosed herein;

[0032] Figure 4 A block diagram showing some embodiments of the transmission device of the image features of the present disclosure;

[0033] Figure 5 A block diagram showing some other embodiments of the image feature transmission device disclosed in the present invention;

[0034] Figure 6 A block diagram showing some further embodiments of the apparatus for transmitting image features disclosed herein;

[0035] Figure 7 A block diagram illustrating some embodiments of the image feature transmission system of the present disclosure. DETAILED DESCRIPTION

[0036] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present disclosure unless otherwise specifically stated.

[0037] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0038] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present disclosure, its application, or uses.

[0039] Technologies, methods and equipment known to ordinary technicians in the relevant field may not be discussed in detail, but where appropriate, the technologies, methods and equipment should be considered part of the authorization specification.

[0040] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0041] It should be noted that like reference numerals and letters refer to similar items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0042] As mentioned above, the growth of data volume directly leads to the inefficiency of previous encoding methods for human vision, which are also difficult to meet the reality in terms of latency and scale. Therefore, feature encoding methods for intelligent machines are needed.

[0043] The premise of coding compression is redundancy, and the purpose of coding compression is to remove redundancy, thereby achieving the purpose of compression. Therefore, the present invention improves the compression rate of feature coding based on the redundant features of neural networks.

[0044] For example, there are three types of channels in a convolutional neural network: the channels of the initial input image sample depend on the image type (such as RGB red, green and blue channels); the out_channels output after the convolution operation is completed depends on the number of convolution kernels. The out_channels at this time will also be used as the in_channels of the convolution kernel in the next convolution; the in_channels in the convolution kernel is the out_channels of the previous convolution.

[0045] In other words, the same channel feature is the result of the convolution operation of all output channels of the previous layer with the convolution kernels with the same number of layers. The neural network will autonomously learn the weight values ​​of each convolution kernel during the training process and extract the information of interest through different convolution kernel weight values. Therefore, the feature map of the middle layer of the convolutional neural network must contain a large amount of non-attention information, that is, feature redundant information. Redundancy in feature maps is an important feature of convolutional neural networks.

[0046] From the perspective of biomimetic biology, relevant brain research shows that the information encoding of biological neurons is usually relatively scattered and sparse. Usually, only about 1% to 4% of neurons in the brain are active at the same time. Linear correction and regularization can be used to debug the activity of neurons in machine neural networks (i.e., the output is positive).

[0047] In contrast, the logistic function reaches 1 / 2 when the input is 0, which is a semi-saturated stable state, which is not in line with the expectations of actual biology for simulated neural networks. Generally speaking, in a neural network using a rectified linear unit (i.e., linear rectification), about 50% of the neurons are in an activated state.

[0048] Convolutional neural networks have the characteristics of local perception. Each neuron only perceives a part of the image, not the entire image. Local pixels are closely related, while distant pixels are less related. The eigenvalues ​​will tilt toward the perceived target object area.

[0049] Therefore, the shallow network contains more position information, and the neurons (feature channels) that do not perceive the target object in the shallow network contain redundant information. In feature coding, these neurons that do not perceive the target object can be treated as redundant information and not enter the coding object, reducing the number of codes and improving the compression rate.

[0050] The present disclosure proposes a method for removing redundancy from the feature matrix of redundant channels by calculating the amount of shallow feature information based on the channel feature distribution characteristics of the shallow intermediate layer of the neural network. For example, this can be achieved through the following embodiments.

[0051] Figure 1 A flow chart showing some embodiments of the image feature transmission method disclosed herein.

[0052] like Figure 1 As shown, in step 110, the first feature information of the image to be processed is extracted using a machine learning model. The first feature information includes a feature matrix of each channel.

[0053] In some embodiments, Cascade R-CNN (Regions with Convolutional Neural Network) ResNet101 (Residual Network) is used to process an RGB (Red Green Blue) image to be processed with a size of 4864×3648×3.

[0054] For example, the feature layer output contained in the intermediate layer features after the first pooling layer of Cascade R-CNN ResNet101 can be used as the object to be encoded. The intermediate layer can be a 64×200×272 layer, 64 is the number of channels (i.e. the number of feature layers contained), and 200×272 is the size of the feature matrix output by each feature layer. In other words, the first feature information F output by the intermediate layer contains N feature matrices corresponding to N channels.

[0055] In step 120, a redundant feature matrix is ​​determined according to the information content of each feature matrix.

