A target detection method and system based on multi-scale mixed dimension network

Abstract

The present invention belongs to the technical field of target recognition, and provides a target detection method and system based on a multi-scale mixed dimensional network. The method comprises: inputting a target image into a trained multi-scale mixed dimensional network, removing redundant frames from the obtained image by an NMS method, and obtaining an image marked with a target position and category; the multi-scale mixed dimensional network comprises a pyramid network, and a mixed dimensional module is introduced between every two convolutional layers from top to bottom of the pyramid network.

Description

Technical Field

[0001] The present invention belongs to the technical field of target recognition, and in particular relates to a target detection method and system based on a multi-scale mixed dimensional network. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Traffic sign detection is a hot topic in the current field of computer vision and pattern recognition. Intelligent transportation and unmanned driving technologies have high requirements for accuracy and real-time performance, which brings great difficulty to the research of traffic sign detection technology. Traditional traffic sign recognition algorithms generally use machine learning algorithms and image processing technologies to extract features and classify and regress the color, shape, and edges of images, such as HOG and SIFT. However, due to their shortcomings such as sensitivity to noise and large amount of calculation, they generally have low adaptability, poor detection effect, and imbalance between real-time performance and accuracy.

[0004] With the rapid development of deep learning, although the object detection algorithm based on convolutional neural network has richer feature information and stronger feature expression ability, it still needs to be continuously improved and perfected to achieve better detection effect. The Faster-RCNN algorithm completes classification and regression by alternately training SPPNET and RPN networks. Since the selective search method is used to act on the candidate box and repeated convolution is performed for network calculation, it leads to large memory usage and slow speed. The SSD algorithm uses multi-scale feature layers for convolution detection. Although the speed is improved to a certain extent, the low-level feature layer with a small number of convolution layers is used, resulting in insufficient feature extraction. The YOLOv1 algorithm uses image segmentation to output position and category, making full use of feature information, but a cell can only predict two boxes and one category, limiting the number of predictions. The YOLOv9000 algorithm uses batch normalization, adds BN layer, and introduces anchor boxes to make the network more stable, but due to label restrictions, the detection performance is low. The YOLOv3 algorithm uses an FPN network and adds NMS screening. It has better detection performance in terms of speed and accuracy. However, due to the small pixels and low resolution, the feature information of small targets is not obvious after convolution. Therefore, there is still a lot of room for improvement in the research on traffic sign detection with a large number of small target objects. Designing a high-accuracy and high-efficiency small target detection algorithm is still a meaningful open problem. Summary of the invention

[0005] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a target detection method and system based on a multi-scale mixed dimensional network, which designs and reconstructs the structure of a feature pyramid network (FPN), constructs a multi-scale mixed domain network by screening the feature fusion target detection layer scale, improves the learning ability of high-value information, and can solve the problems of low detection accuracy, missed detection, false detection, etc. of the original algorithm in the presence of small target objects, and the model is more stable and has higher robustness.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] The first aspect of the present invention provides a target detection method based on a multi-scale mixed dimensional network.

[0008] A target detection method based on a multi-scale mixed dimension network, comprising:

[0009] The target image is input into the trained multi-scale mixed dimension network. After the redundant frames are removed by the NMS method, an image with the target position and category is obtained.

[0010] The multi-scale mixed dimension network includes a pyramid network, and a mixed dimension module is introduced between every two convolutional layers from top to bottom in the pyramid network.

[0011] Furthermore, the mixed dimension module includes a channel attention module and a spatial attention module;

[0012] The channel attention module is configured to: take the features of the previous convolutional layer in the pyramid network as input, pass the feature map after global maximum and average pooling to the shared fully connected layer MLP, add the output features and perform sigmoid activation to obtain a weighted feature layer;

[0013] The spatial attention module is configured to: use the feature layer obtained by the channel attention module as the input feature of the spatial module, perform tensor splicing, then perform dimensionality reduction through convolution operation, generate a spatial feature layer through activation operation, and finally multiply the feature layer with the input feature to obtain the final generated feature as the input of the next convolution layer in the pyramid network.

