Target detection method and device, storage medium and program product

Abstract

The invention discloses a target detection method and device, a storage medium and a program product. The method comprises the following steps: receiving a target detection request; in response to the target detection request, acquiring an image of a target area in the machine room to obtain a target image; the target image is input into a target detector for N-scale detection, a target detection result is obtained, a network structure of the target detector comprises a backbone network and a neck network, the backbone network at least comprises M C2f extraction modules used for feature extraction, and the C2f extraction modules at least comprise deformable convolutional neural networks, the neck network at least comprises T C2f fusion modules used for carrying out feature fusion, and the C2f fusion modules comprise grouping convolution and point-by-point convolution. According to the method and the device, the problem that the detection effect is poor when water leakage detection is carried out on the machine room in a manner of placing the water leakage induction line in the prior art is solved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and more specifically, to a target detection method, device, storage medium, and program product. Background Art

[0002] In recent years, with the rapid development of financial institutions, data centers, as the core of business, have been growing in size, and security management requirements have become increasingly stringent. There are many infrastructures and equipment in data centers. Once water seepage or leakage occurs, it will greatly threaten the safety of equipment operation.

[0003] In the related technology, the common methods for detecting water leakage in the computer room of the data center include placing water leakage sensing lines and camera recognition. However, placing water leakage sensing lines has certain limitations. It can only be recognized when water spreads to the sensing lines, and it is impossible to detect water leakage on the top of the computer room. Although traditional camera recognition can detect leakage on the top of the computer room, the top of the computer room has complex wiring, and is often distributed with fire protection, fresh air and other pipes, which cause a lot of obstruction to the camera. In the case of a small leakage area on the top, the camera often cannot quickly identify leakage on the top.

[0004] With regard to the problem that the detection effect of water leakage detection in the computer room by placing water leakage sensing lines in the related technology is poor, no effective solution has been proposed so far. Summary of the invention

[0005] The main purpose of the present application is to provide a target detection method, device, storage medium and program product to solve the problem of poor detection effect when placing water leakage sensing lines to detect water leakage in the computer room in the related art.

[0006] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a target detection method is provided. The method comprises: receiving a target detection request, wherein the target detection request comprises: a detection request for detecting a target in a computer room, wherein the target comprises: a water leakage location; in response to the target detection request, collecting an image of a target area in the computer room to obtain a target image, wherein the target area comprises: the top of the computer room; inputting the target image into a target detector for N scale detections to obtain a target detection result, wherein the network structure of the target detector comprises: a backbone network, a neck network and a head network, wherein the backbone network comprises at least: M C2f extraction modules for feature extraction, wherein the C2f extraction module comprises at least: a deformable convolutional neural network, wherein the neck network comprises at least: T C2f fusion modules for feature fusion, wherein the C2f fusion module comprises: group convolution and point-by-point convolution, wherein the target detection result is used to indicate whether the target exists in the target image, and M, N and T are positive integers.

[0007] Furthermore, the backbone network also includes: a feature enhancement module, which is used to perform feature enhancement processing on the extracted features, input the target image into the target detector for N scale detections, and obtain a target detection result, including: inputting the target image into the backbone network, extracting features of the target image based on the feature enhancement module and M C2f extraction modules, and obtaining N first feature maps, wherein the N first feature maps carry different amounts of information; inputting the N first feature maps into the neck network, and based on T C2f fusion modules, performing feature fusion on the N first feature maps and the feature maps generated in the neck network, and obtaining N second feature maps, wherein the N second feature maps carry different amounts of information; inputting the N second feature maps into the head network for N scale detections to obtain the target detection result.

[0008] Furthermore, the backbone network also includes: S CBS modules, the CBS modules include: convolution layers, batch normalization layers and activation functions, S is a positive integer, and based on the feature enhancement module and the M C2f extraction modules, feature extraction is performed on the target image to obtain N first feature maps, including: based on the first CBS module in the S CBS modules and the second CBS module in the S CBS modules, feature extraction and nonlinear transformation are performed on the target image to obtain a target feature map, wherein the second CBS module and the second CBS module are two adjacent CBS modules in the backbone network; feature extraction is performed on the target feature map based on the S CBS modules and the M C2f extraction modules, and feature enhancement processing is performed on the feature map obtained by feature extraction by the feature enhancement module to obtain N first feature maps.

[0009] Further, feature extraction is performed on the target feature map based on the S CBS modules and the M C2f extraction modules, and feature enhancement processing is performed on the feature map obtained by feature extraction by the feature enhancement module to obtain N first feature maps, including: feature extraction is performed on the target feature map based on the first C2f extraction module in the M C2f extraction modules, and the extracted feature map is input into the third CBS module in the S CBS modules for feature extraction to obtain the feature map output by the third CBS module; the feature map output by the third CBS module is input into the second C2f extraction module in the M C2f extraction modules for feature extraction to obtain the feature map output by the second C2f extraction module; the feature map output by the second C2f extraction module is input into the fourth CBS module in the S CBS modules for feature extraction and input the feature map output by the fourth CBS module into the third C2f extraction module among the M C2f extraction modules for feature extraction to obtain the feature map output by the third C2f extraction module; input the feature map output by the third C2f extraction module into the fifth CBS module among the S CBS modules for feature extraction, and input the feature map output by the fifth CBS module into the fourth C2f extraction module among the M C2f extraction modules for feature extraction to obtain the feature map output by the fourth C2f extraction module; input the feature map output by the fourth C2f extraction module into the feature enhancement module for feature enhancement processing to obtain the feature map after feature enhancement; determine N first feature maps based on the feature map output by the second C2f extraction module, the feature map output by the third C2f extraction module and the feature map after feature enhancement.