[0056] In some embodiments, the sum of the eigenvalues ​​in each feature matrix is ​​calculated respectively as the information content of each feature matrix; and a feature matrix with an information content of 0 is determined as a redundant feature matrix.

[0057] For example, the feature matrices with non-zero information amounts may be sorted in order of small to large information amounts; feature matrices with sorted sequence numbers less than or equal to a sequence number threshold and feature matrices with zero information amounts may be determined as redundant feature matrices.

[0058] In some embodiments, the sum S of all eigenvalues ​​in the feature matrix i output by the feature layer i can be calculated i , we get the queue S{S n}, n = 0, 1,…, N-1 (such as N = 64).

[0059] All S in S can be sorted from small to large. n Sort and count the number of channels T whose sum of eigenvalues ​​is 0. For example, if T is 15, the corresponding channel (i.e., feature matrix) numbers are 5, 7, 8, 17, 27, 37, 41, 42, 43, 46, 48, 50, 53, 55, and 62.

[0060] In some embodiments, the sum of the eigenvalues ​​or the mean of the eigenvalues ​​in each feature matrix is ​​calculated as the information content of each feature matrix; a feature matrix whose information content is less than or equal to an information content threshold is determined as a redundant feature matrix.

[0061] In step 130, the redundant feature matrix is ​​deleted from the first feature information to generate second feature information.

[0062] In some embodiments, redundant channels may be eliminated from the first feature information F containing N channel feature matrices. For example, S{S n}The first T+M S n The feature matrix of the corresponding channel is obtained, and the redundant intermediate layer feature F0 with a total number of channels (NTM) is obtained, that is, the second feature information.

[0063] M is a sequence number threshold set according to actual conditions (e.g., M can be 13), which is used to determine the number of redundant channels to be removed. For example, the sequence numbers of the 13 redundant channels are 51, 13, 14, 22, 25, 19, 49, 9, 39, 44, 18, 38, and 35.

[0064] In step 140, the encoded second feature information and related information of the redundant feature matrix are transmitted to a decoding end.

[0065] In some embodiments, the second feature information and the related information of the redundant feature matrix are quantized and encoded and then transmitted to the decoding end. For example, F0 and the index list of (T+M) redundant channels can be quantized and encoded to obtain a binary stream to be transmitted. For example, the index list is used as the related information of the redundant feature matrix, which records the channel numbers corresponding to all deleted redundant feature matrices. Quantization can be a quantization method such as uniform quantization.

[0066] In some embodiments, the decoding end can Figure 2 The embodiments in the present invention implement the technical solution of the present disclosure.

[0067] Figure 2 Flowcharts showing other embodiments of the image feature transmission method disclosed in the present invention.

[0068] like Figure 2 As shown, in step 210, the second feature information and related information of the redundant feature matrix are obtained through decoding processing at the decoding end.

[0069] In some embodiments, decoding and dequantization are performed at the decoding end to obtain relevant information of the second feature information and the redundant feature matrix. For example, the binary code stream is decoded and dequantized to obtain the second information feature F0 and the index list of the redundant channel with the total number of channels (NTM) removed from the redundant feature matrix.

[0070] For example, the index list records the serial numbers of T feature matrices with an information content of 0: 5, 7, 8, 17, 27, 37, 41, 42, 43, 46, 48, 50, 53, 55, 62, and the serial numbers of M feature matrices with a higher ranking (i.e., smaller information content): 51, 13, 14, 22, 25, 19, 49, 9, 39, 44, 18, 38, 35.

[0071] In step 220, the decoding end generates third feature information based on the second feature information and related information of the redundant feature matrix for processing the image to be processed.

[0072] In some embodiments, a corresponding number of all-0 matrices are generated according to relevant information of the redundant feature matrix, and channels corresponding to the redundant feature matrix are determined; and according to the corresponding channels, the all-0 matrices are inserted into the second feature information to generate third feature information.

[0073] For example, a channel adding operation is performed on the second feature information F0. According to the sequence number in the index list, each replacement feature matrix with all feature values ​​being 0 is generated and inserted into the corresponding feature matrix position of F0.

[0074] Figure 3 Flowcharts showing still other embodiments of the method for transmitting image features disclosed herein.

[0075] like Figure 3 As shown, in step 310, each coding element contained in the intermediate layer of the neural network used to extract image features is selected as the coding layer to be encoded.

[0076] In step 320, the sum of all eigenvalues ​​in the feature matrix of each channel output by each feature layer is calculated to obtain a queue S{S n}, n = 0, 1, …, N-1, N is the number of channels.