[0014] Furthermore, the pyramid network includes: five residual modules and an output layer, wherein the five residual modules are input to the output layer through feature fusion;

[0015] The numbers of residual blocks in the five residual modules are 1, 2, 4, 4, and 23 respectively;

[0016] The five residual modules respectively output a 2x downsampling layer, a 4x downsampling layer, an 8x downsampling layer, a 16x downsampling layer, and a 32x downsampling layer, and the 4x downsampling layer is input into the output layer through feature fusion;

[0017] The 8x downsampling layer is fused with the features of the 4x downsampling layer after 2x upsampling; the 16x downsampling layer is fused with the features of the 8x downsampling layer after 2x upsampling, and also fused with the features of the 4x upsampling and 4x downsampling layer; the 32x downsampling layer is fused with the features of the 16x downsampling layer after 2x upsampling, and also fused with the features of the 8x downsampling layer after 4x upsampling;

[0018] The output layer includes two residual blocks and one convolutional layer.

[0019] Furthermore, the training process of the multi-scale mixed dimensional network includes: obtaining a target image detection training set; obtaining the anchor frame coordinates of the target image detection training set through a KMA algorithm; and training the multi-scale mixed dimensional network based on the target image detection training set using cloud computing resources to obtain a target image detection network.

[0020] The second aspect of the present invention provides a target detection system based on a multi-scale mixed dimensional network.

[0021] A target detection system based on a multi-scale mixed dimension network, comprising:

[0022] The detection module is configured to: input the target image into the trained multi-scale mixed dimension network, remove the redundant frames of the obtained image by the NMS method, and obtain an image marked with the target position and category;

[0023] The network building module is configured as: the multi-scale mixed dimension network pyramid network, wherein a mixed dimension module is introduced between every two convolutional layers from top to bottom of the pyramid network.

[0024] A third aspect of the present invention provides a computer-readable storage medium.

[0025] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the target detection method based on a multi-scale mixed dimensional network as described in the first aspect above.

[0026] A fourth aspect of the present invention provides a computer device.

[0027] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the target detection method based on a multi-scale mixed dimensional network as described in the first aspect above are implemented.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] The present invention can correctly locate the position and category of the target object and has better generalization ability. It can not only effectively solve the problems of false detection and missed detection, but also has stronger anti-interference ability and practicality, and performs well in an environment with complex and changeable data.

[0030] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0032] Figure 1 It is a flow chart of the target detection method based on multi-scale mixed dimension network of the present invention;

[0033] Figure 2 It is a multi-scale network structure diagram of the present invention;

[0034] Figure 3 It is the overall structure diagram of the mixed dimension attention module of the present invention;

[0035] Figure 4 It is a multi-scale mixed-dimensional network graph of the present invention;

[0036] Figure 5 This is the multi-target detection flow chart of the present invention

[0037] Figure 6 This is a comparison of the P, R, F1 and PR change curves of the network model in the embodiment of the present invention.

[0038] FIG. 7( a ) is an original image in dark weather according to an embodiment of the present invention;

[0039] FIG7( b ) is a picture processed by the YOLOv3 network in dark weather according to an embodiment of the present invention;

[0040] FIG7( c ) is a picture processed by the network of the present invention under dark weather conditions in an embodiment of the present invention;

[0041] FIG8( a ) is an original image with dense objects in an embodiment of the present invention;

[0042] FIG8( b ) is a picture processed by the YOLOv3 network under dense target conditions in an embodiment of the present invention;

[0043] FIG8( c ) is a picture processed by the network of the present invention under dense target conditions in an embodiment of the present invention;

[0044] FIG9( a ) is an original image under object occlusion in an embodiment of the present invention;

[0045] FIG9( b ) is a picture processed by the YOLOv3 network under object occlusion in an embodiment of the present invention;

[0046] FIG9( c ) is a picture processed by the network of the present invention under object occlusion in an embodiment of the present invention;

[0047] FIG10( a ) is an original image of a target that is far away according to an embodiment of the present invention;

[0048] FIG10( b ) is a picture processed by the YOLOv3 network when the target is far away in an embodiment of the present invention;

[0049] FIG10( c ) is a picture processed by the network of the present invention when the target is far away in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0051] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0052] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0053] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the methods and systems according to various embodiments of the present disclosure. It should be noted that each box in the flowchart or block diagram can represent a module, a program segment, or a part of a code, and the module, program segment, or a part of a code may include one or more executable instructions for implementing the logical functions specified in each embodiment. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, or they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the flowchart and / or block diagram, and the combination of boxes in the flowchart and / or block diagram can be implemented using a dedicated hardware-based system that performs a specified function or operation, or can be implemented using a combination of dedicated hardware and computer instructions.

