A method to improve the ELAN module in the YOLOv7 algorithm

Abstract

The present invention belongs to the field of target recognition, specifically relating to an improved method, system, and device for the ELAN module in the YOLOv7 algorithm. This method aims to address the existing art's unsatisfactory target recognition performance and the frequent occurrence of speckle, missing structures, geometric distortion, and shadows. The method comprises: obtaining the ConCat data K from the ELAN module, performing convolution, normalization, activation function processing, and global response normalization on the data; multiplying the convolved data with a preset learnable parameter; performing a residual connection on the data multiplied by the learnable parameter and the data after global response normalization, and using the residual-processed data as the output of the ELAN module. The present invention improves target detection speed and accuracy, maximizes the preservation of original information features, and is real-time, intensive, and efficient, thus offering excellent practical value.

Description

Technical Field

[0001] The present invention belongs to the field of target recognition, and in particular relates to a method, system and device for improving the ELAN module in the YOLOv7 algorithm. Background Art

[0002] Object recognition is a key research area in computer vision, and its development is crucial for improving human-computer interaction, intelligent robotics, autonomous driving, security surveillance, and other fields. However, traditional object recognition suffers from low efficiency, high repetition, and inadequate performance. With the development of virtual space, artificial intelligence, and software-defined image processing, the use of deep learning to identify ship targets in marine remote sensing imagery has become a major research area in remote sensing. However, some mainstream algorithmic models offer unsatisfactory recognition results, and effective solutions are lacking for common problems such as coherent speckle, missing structures, geometric distortion, and shadows. Improving object detection performance, enhancing the target recognition dimension, detection accuracy, and granularity are pressing challenges.

[0003] YOLOv7, a new algorithm, provides a solution to these challenges. To avoid the loss of effective information after convolution of small targets like SAR, a residual module is added to the corresponding position, preserving the original information to the greatest extent possible while extracting effective information.

[0004] Guided by the idea of ​​fusing high-level information with low-level network features, Y. Chen, J. Yu, and others improved the accuracy of SAR ship target detection by adding deconvolution and prediction modules to the SSD model (reference: Y. Chen, J. Yu, and Y. Xu, "SAR Ship Target Detection for SSDv2 under Complex Backgrounds," 2020 International Conference on Computer Vision, Image and Deep Learning (CVIDL), Chongqing, China, 2020, pp. 560-565, doi:0.1109 / CVIDL51233.2020.00-27). However, the network structure used in SSD is relatively shallow, consisting of only a few basic modules such as convolutional and pooling layers. This may not fully extract the complex features in SAR images, resulting in relatively low accuracy. Because SAR images typically exhibit complex multi-scale, multi-directional, and multi-angle features, a deeper and more complex network structure is required to extract these complex features. To address the real-time detection of SAR spaceborne equipment, X.Xu, X.Zhang, and others proposed a lightweight target detection model by replacing traditional convolutional modules with lightweight convolutional modules. (Reference: X.Xu, X.Zhang, T.Zhang, J.Shi, S.Wei, and J.Li, "On-Board Ship Detection in SAR Images Based on L-YOLO," 2022 IEEE Radar Conference (RadarConf22), New York City, NY, USA, 2022, pp. 1-5, doi:10.1109 / RadarConf2248738.2022.9764190.) However, this lightweight approach compromises some detection performance, compromising detection accuracy. To address the issue of the coupling head in YOLOv4 degrading its detection performance, Q.Li, D.Xiao, and others decoupled target detection from regression and classification tasks, simplifying the network structure and improving target detection performance.(Reference: Q.Li, D.Xiao and F.Shi,"A Decoupled Head and Coordinate Attention Detection Method for Ship Targets in SAR Images,"in IEEE Access, vol. 10, pp. 128562-128578, 2022, doi: 10.1109 / ACCESS.2022.3222364.) However, due to the inherent network structure and norm of YOLOv4, feature extraction needs to be strengthened and detection accuracy needs to be improved. Based on this, the present invention proposes an improvement method for the ELAN module in the YOLOv7 algorithm. Summary of the Invention

[0005] In order to solve the above-mentioned problems in the prior art, namely, the recognition effect of the target recognition in the prior art is unsatisfactory, and the problems of coherent speckle, structure loss, geometric distortion, and shadow often occur, the present invention provides a method, system and device for improving the ELAN module in the YOLOv7 algorithm.