[0010] Furthermore, the neck network also includes: F splicing layers, H upsampling modules and P CBS modules, wherein F, H and P are positive integers, and based on T C2f fusion modules, feature fusion is performed on N first feature maps and feature maps generated in the neck network to obtain N second feature maps, including: based on the first upsampling module in the H upsampling modules, upsampling the feature maps after feature enhancement in the N first feature maps to obtain feature maps output by the first upsampling module; based on the first splicing layer in the F splicing layers, splicing the feature map output by the first upsampling module and the feature map output by the third C2f extraction module in the N first feature maps to obtain the feature map output by the first splicing layer; based on the first C2f fusion module in the T C2f fusion modules, performing upsampling on the feature map output by the first splicing layer; Feature fusion processing is performed to obtain a feature map output by the first C2f fusion module; based on the second upsampling module in the H upsampling modules, the feature map output by the first C2f fusion module is upsampled to obtain a feature map output by the second upsampling module; based on the second splicing layer in the F splicing layers, the feature map output by the second upsampling module and the feature map output by the second C2f extraction module in the N first feature maps are spliced ​​to obtain a feature map output by the second splicing layer, and based on the second C2f fusion module in the T C2f fusion modules, the feature map output by the second splicing layer is subjected to feature fusion processing to obtain a feature map output by the second C2f fusion module; based on the feature map output by the first C2f fusion module, the feature map output by the second C2f fusion module and the P CBS modules, N second feature maps are determined.

[0011] Further, based on the feature map output by the first C2f fusion module, the feature map output by the second C2f fusion module, and the P CBS modules, determining N second feature maps includes: based on the first target CBS module in the P CBS modules, performing feature extraction on the feature map output by the second C2f fusion module to obtain the feature map output by the first target CBS module, and based on the third splicing layer in the F splicing layers, performing splicing processing on the feature map output by the first C2f fusion module and the feature map output by the first target CBS module to obtain the feature map output by the third splicing layer; based on the third C2f fusion module in the T C2f fusion modules, performing feature fusion processing on the feature map output by the third splicing layer to obtain the feature map output by the third C2f fusion module. Figure; based on the second target CBS module among the P CBS modules, feature extraction is performed on the feature map output by the third C2f fusion module to obtain the feature map output by the second target CBS module, and based on the fourth splicing layer among the F splicing layers, the feature map output by the second target CBS module and the feature map output by the feature enhancement module are spliced ​​to obtain the feature map output by the fourth splicing layer; based on the fourth C2f fusion module among the T C2f fusion modules, feature fusion processing is performed on the feature map output by the fourth splicing layer to obtain the feature map output by the fourth C2f fusion module; based on the feature map output by the second C2f fusion module, the feature map output by the third C2f fusion module and the feature map output by the fourth C2f fusion module, N second feature maps are determined.

[0012] Furthermore, the target detector is trained in the following manner: obtaining training samples, wherein the training samples include: W historical images of the computer room, each of the historical images includes marking information of the leaking area, the W historical images include: images of the computer room collected in a historical time period, and W is a positive integer; performing model training on the initial detector based on the training samples, and when the initial detector training converges, determining the converged initial detector as the target detector, wherein the initial detector includes: a target detector that has not undergone model training.

[0013] In order to achieve the above-mentioned purpose, according to another aspect of the present application, a target detection device is provided. The device includes: a receiving unit, which is used to receive a target detection request, wherein the target detection request includes: a detection request for detecting a target in a computer room, wherein the target includes: a water leakage location; a collection unit, which is used to respond to the target detection request, collect an image of a target area in the computer room, and obtain a target image, wherein the target area includes: the top of the computer room; a detection unit, which is used to input the target image into a target detector for N scale detections to obtain a target detection result, wherein the network structure of the target detector includes: a backbone network, a neck network, and a head network, wherein the backbone network includes at least: M C2f extraction modules for feature extraction, wherein the C2f extraction module includes at least: a deformable convolutional neural network, wherein the neck network includes at least: T C2f fusion modules for feature fusion, wherein the C2f fusion module includes: group convolution and point-by-point convolution, and the target detection result is used to indicate whether the target exists in the target image, and M, N, and T are positive integers.

[0014] Furthermore, the backbone network also includes: a feature enhancement module, which is used to perform feature enhancement processing on the extracted features. The detection unit includes: an extraction subunit, which is used to input the target image into the backbone network, and perform feature extraction on the target image based on the feature enhancement module and M C2f extraction modules to obtain N first feature maps, wherein the N first feature maps carry different amounts of information; a fusion subunit, which is used to input the N first feature maps into the neck network, and perform feature fusion on the N first feature maps and the feature maps generated in the neck network based on T C2f fusion modules to obtain N second feature maps, wherein the N second feature maps carry different amounts of information; a detection subunit, which is used to input the N second feature maps into the head network for N scale detection to obtain the target detection result.

[0015] Furthermore, the backbone network also includes: S CBS modules, the CBS modules include: convolution layers, batch normalization layers and activation functions, S is a positive integer, and the extraction subunit includes: a first processing module, used to perform feature extraction and nonlinear transformation on the target image based on the first CBS module in the S CBS modules and the second CBS module in the S CBS modules to obtain a target feature map, wherein the second CBS module and the second CBS module are two adjacent CBS modules in the backbone network; a second processing module, used to perform feature extraction on the target feature map based on the S CBS modules and the M C2f extraction modules, and perform feature enhancement processing on the feature map obtained by feature extraction by the feature enhancement module to obtain N first feature maps.

[0016] Further, the second processing module includes: a first processing submodule, used to extract features from the target feature map based on a first C2f extraction module among the M C2f extraction modules, and input the extracted feature map into a third CBS module among the S CBS modules for feature extraction, so as to obtain a feature map output by the third CBS module; a second processing submodule, used to input the feature map output by the third CBS module into a second C2f extraction module among the M C2f extraction modules for feature extraction, so as to obtain a feature map output by the second C2f extraction module; a third processing submodule, used to input the feature map output by the second C2f extraction module into a fourth CBS module among the S CBS modules for feature extraction, and input the feature map output by the fourth CBS module into the third C2f extraction module among the M C2f extraction modules. a fourth processing submodule, for inputting the feature map output by the third C2f extraction module into a fifth CBS module among the S CBS modules for feature extraction, and inputting the feature map output by the fifth CBS module into a fourth C2f extraction module among the M C2f extraction modules for feature extraction, to obtain a feature map output by the fourth C2f extraction module; a fifth processing submodule, for inputting the feature map output by the fourth C2f extraction module into the feature enhancement module for feature enhancement processing, to obtain a feature map after feature enhancement; and a sixth processing submodule, for determining N first feature maps based on the feature map output by the second C2f extraction module, the feature map output by the third C2f extraction module, and the feature map after feature enhancement.