[0054] Embodiment 1

[0055] like Figure 1 As shown, this embodiment provides a target detection method based on a multi-scale mixed dimensional network. This embodiment uses the method applied to a server as an example for illustration. It can be understood that the method can also be applied to a terminal, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network servers, cloud communications, middleware services, domain name services, security services CDN, and big data and artificial intelligence platforms. The terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited to this. The terminal and the server can be directly or indirectly connected via wired or wireless communication, which is not limited in this application. In this embodiment, the method includes the following steps:

[0056] S101: Input the target image into the trained multi-scale mixed dimension network, and remove the redundant frames of the obtained image by the NMS method to obtain an image marked with the target position and category;

[0057] S102: The multi-scale mixed dimension network includes a pyramid network, and a mixed dimension module is introduced between every two convolutional layers from top to bottom of the pyramid network.

[0058] Specifically, the specific solution of this embodiment is as follows:

[0059] In the detection technology for small targets, the common practice is to continuously add shallow feature information to form a four-scale or higher-scale detection layer. Although the detection performance of small targets is improved to a certain extent, simply adding a higher scale will lead to a significant increase in the amount of calculation and a slowdown in speed. This embodiment proposes a design method based on a multi-scale hybrid domain network. The multi-scale network structure is as follows: Figure 2 As shown. Based on the multi-scale hybrid domain network, Darknet-53 is used as the backbone network, and the ResNet structure is used to extract features. The high-scale feature layer is upsampled and the shallow-scale splicing is performed to form a high-information feature layer. Compared with the original YOLOv3 network structure, this design performs tensor splicing on the 8-fold downsampled feature map output by the original network and the feature layer of six convolutions plus sampling, expands the tensor dimension, realizes the feature pyramid structure of feature fusion, and enhances the detection effect of objects of different sizes and occluded objects. The spliced ​​feature layer is then convolved six times and sampled to extract effective features while reducing the number of network parameters. A new 104*104 scale layer is added, that is, a large-scale feature map is superimposed to enhance the ability to pay attention to smaller targets. The feature feature information output by the fourth residual block from the end is added to fully tap the potential of the model in detecting smaller targets. At the same time, the 13*13 scale layer is removed, that is, the classification and regression prediction of the 4-fold downsampled feature layer is removed, aiming to avoid the increase in additional calculations caused by a single increase in scale and the omission of small targets caused by the excessive depth of the feature network. The previous three feature scales of 13*13, 26*26 and 52*52 are converted to three feature scales of 26*26, 52*52 and 104*104. In order to detect more fine-grained features, this embodiment screens the original YOLOv3 convolutional neural network structure, reconstructs the feature pyramid module, and the model network layer obtained by the target detection layer has stronger representation ability and more accurate regression ability than before.

[0060] That is, the pyramid network of YOLOv3 includes: five residual modules and an output layer, and the five residual modules are input to the output layer through feature fusion; the numbers of residual blocks in the five residual modules are 1, 2, 4, 4, and 23 respectively; the five residual modules output a 2x downsampling layer, a 4x downsampling layer, an 8x downsampling layer, a 16x downsampling layer, and a 32x downsampling layer respectively, and the 4x downsampling layer is input to the output layer through feature fusion; the 8x downsampling layer is fused with the 4x downsampling layer feature after 2x upsampling; the 16x downsampling layer is fused with the 8x downsampling layer feature after 2x upsampling, and is also fused with the 4x downsampling layer feature after 4x upsampling; the 32x downsampling layer is fused with the 16x downsampling layer feature after 2x upsampling, and is also fused with the 8x downsampling layer feature after 4x upsampling; the output layer includes two residual blocks and a convolutional layer.

[0061] In order to further enhance the neural network's ability to process information and extract features, this embodiment proposes a new feature attention mode, which uses a series method to successively integrate channel and spatial attention modules to capture and filter out high-value information, that is, after being independently embedded in the convolution layer of the existing backbone network, the learned weight distribution is applied to the original features. We call this new splicing structure a mixed dimension module. Figure 3 As shown in the figure, it is the overall structure of the mixed dimension module, which is mainly divided into two steps. First, the features of the convolution layer in the network architecture are used as input, and the feature map after global maximum and average pooling is passed to the shared fully connected layer MLP. The output features are summed and sigmoid activated to finally obtain the weighted weight, that is, the process of re-operating the one-dimensional vector after the feature map is compressed in the spatial dimension. What kind of features are considered as high-value information? At the same time, the feature layer after the channel module is multiplied by the input feature layer for adaptive feature optimization. Second, the feature layer obtained by the channel module is used as the input feature of the spatial module, and tensor splicing is performed. Then the convolution operation is performed for dimensionality reduction. The spatial feature layer is generated through the activation operation. Finally, the feature layer is multiplied with the input feature, that is, the channel dimension is compressed to obtain the final generated features. The focus is on the position of high-value features. The mixed dimension attention module is formed through adaptive splicing of the channel and spatial modules. The module expressions are shown in equations (1) and (2). σ is the sigmoid activation operation, f 7*7 Represents the convolution kernel size.