[0006] In one aspect of the present invention, a method for improving the ELAN module in the YOLOv7 algorithm is proposed, wherein:

[0007] Step S1, obtaining the ConCat data K in the ELAN module, and performing convolution processing on the K to obtain first data a;

[0008] Step S2: normalize the first data a to obtain second data b;

[0009] Step S3: Processing the second data b through an activation function to obtain third data m; the activation function includes a GELU activation function;

[0010] Step S4: performing global response normalization processing on the third data m to obtain fourth data d;

[0011] Step S5: multiplying the first data a by a preset learnable parameter to obtain fifth data e;

[0012] Step S6: Perform residual connection processing on the fourth data d and the fifth data e, and use the data after residual processing as the output of the ELAN module.

[0013] In some preferred embodiments, the first data a is obtained by:

[0014] a=Conv(K)=K∈R H*W*4C →K∈RH*W*C ;

[0015] Wherein, H is the length of the data of K before ConCat processing, W is the width of the data of K before ConCat processing, and C is the number of channels.

[0016] In some preferred embodiments, the method for obtaining the second data b is:

[0017] Use LayerNorm to normalize it;

[0018] b=LN(a)=LayerNorm(a).

[0019] In some preferred embodiments, the method for obtaining the third data m is:

[0020] b~N(0,1);

[0021] in, X is the input feature value, X follows a Gaussian distribution, P(X<=b) is the probability that X is less than or equal to b; the approximate calculation formula is:

[0022]

[0023] In some preferred embodiments, the method for obtaining the fourth data d is:

[0024] Step S41: perform L2 norm processing on the m, and the calculation formula is:

[0025] G(m)={||m1||,||m2||,...,||m C ||}∈R C ;

[0026] Step S42: normalize the m after L2 norm processing by standard division, and the calculation formula is:

[0027]

[0028] Among them, the m i is the input feature value in the i-th channel, the m j To accumulate the input characteristic values ​​in channels 1 to C.

[0029] Step S42: calibrate the normalized data to obtain m. The calculation formula is:

[0030] d=m i =g*m i *N(G(m))+n+m i∈R;

[0031] Among them, g is the scaling factor and n is the deviation factor.

[0032] In some preferred embodiments, the method for obtaining the fifth data e is:

[0033] e=ResNet(a)=l*a;

[0034] Wherein, l is a preset learnable parameter.

[0035] In some preferred implementations, the fourth data d is residually connected with the fifth data e by:

[0036] Output = d + e.

[0037] In another aspect of the present invention, a system for improving the ELAN module in the YOLOv7 algorithm is proposed, wherein the system includes: a convolution module, a normalization module, an activation module, a GRN processing module, a learning module, and a residual connection module;

[0038] The convolution module is configured to obtain the data K after ConCat in the ELAN module and perform convolution processing on the K to obtain the first data a;

[0039] The normalization module is configured to perform normalization processing on the first data a to obtain second data b;

[0040] The activation module is configured to process the second data b through an activation function to obtain third data m; the activation function includes a GELU activation function;

[0041] The GRN processing module is configured to perform global response normalization processing on the third data m to obtain fourth data d;

[0042] The learning module is configured to multiply the first data a by a preset learnable parameter to obtain fifth data e;

[0043] The residual connection module is configured to perform residual connection processing on the fourth data d and the fifth data e, and use the data after residual processing as the output of the ELAN module.