[0017] Furthermore, the neck network also includes: F splicing layers, H upsampling modules and P CBS modules, wherein F, H and P are positive integers, and the fusion subunit includes: a first sampling processing module, which is used to upsample the feature maps after feature enhancement in the N first feature maps based on the first upsampling module in the H upsampling modules, so as to obtain the feature maps output by the first upsampling module; a first splicing processing module, which is used to splice the feature maps output by the first upsampling module and the feature maps output by the third C2f extraction module in the N first feature maps based on the first splicing layer in the F splicing layers, so as to obtain the feature maps output by the first splicing layer; a fusion processing module, which is used to perform feature fusion processing on the feature maps output by the first splicing layer based on the first C2f fusion modules in the T C2f fusion modules, so as to obtain the feature maps output by the first C2f fusion module. Figure; a second sampling processing module, which is used to perform upsampling processing on the feature map output by the first C2f fusion module based on the second upsampling module in the H upsampling modules, so as to obtain the feature map output by the second upsampling module; a second splicing processing module, which is used to perform splicing processing on the feature map output by the second upsampling module and the feature map output by the second C2f extraction module in the N first feature maps based on the second splicing layer in the F splicing layers, so as to obtain the feature map output by the second splicing layer, and perform feature fusion processing on the feature map output by the second splicing layer based on the second C2f fusion module in the T C2f fusion modules, so as to obtain the feature map output by the second C2f fusion module; a determination module, which is used to determine N second feature maps based on the feature map output by the first C2f fusion module, the feature map output by the second C2f fusion module and the P CBS modules.

[0018] Further, the determination module includes: a seventh processing submodule, for performing feature extraction on the feature map output by the second C2f fusion module based on the first target CBS module in the P CBS modules, to obtain the feature map output by the first target CBS module, and performing splicing processing on the feature map output by the first C2f fusion module and the feature map output by the first target CBS module based on the third splicing layer in the F splicing layers, to obtain the feature map output by the third splicing layer; an eighth processing submodule, for performing feature fusion processing on the feature map output by the third splicing layer based on the third C2f fusion module in the T C2f fusion modules, to obtain the feature map output by the third C2f fusion module; a ninth processing submodule, for performing feature fusion processing on the feature map output by the third splicing layer based on the second target CBS module in the P CBS modules The block performs feature extraction on the feature map output by the third C2f fusion module to obtain the feature map output by the second target CBS module, and based on the fourth splicing layer in the F splicing layers, splices the feature map output by the second target CBS module and the feature map output by the feature enhancement module to obtain the feature map output by the fourth splicing layer; a fusion processing submodule is used to perform feature fusion processing on the feature map output by the fourth splicing layer based on the fourth C2f fusion module in the T C2f fusion modules to obtain the feature map output by the fourth C2f fusion module; a determination submodule is used to determine N second feature maps based on the feature map output by the second C2f fusion module, the feature map output by the third C2f fusion module and the feature map output by the fourth C2f fusion module.

[0019] Furthermore, the target detector is obtained by training the following units: an acquisition unit, used to acquire training samples, wherein the training samples include: W historical images of the computer room, each of the historical images includes marking information of the leakage area, the W historical images include: images of the computer room collected in a historical time period, and W is a positive integer; a training unit, used to perform model training on the initial detector based on the training samples, and when the initial detector training converges, determine the converged initial detector as the target detector, wherein the initial detector includes: a target detector that has not undergone model training.

[0020] According to another aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium includes a stored executable program, wherein when the executable program is run, the device where the computer-readable storage medium is located is controlled to execute the target detection method.

[0021] According to another aspect of the present application, an electronic device is provided, comprising: a memory storing an executable program; and a processor for running the program, wherein the target detection method is executed when the program is running.

[0022] According to another aspect of the present application, a computer program product is provided, comprising computer instructions, which implement the steps of the target detection method when executed by a processor.

[0023] In the disclosure of the present application, a target detection request may be received first, wherein the target detection request includes: a detection request for detecting a target in a computer room, wherein the target includes: a water leakage location; in response to the target detection request, an image of a target area in the computer room is collected to obtain a target image, wherein the target area includes: the top of the computer room; the target image is input into a target detector for N scale detections to obtain a target detection result, wherein the network structure of the target detector includes: a backbone network, a neck network, and a head network, wherein the backbone network includes at least: M C2f extraction modules for feature extraction, wherein the C2f extraction module includes at least: a deformable convolutional neural network, wherein the neck network includes at least: T C2f fusion modules for feature fusion, wherein the C2f fusion module includes: group convolution and point-by-point convolution, and the target detection result is used to indicate whether there is a target in the target image, thereby solving the technical problem in the related art of using a method of placing a water leakage sensing line to detect water leakage in the computer room, and the detection effect is poor.

[0024] According to the above-mentioned disclosed content, based on the images in the computer room, multi-scale target (leakage location) detection is performed on the computer room through a target detector containing a C2f extraction module and a C2f fusion module, thereby avoiding the situation in which the leakage area in the computer room is detected by placing a leakage sensing line in the related technology, and there is a situation where identification is unavailable, thereby achieving the technical effect of improving the accuracy of detection of the leakage area in the computer room. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0026] Figure 1 A hardware structure block diagram of a computer terminal for implementing a target detection method is shown;

[0027] Figure 2 is a flow chart of a target detection method provided according to an embodiment of the present application;

[0028] Figure 3 is a schematic diagram of the structure of C2f_DCNv2 provided according to an embodiment of the present application;

[0029] Figure 4 is a schematic diagram of the structure of C2f_Faster provided according to an embodiment of the present application;

[0030] Figure 5 is a schematic diagram of the structure of a target detector provided according to an embodiment of the present application;

[0031] Figure 6 is a schematic diagram of a target detection device provided according to an embodiment of the present application;

[0032] Figure 7 It is a structural block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0033] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0034] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0035] First, some nouns or terms that appear in the description of the embodiments of the present application are subject to the following explanations:

[0036] YOLOv8n: It is a real-time target detector that inherits the fast and accurate characteristics of the target detection algorithm. The network structure of YOLOv8n consists of Backbone, Neck and Head to achieve efficient feature extraction, multi-scale feature fusion and accurate target detection and classification.

[0037] Backbone layer: Convolution and deconvolution layers are used to extract features, while residual connections and bottleneck structures are used to reduce network size and improve performance. This part uses C2f modules as the basic building block.