[0062] M c (F)=σ(MLP(AvgPool(F)+MLP(MaxPool(F))) (1)

[0063] M s (F) = σ(f 7*7 ([AvgPool(F);MaxPool(F)])) (2)

[0064] The hybrid dimensional attention module captures and filters out high-value information by emphasizing key areas and suppressing secondary areas, giving priority to feature extraction of effective features in key areas, thus reducing the amount of parameters and computation.

[0065] We describe the proposed multi-scale mixed dimension module (FPCS) in detail. This module mainly adaptively fuses and splices the three scale layers with the mixed dimension module. The splicing method is as follows: Figure 4As shown in the figure, a mixed dimension module is added to the original YOLOv3 pyramid module. This new splicing structure is called a multi-scale mixed dimension module (FPCS). By introducing a mixed domain mechanism from top to bottom after the convolution layer, it aims to generate a more expressive network model, which is not only more sensitive to identifying target objects, but also more flexible in the fluidity of network information. Conventional clustering algorithms such as K-Means are easily affected by the initial point, resulting in inaccurate clustering. Under the premise of multi-scale and mixed dimension adaptive splicing, this embodiment optimizes the initialization process of the cluster center, makes the points far away more likely to become the cluster center point, makes the selected initial point sufficiently discrete, and ensures that each cluster can cluster to a considerable amount of data during the iteration process. We call this new clustering algorithm the KMA algorithm. The KMA algorithm selects anchor frames that have a good match with the target object more accurately, analyzes and updates the anchor frames, modifies network parameters and prior frames of different scales, and thus makes the network model easier to learn and converge.

[0066] like Figure 5 As shown in the figure, it is a multi-target detection flow chart. First, before network training, the KMA clustering algorithm is used to re-cluster and analyze the data information, and the anchor frame with higher matching degree is selected. Then, Darknet-53 convolution down-sampling is performed, and the large-scale mixed dimension attention module is adaptively fused to increase the model expressiveness, which not only ensures higher accuracy and high resolution from the low level, but also integrates the richer semantic information of the high level, so that the neural network model is more suitable for the detection task of specific traffic signs. The final result uses NMS to remove redundant frames and output the position and category of the target image with high accuracy.

[0067] This embodiment uses the CCTSDB data set for experiments, performs format conversion and normalization operations on the data, and annotates it in the PASCAL VOC format. After building a Pytorch framework and comparing it with the original YOLOv3 algorithm, the values ​​of various evaluation indicators obtained are shown in the following table. Compared with the original algorithm, the network algorithm of this embodiment significantly improves the accuracy of the target, and the larger the IoU, the more obvious the algorithm advantage of this embodiment, and the more accurate the target bounding box positioning.

[0068] Table 1 Comparison of experimental results of different algorithms

[0069]

[0070] Table 2 Comparison of mAP values ​​at different IoU thresholds

[0071]

[0072] This example shows a visual comparison of the accuracy of the YOLOv3 algorithm and our algorithm in the figure below. Figure 6As shown, traffic sign samples in the test set are randomly selected for test verification to intuitively demonstrate the practicality of the improved model.

[0073] This embodiment uses the YOLOv3 network and the multi-scale mixed dimension network described in the present invention to process the original image in four situations: dark weather, dense targets, object occlusion, and distant targets. Figure 7(a)-Figure 10(c) It can be seen that the multi-scale mixed dimensional network proposed in the present invention can correctly locate the position and category of the target object, has better generalization ability, can not only effectively solve the problems of false detection and missed detection, but also has stronger anti-interference ability and practicality, and performs well in an environment with complex and changeable data.

[0074] Embodiment 2

[0075] This embodiment provides a target detection system based on a multi-scale mixed dimensional network.