[0077] In step 330, the queue S{S n} are sorted, and the number of channels whose sum of eigenvalues ​​is 0 is counted as T.

[0078] In step 340, redundant channels are removed from the intermediate layer output feature F containing N channels. n}, and obtain the feature F0 of the redundant middle layer with a total number of channels (NTM). M is the threshold for removing redundant channels.

[0079] In step 350, quantization encoding operations are performed on F0 and (T+M) redundant channel index lists to obtain a binary stream.

[0080] In step 360, the binary code stream is de-encoded and de-quantized to obtain a de-redundant intermediate layer feature F0 with a total number of channels (NTM) and a redundant channel index list. A channel addition operation is performed on the intermediate layer feature F0 to add a channel with a sequence number of the element in the list and a feature value of all 0.

[0081] In the above embodiment, redundant information in the information to be transmitted is eliminated according to the amount of information in each channel feature matrix of the machine learning model, thereby improving the coding compression rate and ensuring the communication quality.

[0082] Figure 4 A block diagram showing some embodiments of the apparatus for transmitting image features of the present disclosure.

[0083] like Figure 4 As shown, the image feature transmission device 4 includes an extraction unit 41 , a determination unit 42 , a first generation unit 43 and a transmission unit 44 .

[0084] The extraction unit 41 uses a machine learning model to extract first feature information of the image to be processed. The first feature information includes a feature matrix of each channel.

[0085] The determination unit 42 determines a redundant feature matrix according to the information amount of each feature matrix.

[0086] In some embodiments, the determination unit 42 calculates the sum of the eigenvalues ​​or the mean of the eigenvalues ​​in each feature matrix as the information content of each feature matrix; and determines a feature matrix with an information content of 0 as a redundant feature matrix.

[0087] In some embodiments, the determination unit 42 sorts the feature matrices whose information amount is not 0 in order from small to large; and determines the feature matrices whose sorted sequence numbers are less than or equal to the sequence number threshold and the feature matrices whose information amount is 0 as redundant feature matrices.

[0088] In some embodiments, the determination unit 42 calculates the sum of the eigenvalues ​​in each feature matrix as the information content of each feature matrix; and determines a feature matrix whose information content is less than or equal to an information content threshold as a redundant feature matrix.

[0089] The first generating unit 43 deletes the redundant feature matrix from the first feature information to generate second feature information.

[0090] The transmission unit 44 transmits the encoded second feature information and related information of the redundant feature matrix to the decoding end.

[0091] In some embodiments, the transmission unit 44 performs quantization processing and encoding processing on the second feature information and related information of the redundant feature matrix and transmits the result to the decoding end.

[0092] In some embodiments, the transmission device 4 also includes: an acquisition unit 45, which is used to acquire the second feature information and related information of the redundant feature matrix through decoding processing at the decoding end; a second generation unit 46, which is used to generate third feature information at the decoding end based on the second feature information and related information of the redundant feature matrix for processing the image to be processed.

[0093] In some embodiments, the second generating unit 46 generates a corresponding number of all-0 matrices according to relevant information of the redundant feature matrix, determines the channels corresponding to the redundant feature matrix, and inserts the all-0 matrices into the second feature information according to the corresponding channels to generate third feature information.

[0094] In some embodiments, the acquisition unit 45 performs decoding processing and inverse quantization processing at the decoding end to acquire the second feature information and related information of the redundant feature matrix.

[0095] Figure 5 A block diagram showing some other embodiments of the apparatus for transmitting image features of the present disclosure.

[0096] like Figure 5As shown, the image feature transmission device 5 of this embodiment includes: a memory 51 and a processor 52 coupled to the memory 51, and the processor 52 is configured to execute the image feature transmission method in any embodiment of the present disclosure based on the instructions stored in the memory 51.

[0097] The memory 51 may include, for example, a system memory, a fixed non-volatile storage medium, etc. The system memory may store, for example, an operating system, an application program, a boot loader, a database, and other programs.

[0098] Figure 6 A block diagram showing some further embodiments of the apparatus for transmitting image features of the present disclosure.

[0099] like Figure 6 As shown, the image feature transmission device 6 of this embodiment includes: a memory 610 and a processor 620 coupled to the memory 610, and the processor 620 is configured to execute the image feature transmission method in any of the aforementioned embodiments based on the instructions stored in the memory 610.