[0038] C2f module: Convolutional Block 2Fused, a module that fuses two convolutional blocks, is an efficient feature extraction component introduced in YOLOv5 (and its subsequent version YOLOv8). The C2f module consists of multiple core components, mainly including: Convolutional Layer (Conv for short) is used to capture local features of the input image, such as texture, edges, and parts of objects. Batch Normalization (BN for short): adjusts the distribution of the output of the convolutional layer, accelerates the training process and improves the stability of the model. Activation function (such as SiLU or ReLU): introduces nonlinearity, allowing the network to learn more complex features.

[0039] Neck layer: Through multi-scale feature fusion technology, feature maps from different stages of Backbone are fused to improve feature expression capabilities. The technologies include SPPF (Spatial Pyramid Pooling-Fast) module, PAA (Probabilistic Ancho Assignment) module and PAN (Path Aggregation Network) module. The Head layer is mainly responsible for the final target detection and classification tasks. This layer contains the detection head and the classification head. The detection head includes a series of convolutional layers and deconvolutional layers to generate detection results; the classification head uses global average pooling to classify each feature map T.

[0040] CBS module: It is a basic and important module in YOL0v8n. It is mainly composed of three parts: convolution layer (Conv), batch normalization layer (Batch Normalization, referred to as BN) and activation function (SiLU). First, the convolution layer (Conv) is responsible for convolution operation on the input feature image to extract the features in the image. The parameters of the convolution layer include the number of input channels, the number of output channels, the size of the convolution kernel, the step size, the padding, etc. Through the sliding window method, the convolution kernel is used to perform local weighted summation on the input feature map, and the bias term is added to obtain the output feature map. Then, the output of the convolution is normalized through the batch normalization layer (BatchNormalization, BN) so that the data meets a certain distribution law (usually the mean is 0 and the variance is 1), thereby accelerating the convergence speed of the model and improving the stability and generalization ability of the model. Each small batch of data is normalized, including calculating the mean and variance, and then normalizing and scaling. Then, the activation function (SiLU) introduces nonlinear factors so that the model can learn more complex mapping relationships. The SiLU activation function has smooth and non-monotonic properties, which helps alleviate the gradient vanishing problem.

[0041] SPPF (Spatial Pyramid Pooling-Fast) module: A technology used in the field of deep learning and computer vision that extracts features at different scales to capture detailed information of the target object at different scales, enhance the model's perception of the target object, and improve detection accuracy.

[0042] PAA (Probabilistic Anchor Assignment) module: used to intelligently assign anchor boxes, optimize sample selection, and improve model training efficiency and final detection performance.

[0043] PAN (Path Aggregation Network) module: A deep neural network architecture for image semantic segmentation and target detection. It improves model performance through multi-scale feature fusion and enhances the expressiveness of feature maps through bottom-up and top-down paths. This module can improve the model's detection ability for small targets.

[0044] DCNv2 (Deformable Convolutional Networks v2) module: An improved deformable convolutional network that further optimizes the original deformable convolutional network (DCNv1) to better handle complex spatial transformations and feature extraction tasks. DCNv2 is widely used in the field of computer vision, especially in tasks such as object detection and image segmentation.

[0045] BN (Batch Normalization) module: It is a technique used in deep learning to accelerate model training and improve stability. It normalizes the input of each layer, reduces the problem of internal covariate shift, and keeps the input distribution of each layer consistent, thereby significantly accelerating the training process and improving the convergence speed of the model. BN introduces learnable scaling and translation parameters, retains the network's expressive power, and can handle complex neural network structures. It is widely used in tasks such as image classification, object detection, natural language processing, and generative adversarial networks.

[0046] ReLU (Rectified Linear Unit) module: The ReLU module is one of the most commonly used activation functions in deep learning. By introducing nonlinear characteristics, the neural network can learn and represent complex nonlinear relationships. Its core idea is simple and efficient calculation (only one threshold operation is required) and effective mitigation of the gradient vanishing problem (the gradient is always 1 when the input is greater than 0). Compared with traditional activation functions such as Sigmoid and Tanh, ReLU can better capture the nonlinear characteristics of data and is widely used in various neural network architectures, significantly improving training efficiency and model performance.

[0047] SiLU (Sigmoid Linear Unit) module: The SiLU module is also called the Swish activation function, which is a new type of activation function that combines the smoothness of the Sigmoid function (S-shaped function) with the simplicity of the linear function, and is designed to improve the performance of neural networks. By introducing a smooth Sigmoid transformation, the activation function approaches linearity when the input value is large, and maintains nonlinear characteristics when the input value is small, thereby providing a more flexible nonlinear representation and significantly improving the performance of the model in tasks such as image classification and natural language processing.

[0048] Split module: refers to dividing the input data according to certain features or dimensions to obtain multiple sub-datasets. In deep learning, the split operation is often used to process large-scale data sets to reduce data load and improve computing efficiency. In addition, the split operation can also be used for data preprocessing and feature extraction to better train neural network models.

[0049] The Bottleneck module is a commonly used structure in deep convolutional neural networks (CNNs). It aims to improve the depth and computational efficiency of the network by inserting a bottleneck layer into the network, while enhancing the feature extraction capability. The core idea is to use a 1x1 convolutional layer to compress the input features, reduce the feature dimension, and thus reduce computational complexity. It then uses a 3x3 convolutional layer for nonlinear transformation to extract complex high-level features, and finally restores the feature dimension through another 1x1 convolutional layer. This structure increases the depth of the network and enhances its ability to learn complex feature representations without significantly increasing the amount of computation. Therefore, it has been widely used in network architectures such as ResNet and has become an effective means to improve network performance.

[0050] FasterNet Block module: is a commonly used module in deep convolutional neural networks (CNNs). It aims to significantly improve the computational efficiency and reasoning speed of the network while maintaining high performance by optimizing the network structure and introducing new operations such as grouped convolution and point-by-point convolution. Grouped convolution is used to divide the input feature map into multiple groups, and each group performs convolution operations independently, thereby reducing the amount of calculation for each convolution operation; point-by-point convolution processes the feature map using a 1x1 convolution kernel, effectively reducing the computational complexity while retaining rich feature information. In addition, FasterNet Block uses residual connections to enable the network to better learn residual information and improve feature extraction capabilities, and uses batch normalization to stabilize and accelerate the training process and improve the generalization ability of the network.