[0076] A target detection system based on a multi-scale mixed dimension network, comprising:

[0077] The detection module is configured to: input the target image into the trained multi-scale mixed dimension network, remove the redundant frames of the obtained image by the NMS method, and obtain an image marked with the target position and category;

[0078] The network building module is configured as follows: the multi-scale mixed dimension network includes a pyramid network, and a mixed dimension module is introduced between every two convolutional layers from top to bottom of the pyramid network.

[0079] It should be noted that the above detection module and network construction module correspond to steps S101 to S102 in Embodiment 1, and the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the contents disclosed in Embodiment 1. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer executable instructions.

[0080] Embodiment 3

[0081] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps in the target detection method based on a multi-scale mixed dimensional network as described in the first embodiment above are implemented.

[0082] Embodiment 4

[0083] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the target detection method based on a multi-scale mixed dimensional network as described in the first embodiment are implemented.

[0084] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0085] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0086] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0087] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0088] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0089] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A target detection method based on a multi-scale mixed dimension network, characterized in that: include: The target image is input into the trained multi-scale mixed dimension network. After the redundant frames are removed by the NMS method, an image with the target position and category is obtained. The multi-scale mixed dimension network includes a pyramid network, and a mixed dimension module is introduced between every two convolutional layers from top to bottom of the pyramid network; The training process of the multi-scale mixed dimension network includes: obtaining a target image detection training set; obtaining the anchor frame coordinates of the target image detection training set through the KMA algorithm; based on the target image detection training set, using cloud computing resources to train the multi-scale mixed dimension network to obtain a target image detection network; The mixed dimension module includes a channel attention module and a spatial attention module; The spatial attention module is configured to: use the feature layer obtained by the channel attention module as the input feature of the spatial module, perform tensor splicing, then perform dimensionality reduction through convolution operation, generate a spatial feature layer through activation operation, and finally multiply the feature layer with the input feature to obtain the final generated feature as the input of the next convolution layer in the pyramid network.

2. The target detection method based on a multi-scale mixed dimension network according to claim 1 is characterized in that: The channel attention module is configured to: take the features of the previous convolutional layer in the pyramid network as input, pass the feature map after global maximum and average pooling to the shared fully connected layer MLP, add the output features and perform sigmoid activation to obtain a weighted feature layer.

3. The target detection method based on a multi-scale mixed dimension network according to claim 1 is characterized in that: The pyramid network includes: five residual modules and an output layer, and the five residual modules are input to the output layer through feature fusion.

4. The target detection method based on a multi-scale mixed dimension network according to claim 3 is characterized in that: The numbers of residual blocks in the five residual modules are 1, 2, 4, 4, and 23 respectively; The five residual modules respectively output a 2x downsampling layer, a 4x downsampling layer, an 8x downsampling layer, a 16x downsampling layer, and a 32x downsampling layer, and the 4x downsampling layer is input into the output layer through feature fusion; The 8x downsampling layer is fused with the features of the 4x downsampling layer after 2x upsampling; the 16x downsampling layer is fused with the features of the 8x downsampling layer after 2x upsampling, and also fused with the features of the 4x upsampling and 4x downsampling layer; the 32x downsampling layer is fused with the features of the 16x downsampling layer after 2x upsampling, and also fused with the features of the 8x downsampling layer after 4x upsampling; The output layer includes two residual blocks and one convolutional layer.

5. A target detection system based on a multi-scale mixed dimension network, characterized in that: include: The detection module is configured to: input the target image into the trained multi-scale mixed dimension network, remove the redundant frames of the obtained image by the NMS method, and obtain an image marked with the target position and category; A network construction module is configured as follows: the multi-scale mixed dimension network includes a pyramid network, and a mixed dimension module is introduced between every two convolutional layers from top to bottom of the pyramid network; The training process of the multi-scale mixed dimension network includes: obtaining a target image detection training set; obtaining the anchor frame coordinates of the target image detection training set through the KMA algorithm; based on the target image detection training set, using cloud computing resources to train the multi-scale mixed dimension network to obtain a target image detection network; The mixed dimension module includes a channel attention module and a spatial attention module; The spatial attention module is configured to: use the feature layer obtained by the channel attention module as the input feature of the spatial module, perform tensor splicing, then perform dimensionality reduction through convolution operation, generate a spatial feature layer through activation operation, and finally multiply the feature layer with the input feature to obtain the final generated feature as the input of the next convolution layer in the pyramid network.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the target detection method based on a multi-scale mixed dimensional network as described in any one of claims 1 to 4 are implemented.

7. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the target detection method based on a multi-scale mixed dimensional network as described in any one of claims 1-4 are implemented.

Images