[0044] A third aspect of the present invention provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor;

[0045] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the above-mentioned method for improving the ELAN module in the YOLOv7 algorithm.

[0046] In a fourth aspect of the present invention, a computer-readable storage device is proposed, wherein the computer-readable storage device stores computer instructions, and the computer instructions are used to be executed by the computer to implement the above-mentioned method for improving the ELAN module in the YOLOv7 algorithm.

[0047] Beneficial effects of the present invention:

[0048] The present invention is mainly used for target detection under complex scene conditions in SAR remote sensing images, and can also be expanded to remote sensing related fields. Based on the improvement of the ELAN module of the YOLOv7 algorithm, in particular, the introduction of the GELU activation function probability regularization, the expected value is output in a probabilistic manner for the input neuron. At the same time, since its derivative is continuous, the gradient disappearance is reduced during the training process, which helps to accelerate the convergence of the model. The improvement avoids the loss of effective information caused by the convolution of small SAR targets, improves the target detection speed and accuracy, retains the original information characteristics to the greatest extent, and has the characteristics of real-time, intensive, and efficient. The extracted effective information can be used in military reconnaissance of the enemy's strategic deployment and activity information of port terminals, and can also provide effective reference for maritime transportation management in civilian applications, and has good use value. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Other features, objects and advantages of the present application will become more apparent upon reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0050] Figure 1 This is a flow chart of a method for improving the ELAN module in the YOLOv7 algorithm according to the present invention;

[0051] Figure 2 The present invention is a method for improving the ELAN module in the YOLOv7 algorithm.

[0052] Figure 3 This is a flow chart comparing the improved ELAN and the ELAN before improvement in a method for improving the ELAN module in the YOLOv7 algorithm according to the present invention;

[0053] Figure 4 This is a schematic diagram of experimental results of an improved method for the ELAN module in the YOLOv7 algorithm of the present invention;

[0054] Figure 5 This is an experimental result analysis table of an improved method for the ELAN module in the YOLOv7 algorithm of the present invention;

[0055] Figure 6It is a structural diagram of a computer system of a server for implementing the method, system, and device embodiments of the present application. DETAILED DESCRIPTION

[0056] The present application will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are intended only to illustrate the relevant invention and are not intended to limit the invention. It should also be noted that, for ease of description, only portions relevant to the relevant invention are shown in the accompanying drawings.

[0057] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0058] See also Figure 1-5 As shown, a method for improving the ELAN module in the YOLOv7 algorithm according to the first embodiment of the present invention includes the following steps:

[0059] Step S1, obtaining the ConCat data K in the ELAN module, and performing convolution processing on the K to obtain first data a;

[0060] Step S2: normalize the first data a to obtain second data b;

[0061] Step S3: Processing the second data b through an activation function to obtain third data m; the activation function includes a GELU activation function;

[0062] Step S4: performing global response normalization processing on the third data m to obtain fourth data d;

[0063] Step S5: multiplying the first data a by a preset learnable parameter to obtain fifth data e;

[0064] Step S6: Perform residual connection processing on the fourth data d and the fifth data e, and use the data after residual processing as the output of the ELAN module.

[0065] Preferably, the method for obtaining the first data a is:

[0066] a=Conv(K)=K∈R H*W*4C →K∈R H*W*C ;

[0067] Wherein, H is the length of the data of K before ConCat processing, W is the width of the data of K before ConCat processing, and C is the number of channels.

[0068] Preferably, the method for obtaining the second data b is:

[0069] Use LayerNorm to normalize it;

[0070] b=LN(a)=LayerNorm(a).

[0071] Preferably, the method for obtaining the third data m is:

[0072] b~N(0,1);

[0073] in, X is the input feature value, X follows a Gaussian distribution, P(X<=b) is the probability that X is less than or equal to b; the approximate calculation formula is:

[0074]

[0075] Preferably, the method for obtaining the fourth data d is:

[0076] Step S41: perform L2 norm processing on the m, and the calculation formula is:

[0077] G(m)={||m1||,||m2||,...,||m C ||}∈R C ;

[0078] Step S42: normalize the m after L2 norm processing by standard division, and the calculation formula is:

[0079]

[0080] Among them, the m i is the input feature value in the i-th channel, the m j To accumulate the input characteristic values ​​in channels 1 to C.