[0051] It should be noted that the collected information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, data for analysis, etc.) involved in this application are information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data are in compliance with relevant laws, regulations and standards, necessary confidentiality measures are taken, and public order and good customs are not violated, and corresponding operation entrances are provided for users to choose to authorize or refuse. For example, an interface is set up between this system and relevant users or institutions to provide users with corresponding operation entrances for users to choose to agree or refuse the results of automated decision-making; if the user chooses to refuse, the expert decision-making process will be entered.

[0052] Embodiment 1

[0053] According to an embodiment of the present application, a method embodiment of a target detection method is also provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0054] The method embodiment provided in the first embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computer terminal (or mobile device) for implementing a target detection method. Figure 1 As shown, the computer terminal 10 (or mobile device) may include one or more (102a, 102b, ..., 102n are used to illustrate) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0055] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10 (or mobile device). As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0056] The memory 104 can be used to store software programs and modules of application software, such as program instructions / data storage devices corresponding to the target detection method in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, realizing the above-mentioned target detection method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0057] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0058] The display may be, for example, a touch screen liquid crystal display (LCD), which may enable a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0059] Under the above operating environment, this application provides Figure 2 The target detection method shown. Figure 2 It is a flow chart of the target detection method according to the first embodiment of the present application.

[0060] Step S201, receiving a target detection request, wherein the target detection request includes: a detection request for detecting a target in a computer room, and the target includes: a water leakage location.

[0061] The above target detection situation can be used to request a detection request for detecting a target in a machine room (for example, whether there is a water leakage location).

[0062] Step S202, in response to the target detection request, an image of a target area in the computer room is collected to obtain a target image, wherein the target area includes: a top of the computer room.

[0063] The above-mentioned target area may include: the top of the computer room. In this embodiment, an image acquisition device may be used to acquire an image of the target area in the computer room. For example, a camera on the top of the cabinet may be used to capture a top picture to obtain a target image.

[0064] Step S203, input the target image into the target detector for N scale detections to obtain the target detection result, wherein the network structure of the target detector includes: a backbone network, a neck network and a head network, the backbone network includes at least: M C2f extraction modules for feature extraction, the C2f extraction module includes at least: a deformable convolutional neural network, the neck network includes at least: T C2f fusion modules for feature fusion, the C2f fusion module includes: group convolution and point-by-point convolution, the target detection result is used to indicate whether there is a target in the target image, and M, N and T are positive integers.

[0065] The above-mentioned target detector may include three parts: Backbone, Neck and Head. The backbone network includes at least M C2f extraction modules for feature extraction, such as C2f_DCNv2 architecture. The above-mentioned neck network includes at least T C2f fusion modules for feature fusion, such as C2f_Faster architecture. The above-mentioned C2f_DCNv2 can be obtained by introducing the DCNv2 module into the C2f module, and C2f_Faster can be obtained by introducing the FasterNet Block module with grouped convolution and point-by-point convolution into the C2f module.

[0066] Figure 3 is a schematic diagram of the structure of C2f_DCNv2 provided according to an embodiment of the present application, such as Figure 3 As shown, C2f_DCNv2 includes: DCBS, Split, Bottleneck, Concat, wherein DCBS is composed of DCNv2, BN and ReLU / SiLU, that is, C2f_DCNv2 in this application replaces CBS in C2f module with DCBS module, wherein, Figure 3 The C in represents the number of channels.

[0067] Figure 4 is a schematic diagram of the structure of C2f_Faster provided according to an embodiment of the present application, such as Figure 4 As shown, it includes: CBS, Split, FasterNet Block, Concat, that is, C2f_Faster in this application replaces Bottleneck in the C2f module with the FasterNet Block module.

[0068] In this embodiment, the target image is input into the target detector, and the target detector is used to detect whether there is a target in the target image at N scales to obtain the target detection result. The detection of N scales can include the detection of small, medium and large targets. For example, after the target image is input into the target detector, the Backbone layer extracts image features through the C2f_DCNv2 architecture, and enhances the extracted features with the help of the Focal Modulation (feature enhancement module) technology; the Neck layer uses the C2f_Faster module to achieve information fusion between each feature level of the backbone network; the Head layer consists of three different detection heads, which detect small, medium and large targets respectively, thereby improving the speed of water seepage detection.

[0069] Through the above steps, in this embodiment, based on the image in the computer room, the target detector containing the C2f extraction module and the C2f fusion module performs multi-scale target (leakage position) detection on the computer room, avoiding the situation that the leakage area in the computer room cannot be identified by placing a leakage sensing line in the related art, thereby achieving the technical effect of improving the accuracy of the detection of the leakage area in the computer room. In addition, the technical problem that the detection effect of the leakage in the computer room is poor by placing a leakage sensing line in the related art is solved.

[0070] Optionally, in the target detection method provided in the embodiment of the present application, the backbone network also includes: a feature enhancement module, the feature enhancement module is used to perform feature enhancement processing on the extracted features, and the target image is input into the target detector for N scale detections to obtain a target detection result, including: inputting the target image into the backbone network, performing feature extraction on the target image based on the feature enhancement module and M C2f extraction modules, and obtaining N first feature maps, wherein the N first feature maps carry different amounts of information; inputting the N first feature maps into the neck network, and based on T C2f fusion modules, performing feature fusion on the N first feature maps and the feature maps generated in the neck network to obtain N second feature maps, wherein the N second feature maps carry different amounts of information; inputting the N second feature maps into the head network for N scale detections to obtain a target detection result.

[0071] The above-mentioned feature enhancement module includes: Focal Modulation. In this embodiment, the target image can be input into the backbone network, and the feature of the target image can be extracted based on the feature enhancement module and the C2f extraction module (C2f_DCNv2) in the backbone network to obtain N first feature maps containing different amounts of information. Then, the N first feature maps can be input into the neck network of the target detector, and the C2f fusion module (C2f_Faster) in the neck network is used to perform feature fusion to obtain N second feature maps of different scales. Then, the N second feature maps of different scales can be input into the head network, and target detection heads of different scales are used to perform target detection on the second feature maps to obtain the target detection result.

[0072] In this embodiment, the target detector is used to detect targets of different scales, thereby improving the technical effect of the accuracy of water seepage position detection.