[0100] The memory 610 may include, for example, a system memory, a fixed non-volatile storage medium, etc. The system memory may store, for example, an operating system, an application program, a boot loader, and other programs.

[0101] The image feature transmission device 6 may also include an input / output interface 630, a network interface 640, a storage interface 650, etc. These interfaces 630, 640, 650, the memory 610, and the processor 620 may be connected, for example, via a bus 660. The input / output interface 630 provides a connection interface for input / output devices such as a display, a mouse, a keyboard, a touch screen, a microphone, and a speaker. The network interface 640 provides a connection interface for various networked devices. The storage interface 650 provides a connection interface for external storage devices such as an SD card and a USB flash drive.

[0102] Figure 7 A block diagram illustrating some embodiments of the image feature transmission system of the present disclosure.

[0103] like Figure 7 As shown, the image feature transmission system 7 includes an encoding end 71 and a decoding end 72 .

[0104] The encoding end 71 is used to use a machine learning model to extract the first feature information of the image to be processed, where the first feature information includes a feature matrix of each channel; determine a redundant feature matrix based on the amount of information of each feature matrix; delete the redundant feature matrix from the first feature information to generate second feature information; and transmit the encoded second feature information and related information of the redundant feature matrix to the decoding end.

[0105] The decoding end 72 is used to obtain the second feature information and the related information of the redundant feature matrix through decoding processing; at the decoding end, the third feature information is generated according to the related information of the second feature information and the redundant feature matrix for processing the image to be processed.

[0106] Those skilled in the art will appreciate that the embodiments of the present disclosure may be provided as methods, systems, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present disclosure may take the form of a computer program product implemented on one or more computer-usable non-transient storage media including but not limited to disk storage, CD-ROM, optical storage, etc., which contain computer-usable program code.

[0107] So far, the image feature transmission method, image feature transmission device, image feature transmission system and non-volatile computer-readable storage medium according to the present disclosure have been described in detail. In order to avoid obscuring the concept of the present disclosure, some details known in the art are not described. Based on the above description, those skilled in the art can fully understand how to implement the technical solution disclosed herein.

[0108] The method and system of the present disclosure may be implemented in many ways. For example, the method and system of the present disclosure may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is for illustration only, and the steps of the method of the present disclosure are not limited to the order specifically described above, unless otherwise specifically stated. In addition, in some embodiments, the present disclosure may also be implemented as a program recorded in a recording medium, which includes machine-readable instructions for implementing the method according to the present disclosure. Therefore, the present disclosure also covers a recording medium storing a program for executing the method according to the present disclosure.

[0109] Although some specific embodiments of the present disclosure have been described in detail by way of example, it should be understood by those skilled in the art that the above examples are for illustration only and are not intended to limit the scope of the present disclosure. It should be understood by those skilled in the art that the above embodiments may be modified without departing from the scope and spirit of the present disclosure. The scope of the present disclosure is defined by the appended claims.

Claims

1. A method for transmitting image features, comprising: Using a machine learning model, extracting first feature information of the image to be processed, where the first feature information includes a feature matrix of each channel; Determine the redundant feature matrix according to the information amount of each feature matrix; Deleting the redundant feature matrix from the first feature information to generate second feature information; The encoded second feature information and the related information of the redundant feature matrix are transmitted to the decoding end, The step of determining the redundant feature matrix according to the information amount of each feature matrix includes at least one of the following: The sum of the eigenvalues ​​or the mean of the eigenvalues ​​in each of the characteristic matrices is calculated respectively as the information content of each of the characteristic matrices, and the characteristic matrix with an information content of 0 is determined as the redundant characteristic matrix; Sorting the feature matrices whose information amount is not 0 in order of small to large information amount, and determining the feature matrices whose sequence numbers are less than or equal to the sequence number threshold and the feature matrices whose information amount is 0 as the redundant feature matrices; The sum of the eigenvalues ​​or the mean of the eigenvalues ​​in each of the feature matrices is calculated respectively as the information content of each of the feature matrices, and the feature matrix whose information content is less than or equal to the information content threshold is determined as the redundant feature matrix.

2. The transmission method according to claim 1, further comprising: Obtaining the second feature information and the related information of the redundant feature matrix through decoding processing at the decoding end; At the decoding end, third feature information is generated according to the second feature information and related information of the redundant feature matrix, for processing the image to be processed.