[0081] Step S42: calibrate the normalized data to obtain m. The calculation formula is:

[0082] d=m i =g*m i *N(G(m))+n+m i ∈R;

[0083] Among them, g is the scaling factor and n is the deviation factor.

[0084] Preferably, the method for obtaining the fifth data e is:

[0085] e=ResNet(a)=l*a;

[0086] Wherein, l is a preset learnable parameter.

[0087] Preferably, the fourth data d and the fifth data e are subjected to residual connection, and the method is:

[0088] Output = d + e.

[0089] Typical Example 1 of the present invention: an improved method for the ELAN module based on the YOLOv7 algorithm, see Figure 2 、 3 ,4,5,Experimental results analysis table:

[0090] To verify the experimental effect of the improved algorithm in this paper, we conducted ablation experiments on the original YOLOv7 network model with the ELAN module in different network layers replaced by the Res-ELAN module.

[0091] Basic approach: Based on the SIRSDD dataset, the Res-ELAN network module is added to YOLOv7, and a comparative ablation experiment is performed by replacing the 2nd, 3rd, and 4th ELAN modules in the Head layer and the 2nd, 3rd, and 4th ELAN modules in the Body layer.

[0092] Experimental results show that replacing the ELAN modules in different network layers can produce different improvements. Replacing the 2nd, 3rd, and 4th ELAN modules in the head and the 3rd ELAN module in the body with the Res-ELAN module achieves the best results, with mAP_0.5 improving by approximately 0.7 percentage points and mAP_0.5:0.95 improving by approximately 1.58%, both of which are significant improvements.

[0093] Although the various steps in the above embodiment are described in the above-mentioned order, those skilled in the art will understand that in order to achieve the effect of this embodiment, different steps do not have to be executed in such an order. They can be executed simultaneously (in parallel) or in a reverse order. These simple changes are within the scope of protection of the present invention.

[0094] A target detection method according to a second embodiment of the present invention is applied to a YOLOv7 algorithm model. The YOLOv7 algorithm includes a convolutional layer, an ELAN module, and a pooling layer. The method includes:

[0095] Step A1, obtaining an image to be subjected to target detection as an input image;

[0096] Step A2: inputting the input image into the convolution layer of the YOLOv7 algorithm model to obtain convolution data;

[0097] Step A3: input the convolution data into the ELAN module of the YOLOv7 algorithm model, obtain the ConCat data K in the ELAN module, and perform convolution processing on the X to obtain the first data a;

[0098] Step A4: normalize the first data a to obtain second data b;

[0099] Step A5: Processing the second data b through an activation function to obtain third data m; the activation function includes a GELU activation function;

[0100] Step A6: performing global response normalization processing on the third data m to obtain fourth data d;

[0101] Step A7: multiplying the first data a by a preset learnable parameter to obtain fifth data e;

[0102] Step A8: performing residual connection processing on the fourth data d and the fifth data e, and using the data after residual processing as the output of the ELAN module;

[0103] In step A9, the output of step A8 is input to the pooling layer to obtain the target detection result corresponding to the input image.

[0104] See also Figure 1 , a third embodiment of the present invention is a system for improving the ELAN module in the YOLOv7 algorithm, the system comprising: a convolution module, a normalization module, an activation module, a GRN processing module, a learning module, and a residual connection module;

[0105] The convolution module is configured to obtain the data K after ConCat in the ELAN module and perform convolution processing on the K to obtain the first data a;

[0106] The normalization module is configured to perform normalization processing on the first data a to obtain second data b;

[0107] The activation module is configured to process the second data b through an activation function to obtain third data m; the activation function includes a GELU activation function;

[0108] The GRN processing module is configured to perform global response normalization processing on the third data m to obtain fourth data d;

[0109] The learning module is configured to multiply the first data a by a preset learnable parameter to obtain fifth data e;

[0110] The residual connection module is configured to perform residual connection processing on the fourth data d and the fifth data e, and use the data after residual processing as the output of the ELAN module.