[0073] Optionally, in the target detection method provided in the embodiment of the present application, the backbone network also includes: S CBS modules, the CBS module includes: a convolution layer, a batch normalization layer and an activation function, S is a positive integer, and feature extraction is performed on the target image based on the feature enhancement module and the M C2f extraction modules to obtain N first feature maps, including: based on the first CBS module in the S CBS modules and the second CBS module in the S CBS modules, feature extraction and nonlinear transformation are performed on the target image to obtain a target feature map, wherein the second CBS module and the second CBS module are two adjacent CBS modules in the backbone network; feature extraction is performed on the target feature map based on the S CBS modules and the M C2f extraction modules, and feature enhancement processing is performed on the feature map obtained by feature extraction by the feature enhancement module to obtain N first feature maps.

[0074] Figure 5 is a schematic diagram of the structure of a target detector provided according to an embodiment of the present application, such as Figure 5 As shown, in this embodiment, the target image can be subjected to two CBS modules (i.e., the first CBS module and the second CBS module) to perform feature extraction and nonlinear transformation on the input image to obtain a target feature map. Then, based on S CBS modules (e.g., Figure 5 The third CBS module, the fourth CBS module, the fifth CBS module) and M C2f extraction modules (for example, Figure 5 The first C2f extraction module, the second C2f extraction module, the third C2f extraction module and the fourth C2f extraction module in the target feature map are used to extract features, and feature enhancement processing is performed on the feature map obtained by feature extraction by the feature enhancement module to obtain N first feature maps. The purpose of multi-scale feature extraction of the target image is achieved.

[0075] Optionally, in the target detection method provided in the embodiment of the present application, feature extraction is performed on the target feature map based on S CBS modules and M C2f extraction modules, and feature enhancement processing is performed on the feature map obtained by feature extraction by the feature enhancement module to obtain N first feature maps, including: feature extraction is performed on the target feature map based on the first C2f extraction module in the M C2f extraction modules, and the extracted feature map is input into the third CBS module in the S CBS modules for feature extraction to obtain the feature map output by the third CBS module; the feature map output by the third CBS module is input into the second C2f extraction module in the M C2f extraction modules for feature extraction to obtain the feature map output by the second C2f extraction module; the feature map output by the second C2f extraction module is input into the fourth CBS module in the S CBS modules The module performs feature extraction, and inputs the feature map output by the fourth CBS module into the third C2f extraction module among the M C2f extraction modules for feature extraction, so as to obtain the feature map output by the third C2f extraction module; the feature map output by the third C2f extraction module is input into the fifth CBS module among the S CBS modules for feature extraction, and the feature map output by the fifth CBS module is input into the fourth C2f extraction module among the M C2f extraction modules for feature extraction, so as to obtain the feature map output by the fourth C2f extraction module; the feature map output by the fourth C2f extraction module is input into the feature enhancement module for feature enhancement processing, so as to obtain the feature map after feature enhancement; based on the feature map output by the second C2f extraction module, the feature map output by the third C2f extraction module and the feature map after feature enhancement, N first feature maps are determined.

[0076] like Figure 5 As shown, in this embodiment, feature extraction can be performed on the target feature map based on the first C2f extraction module (C2f_DCNv2) among the M C2f extraction modules, and the extracted feature map is input into the third CBS module among the S CBS modules for feature extraction to obtain a feature map output by the third CBS module.

[0077] It should be noted that the processed image (corresponding to the target feature map) is input into the C2f_DCNv2 module (corresponding to the first C2f extraction module). In order to improve the performance of the network model, the DCNv2 module is introduced into the C2f module, which can enable it to better utilize contextual information, thereby improving the accuracy and robustness of target detection. DCNv2 can introduce a weight coefficient at each sampling point to calculate the output eigenvalue Po. The introduction of this weight coefficient can suppress irrelevant background information and help the model reduce the interference of irrelevant content when learning features. In order to improve the performance of the network model, the DCNv2 module is introduced into the C2f module, so that it can better utilize contextual information, thereby improving the accuracy and robustness of target detection. The structure of the C2f_DCNv2 module is as follows: Figure 3 shown.

[0078] The output image of the C2f_DCNv2 module (the feature map output by the first C2f extraction module) is processed by one CBS module (the third CBS module) and one C2f_DCNv2 module (the second C2f extraction module) to obtain the feature map output by the second C2f extraction module. The feature map output by the second C2f extraction module is output to the fourth CBS module and the second splicing module in the neck network respectively.

[0079] The feature map output by the fourth CBS module is processed by one CBS module (fourth CBS module) and one C2f_DCNv2 module (third C2f extraction module) to obtain the feature map output by the third C2f extraction module. The feature map output by the third C2f extraction module can be output to the fifth CBS module and the first splicing layer in the neck network respectively.

[0080] The feature map output by the third C2f extraction module is processed by one CBS module (fifth CBS module) and one C2f_DCNv2 module (fourth C2f extraction module) to obtain the feature map output by the fourth C2f extraction module. The feature map output by the fourth C2f extraction module can be output to the feature enhancement module.

[0081] The feature map output by the fourth C2f extraction module is input into the Focal Modulation module (corresponding to the feature enhancement module) for processing to obtain the feature map after feature enhancement. The feature map after feature enhancement is output to the first upsampling module and the fourth splicing layer in the neck network respectively.

[0082] It should be noted that the Focal Modulation module uses the focal modulation module to replace the self-attention mechanism to simulate token interactions in vision. Focal modulation consists of the following three components: focal contextualization, gated aggregation, and element-by-element affine transformation. Focal contextualization mainly uses deep convolutional layers to encode visual context information of different scales. Convolutional layers are good at capturing visual features from close to long distances, enabling the network to understand image content from multiple levels. This method enables the network to maintain sensitivity to local complexity and enhance the extraction of global information when aggregating contextual information; gated aggregation is mainly responsible for aggregating contextual information of a single query token. This function enables the network to identify key contextual information related to the query token, thereby focusing on the most relevant aspects. By adopting gated aggregation, FocalNets (FocalModulation Networks) can focus on the key information of a given task more efficiently and in a more targeted manner. This method can minimize the processing of redundant information and strengthen the focus on basic features, thereby improving the efficiency and performance of the model. In visual tasks, it can improve the performance of object detection and image classification, especially in complex or dynamic visual environments; the element-wise affine transformation mainly injects the obtained modulator into each query token through gated aggregation. This aggregation process combines the contextual information with the original features of the query token, so that the contextual information contained in the modulator can have a direct impact on the representation of the token.