3. The transmission method according to claim 2, wherein: The decoding end generates the third feature information according to the second feature information and the related information of the redundant feature matrix, including: Generate a corresponding number of all-0 matrices according to the relevant information of the redundant feature matrix, and determine the channel corresponding to the redundant feature matrix; According to the corresponding channel, the all-0 matrix is ​​inserted into the second feature information to generate the third feature information.

4. The transmission method according to any one of claims 1 to 3, wherein: The step of transmitting the encoded second feature information and the related information of the redundant feature matrix to the decoding end comprises: The second feature information and the related information of the redundant feature matrix are quantized and encoded, and then transmitted to the decoding end.

5. The transmission method according to claim 4, further comprising The decoding end performs decoding processing and inverse quantization processing to obtain the second feature information and related information of the redundant feature matrix.

6. A device for transmitting image features, comprising: An extraction unit, used to extract first feature information of the image to be processed by using a machine learning model, wherein the first feature information includes a feature matrix of each channel; A determination unit, used for determining a redundant feature matrix according to the information amount of each feature matrix; A first generating unit, configured to delete the redundant feature matrix from the first feature information to generate second feature information; a transmission unit, configured to transmit the encoded second feature information and related information of the redundant feature matrix to a decoding end, The determining unit is configured to perform at least one of the following: The sum of the eigenvalues ​​or the mean of the eigenvalues ​​in each of the characteristic matrices is calculated respectively as the information content of each of the characteristic matrices, and the characteristic matrix with an information content of 0 is determined as the redundant characteristic matrix; Sorting the feature matrices whose information amount is not 0 in order of small to large information amount, and determining the feature matrices whose sequence numbers are less than or equal to the sequence number threshold and the feature matrices whose information amount is 0 as the redundant feature matrices; The sum of the eigenvalues ​​or the mean of the eigenvalues ​​in each of the feature matrices is calculated respectively as the information content of each of the feature matrices, and the feature matrix whose information content is less than or equal to the information content threshold is determined as the redundant feature matrix.

7. The transmission device according to claim 6, further comprising: An acquiring unit, configured to acquire the second feature information and related information of the redundant feature matrix through decoding processing at the decoding end; The second generating unit is used to generate third feature information at the decoding end according to the second feature information and related information of the redundant feature matrix, so as to process the image to be processed.

8. The transmission device according to claim 7, wherein: The second generating unit generates a corresponding number of all-0 matrices according to the relevant information of the redundant feature matrix, determines the channels corresponding to the redundant feature matrix, and inserts the all-0 matrices into the second feature information according to the corresponding channels to generate the third feature information.

9. The transmission device according to any one of claims 6 to 8, wherein: The transmission unit performs quantization processing and encoding processing on the second feature information and related information of the redundant feature matrix, and then transmits the result to the decoding end.

10. The transmission device according to claim 9, further comprising: An acquisition unit is used to perform decoding processing and inverse quantization processing at the decoding end to obtain the second feature information and related information of the redundant feature matrix.

11. A system for transmitting image features, comprising: The encoding end is used to extract first feature information of the image to be processed by using a machine learning model, where the first feature information includes a feature matrix of each channel, determine a redundant feature matrix according to the information amount of each feature matrix, delete the redundant feature matrix from the first feature information, generate second feature information, and transmit the encoded second feature information and related information of the redundant feature matrix to the decoding end; A decoding end is used to obtain the second feature information and the related information of the redundant feature matrix through decoding processing, and generate third feature information at the decoding end according to the second feature information and the related information of the redundant feature matrix for processing the image to be processed, The encoding end performs at least one of the following: The sum of the eigenvalues ​​or the mean of the eigenvalues ​​in each of the characteristic matrices is calculated respectively as the information content of each of the characteristic matrices, and the characteristic matrix with an information content of 0 is determined as the redundant characteristic matrix; Sorting the feature matrices whose information amount is not 0 in order of small to large information amount, and determining the feature matrices whose sequence numbers are less than or equal to the sequence number threshold and the feature matrices whose information amount is 0 as the redundant feature matrices; The sum of the eigenvalues ​​or the mean of the eigenvalues ​​in each of the feature matrices is calculated respectively as the information content of each of the feature matrices, and the feature matrix whose information content is less than or equal to the information content threshold is determined as the redundant feature matrix.

12. A device for transmitting image features, comprising: Memory; and A processor coupled to the memory, wherein the processor is configured to execute the method for transmitting image features according to any one of claims 1 to 5 based on instructions stored in the memory.

13. A non-volatile computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the method for transmitting image features as described in any one of claims 1 to 5 is implemented.

Images