[0111] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working process and related instructions of the system described above can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0112] It should be noted that the above embodiment provides an improved system for the ELAN module in the YOLOv7 algorithm, which is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be further decomposed or combined. For example, the modules in the above embodiments can be combined into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the modules or steps and are not to be regarded as improper limitations of the present invention.

[0113] An electronic device according to a fourth embodiment of the present invention includes:

[0114] at least one processor; and

[0115] a memory communicatively connected to at least one of the processors; wherein,

[0116] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the above-mentioned method for improving the ELAN module in the YOLOv7 algorithm.

[0117] A computer-readable storage medium according to a fifth embodiment of the present invention stores computer instructions, which are used to be executed by the computer to implement the above-mentioned method for improving the ELAN module in the YOLOv7 algorithm.

[0118] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes and related instructions of the storage device and processing device described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0119] Those skilled in the art should be able to appreciate that, in conjunction with the modules and method steps of each example described in the embodiments disclosed herein, it is possible to implement them with electronic hardware, computer software, or a combination of the two, and the programs corresponding to the software modules and method steps can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art. In order to clearly illustrate the interchangeability of electronic hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0120] Reference below Figure 6 , which shows a structural diagram of a computer system of a server for implementing the method, system, and device embodiments of the present application. Figure 6 The server shown is merely an example and should not limit the functions and scope of use of the embodiments of the present application.

[0121] like Figure 6 As shown, the computer system includes a central processing unit (CPU) 601, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 602 or the program loaded from the storage part 608 to the random access memory (RAM) 603. Various programs and data required for system operation are also stored in the RAM 603. The CPU 601, ROM 602 and RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0122] The following components are connected to the I / O interface 605: an input section 606 including a keyboard, a mouse, and the like; an output section 607 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and a speaker; a storage section 608 including a hard disk and the like; and a communication section 609 including a network interface card such as a LAN (Local Area Network) card or a modem. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as needed. Removable media 611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, and the like, are installed in the drive 610 as needed so that computer programs read therefrom can be installed into the storage section 608 as needed.

[0123] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes 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 such an embodiment, the computer program can be downloaded and installed from the network through the communication part 609, and / or installed from the removable medium 611. When the computer program is executed by the central processing unit (CPU) 601, the above-mentioned functions defined in the method of the present application are executed. It should be noted that the computer-readable medium mentioned above in the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can 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 having one or more conductors, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal may take a variety of forms, including, but not limited to, electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code embodied on a computer-readable medium may be transmitted using any suitable medium, including, but not limited to, wireless, wire, optical cable, RF, etc., or any suitable combination thereof.

[0124] Computer program code for performing the operations of the present application 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 stand-alone 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 through 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., through the Internet using an Internet service provider).

[0125] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a module, program segment or a part of code, and the module, program segment or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two 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 box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented by a dedicated hardware-based system that performs the specified function or operation, or can be implemented by a combination of dedicated hardware and computer instructions.

[0126] The terms "first", "second", etc. are used to distinguish similar objects, rather than to describe or indicate a particular order or sequence.