[0083] In this embodiment, N first feature maps can be composed of the feature map output by the second C2f extraction module, the feature map output by the third C2f extraction module, and the feature map after feature enhancement, thereby achieving the purpose of multi-scale feature extraction of the target image.

[0084] Optionally, in the target detection method provided in the embodiment of the present application, the neck network also includes: F splicing layers, H upsampling modules and P CBS modules, wherein F, H and P are positive integers, and based on T C2f fusion modules, feature fusion is performed on N first feature maps and feature maps generated in the neck network to obtain N second feature maps, including: based on the first upsampling module in the H upsampling modules, upsampling the feature maps after feature enhancement in the N first feature maps to obtain feature maps output by the first upsampling module; based on the first splicing layer in the F splicing layers, splicing the feature maps output by the first upsampling module and the feature maps output by the third C2f extraction module in the N first feature maps to obtain feature maps output by the first splicing layer; based on the first C2f fusion module in the T C2f fusion modules, upsampling the feature maps output by the first upsampling module and the feature maps output by the third C2f extraction module in the N first feature maps to obtain feature maps output by the first splicing layer. The feature map output by the connection layer is subjected to feature fusion processing to obtain the feature map output by the first C2f fusion module; the feature map output by the first C2f fusion module is upsampled based on the second upsampling module in the H upsampling modules to obtain the feature map output by the second upsampling module; based on the second splicing layer in the F splicing layers, the feature map output by the second upsampling module and the feature map output by the second C2f extraction module in the N first feature maps are spliced ​​to obtain the feature map output by the second splicing layer, and the feature map output by the second splicing layer is subjected to feature fusion processing based on the second C2f fusion module in the T C2f fusion modules to obtain the feature map output by the second C2f fusion module; based on the feature map output by the first C2f fusion module, the feature map output by the second C2f fusion module and the P CBS modules, N second feature maps are determined.

[0085] like Figure 5 As shown, in this embodiment, the data output from the FocalModulation module (corresponding to the feature map after feature enhancement in the N first feature maps) enters the upsample module (first upsample module) and the Concat splicing module (first splicing layer) for processing. The Upsample module can be used to enlarge the size of the input data. For example, each data can be copied four times, so that the size of the output data becomes twice the size of the input data. The function of the Concat (splicing) module is to merge the outputs from different layers or different networks along a specific dimension (usually the longest sub-dimension), thereby merging multiple feature vectors or data streams into a single vector or matrix. Doing so allows the neural network to consider information from multiple sources at the same time, enhancing the model's expressiveness and learning capabilities.

[0086] The data that has passed through the upsample module and the Concat module (the feature map output by the first concatenation layer) is input into the C2f_Faster module (the first C2f fusion module) for processing to obtain the feature map output by the first C2f fusion module.

[0087] The data that has passed through the C2f_Faster module (the feature map output by the first C2f fusion module) is input into the upsample module (the second upsample module) and the Concat module (the second concatenation layer) for processing to obtain the feature map output by the second concatenation layer. The feature map output by the second concatenation layer can then be input into the C2f_Faster module (the second C2f fusion module) for processing to obtain the large target detection head module (Detect (80x80x256)) that is output to the head network for detection.

[0088] Based on the feature map output by the first C2f fusion module, the feature map output by the second C2f fusion module, and the P CBS modules, N second feature maps are determined.

[0089] It should be noted that the C2f_Faster module can be constructed by introducing FasterNet Block into the C2f module, thereby improving the accuracy and robustness of target detection. The FasterNet Block module is constructed based on PConv and the inverted residual structure. This module can not only improve network performance by accelerating the training and reasoning process, reducing the number of parameters, and optimizing computational efficiency, but can also be applied to generate adversarial networks or other complex network architectures to improve the adaptability and flexibility of network models.

[0090] Optionally, in the target detection method provided in the embodiment of the present application, based on the feature map output by the first C2f fusion module, the feature map output by the second C2f fusion module, and P CBS modules, N second feature maps are determined, including: based on the first target CBS module in the P CBS modules, feature extraction is performed on the feature map output by the second C2f fusion module to obtain the feature map output by the first target CBS module, and based on the third splicing layer in the F splicing layers, the feature map output by the first C2f fusion module and the feature map output by the first target CBS module are spliced ​​to obtain the feature map output by the third splicing layer; based on the third C2f fusion module in the T C2f fusion modules, feature fusion processing is performed on the feature map output by the third splicing layer to obtain the feature map output by the third C2f fusion module module output; based on the second target CBS module among the P CBS modules, feature extraction is performed on the feature map output by the third C2f fusion module to obtain the feature map output by the second target CBS module, and based on the fourth splicing layer among the F splicing layers, the feature map output by the second target CBS module and the feature map output by the feature enhancement module are spliced ​​to obtain the feature map output by the fourth splicing layer; based on the fourth C2f fusion module among the T C2f fusion modules, feature fusion is performed on the feature map output by the fourth splicing layer to obtain the feature map output by the fourth C2f fusion module; based on the feature map output by the second C2f fusion module, the feature map output by the third C2f fusion module and the feature map output by the fourth C2f fusion module, N second feature maps are determined.

[0091] like Figure 5 As shown, in this embodiment, the data processed by the C2f_Faster module (the feature map output by the second C2f fusion module) can be processed by 1 CBS module (first target CBS module), 1 Concat module (third splicing layer) and C2f_Faster module (the third C2f fusion module among T C2f fusion modules) to obtain the feature map output by the third C2f fusion module. The feature map output by the third C2f fusion module can be output to the medium-sized target detection head module (Detect (40x40x512)) in the head network for detection and the second target CBS module among P CBS modules.

[0092] The data passing through the C2f_Faster module is processed by 1 CBS module (second target CBS module), 1 Concat module (fourth concatenation layer) and C2f_Faster module (fourth C2f fusion module) to obtain the feature map output by the fourth C2f fusion module. The feature map output by the fourth C2f fusion module can be output to the small target detection head module (Detect(20x20x512)) for detection.

[0093] In this embodiment, N second feature maps can be composed of the feature map output by the second C2f fusion module, the feature map output by the third C2f fusion module, and the feature map output by the fourth C2f fusion module, thereby achieving the purpose of feature fusion of the first feature maps of multiple scales of the backbone network, and further improving the technical effect of the accuracy of the target detection results.