[0127] The term "comprise" or any other similar term is intended to cover non-exclusive inclusion such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0128] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

Claims

1. A method for improving the ELAN module in the YOLOv7 algorithm, characterized by: The method comprises the following steps: Step S1, obtaining the ConCat data K in the ELAN module, and performing convolution processing on the K to obtain first data a; Step S2: normalize the first data a to obtain second data b; Step S3: Processing the second data b through an activation function to obtain third data m; the activation function includes a GELU activation function; Step S4: Perform global response normalization processing on the third data m to obtain fourth data d: Step S41: perform L2 norm processing on the m, and the calculation formula is: G(m)={‖m1‖,‖m2‖,……,‖m C ‖}∈R C ; Step S42: normalize the m after L2 norm processing by standard division, and the calculation formula is: Among them, the m i is the input feature value in the i-th channel, the m j To accumulate the input characteristic values ​​in channels 1 to C; Step S43: calibrate the normalized data to obtain m. The calculation formula is: d=m i =g * m i *N(G(m))+n+m i ∈R; Among them, g is the scaling factor and n is the deviation factor; Step S5: multiplying the first data a by a preset learnable parameter to obtain fifth data e; Step S6: Perform residual connection processing on the fourth data d and the fifth data e, and use the data after residual processing as the output of the ELAN module.

2. The method for improving the ELAN module in the YOLOv7 algorithm according to claim 1, wherein: The method for obtaining the first data a is as follows: a=Conv(K)=K∈R H*W*4C →K∈R H*W*C ; Wherein, H is the length of the data of K before ConCat processing, W is the width of the data of K before ConCat processing, and C is the number of channels.

3. The method for improving the ELAN module in the YOLOv7 algorithm according to claim 1, wherein: The method for obtaining the second data b is: Use LayerNorm to normalize it; b=LN(a)=LayerNorm(a).

4. The method for improving the ELAN module in the YOLOv7 algorithm according to claim 1, wherein: The method for obtaining the third data m is: in, X is the input feature value, X follows a Gaussian distribution, P(X<=b) is the probability that X is less than or equal to b; the approximate calculation formula is:

5. The method for improving the ELAN module in the YOLOv7 algorithm according to claim 1, wherein: The method for obtaining the fifth data e is: e=ResNet(a)=l*a; Wherein, l is a preset learnable parameter.

6. The method for improving the ELAN module in the YOLOv7 algorithm according to claim 1, wherein: The fourth data d is residually connected with the fifth data e, and the method is: Output = d + e.

7. A system for improving the ELAN module in the YOLOv7 algorithm, based on the method for improving the ELAN module in the YOLOv7 algorithm according to any one of claims 1 to 6, characterized in that: The system includes: a convolution module, a normalization module, an activation module, a GRN processing module, a learning module, and a residual connection module; The convolution module is configured to obtain the data K after ConCat in the ELAN module and perform convolution processing on the K to obtain the first data a; The normalization module is configured to perform normalization processing on the first data a to obtain second data b; The activation module is configured to process the second data b through an activation function to obtain third data m; the activation function includes a GELU activation function; The GRN processing module is configured to perform global response normalization processing on the third data m to obtain fourth data d: The L2 norm processing is performed on the m, and the calculation formula is: G(m)={‖m1‖,‖m2‖,……,‖m C ‖}∈R C ; The m after L2 norm processing is normalized by standard division, and the calculation formula is: Among them, the m i is the input feature value in the i-th channel, the m j To accumulate the input characteristic values ​​in channels 1 to C; The normalized data is used to calibrate m, and the calculation formula is: d=m i =g * m i *N(G(m))+n+m i ∈R; Among them, g is the scaling factor and n is the deviation factor; The learning module is configured to multiply the first data a by a preset learnable parameter to obtain fifth data e; The residual connection module is configured to perform residual connection processing on the fourth data d and the fifth data e, and use the data after residual processing as the output of the ELAN module.

8. An electronic device, characterized in that: include: at least one processor; and a memory communicatively coupled to at least one of said processors; The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the method for improving the ELAN module in the YOLOv7 algorithm according to any one of claims 1 to 6.

9. A computer-readable storage device, characterized in that: The computer-readable storage device stores computer instructions, and the computer instructions are used to be executed by the computer to implement the method for improving the ELAN module in the YOLOv7 algorithm according to any one of claims 1 to 6.

Images