[0094] Optionally, in the target detection method provided in an embodiment of the present application, the target detector is trained in the following manner: obtaining training samples, wherein the training samples include: W historical images of the computer room, each historical image includes marking information of the leaking area, the W historical images include: images of the computer room collected in a historical time period, and W is a positive integer; performing model training on the initial detector based on the training samples, and when the initial detector training converges, determining the converged initial detector as the target detector, wherein the initial detector includes: a target detector that has not undergone model training.

[0095] In this embodiment, W historical images of the computer room in the historical time period can be obtained, each historical image includes marking information of the leakage area, and training samples are obtained. Then, the initial detector to be trained can be trained based on the training samples. When the training results meet the preset conditions (for example, the convergence conditions of the model), the initial detector can be determined as the target detector. The target detector can be used to detect leakage on the top of the computer room, thereby achieving the technical effect of improving the accuracy of the leakage detection results in the computer room.

[0096] It should be noted that this embodiment has been trained and performance verified on a specified data set. Compared with the traditional model, the number of model parameters has been reduced by 12.5%, and the detection speed has been increased by 9.5 frames per second, achieving good results.

[0097] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0098] Embodiment 2

[0099] The embodiment of the present application also provides a target detection device. It should be noted that the target detection device of the embodiment of the present application can be used to execute the target detection method provided by the embodiment of the present application. The target detection device provided by the embodiment of the present application is introduced below.

[0100] According to an embodiment of the present application, a device for implementing the above target detection method is also provided, such as Figure 6As shown, the device includes: a receiving unit 61, a collecting unit 62 and a detecting unit 63.

[0101] The receiving unit 61 is used to receive a target detection request, wherein the target detection request includes: a detection request for detecting a target in the computer room, the target includes: a water leakage location;

[0102] The acquisition unit 62 is used to respond to the target detection request, acquire an image of a target area in the computer room, and obtain a target image, wherein the target area includes: a top of the computer room;

[0103] The detection unit 63 is used to input the target image into the target detector for N scale detections to obtain the target detection result, wherein the network structure of the target detector includes: a backbone network, a neck network and a head network, the backbone network includes at least: M C2f extraction modules for feature extraction, the C2f extraction module includes at least: a deformable convolutional neural network, the neck network includes at least: T C2f fusion modules for feature fusion, the C2f fusion module includes: group convolution and point-by-point convolution, the target detection result is used to indicate whether there is a target in the target image, and M, N and T are positive integers.

[0104] In the target detection device provided in the embodiment of the present application, a target detection request can be received by a receiving unit 61, wherein the target detection request includes: a detection request for detecting a target in a computer room, the target includes: a water leakage location, and the acquisition unit 62 responds to the target detection request, collects an image of the target area in the computer room, and obtains a target image, wherein the target area includes: the top of the computer room, and the target image is input into the target detector through the detection unit 63 for N scale detections to obtain a target detection result, wherein the network structure of the target detector includes: a backbone network, a neck network, and a head network, the backbone network includes at least: M C2f extraction modules for feature extraction, the C2f extraction module includes at least: a deformable convolutional neural network, the neck network includes at least: T C2f fusion modules for feature fusion, the C2f fusion module includes: group convolution and point-by-point convolution, and the target detection result is used to indicate whether there is a target in the target image, and M, N, and T are positive integers. This solves the technical problem of poor detection effect of placing a water leakage sensing line to detect water leakage in the computer room in the related art. In this embodiment, based on the image in the computer room, multi-scale target (leakage location) detection is performed on the computer room through a target detector containing a C2f extraction module and a C2f fusion module, thereby avoiding the situation in which the leakage area in the computer room is detected by placing a leakage sensing line in the related technology, and the situation that the leakage area cannot be identified is avoided, thereby achieving the technical effect of improving the accuracy of the detection of the leakage area in the computer room.

[0105] Optionally, in the target detection device provided in the embodiment of the present application, the backbone network also includes: a feature enhancement module, the feature enhancement module is used to perform feature enhancement processing on the extracted features, and the detection unit includes: an extraction subunit, used to input the target image into the backbone network, and extract features of the target image based on the feature enhancement module and M C2f extraction modules to obtain N first feature maps, wherein the N first feature maps carry different amounts of information; a fusion subunit, used to input the N first feature maps into the neck network, and based on T C2f fusion modules, perform feature fusion on the N first feature maps and the feature maps generated in the neck network to obtain N second feature maps, wherein the N second feature maps carry different amounts of information; a detection subunit, used to input the N second feature maps into the head network for N scale detections to obtain target detection results.

[0106] Optionally, in the target detection device provided in the embodiment of the present application, the backbone network also includes: S CBS modules, the CBS module includes: a convolution layer, a batch normalization layer and an activation function, S is a positive integer, and the extraction subunit includes: a first processing module, which is used to perform feature extraction and nonlinear transformation on the target image based on the first CBS module in the S CBS modules and the second CBS module in the S CBS modules to obtain a target feature map, wherein the second CBS module and the second CBS module are two adjacent CBS modules in the backbone network; a second processing module, which is used to extract features from the target feature map based on the S CBS modules and the M C2f extraction modules, and perform feature enhancement processing on the feature map obtained by feature extraction by the feature enhancement module to obtain N first feature maps.

[0107] Optionally, in the target detection device provided in the embodiment of the present application, the second processing module includes: a first processing submodule, which is used to perform feature extraction on the target feature map based on the first C2f extraction module among the M C2f extraction modules, and input the extracted feature map into the third CBS module among the S CBS modules for feature extraction, so as to obtain the feature map output by the third CBS module; a second processing submodule, which is used to input the feature map output by the third CBS module into the second C2f extraction module among the M C2f extraction modules for feature extraction, so as to obtain the feature map output by the second C2f extraction module; a third processing submodule, which is used to input the feature map output by the second C2f extraction module into the fourth CBS module among the S CBS modules for feature extraction, and input the feature map output by the fourth CBS module into the M C2f extraction modules. The third C2f extraction module in the extraction module performs feature extraction to obtain a feature map output by the third C2f extraction module; a fourth processing submodule is used to input the feature map output by the third C2f extraction module into the fifth CBS module in the S CBS modules for feature extraction, and input the feature map output by the fifth CBS module into the fourth C2f extraction module in the M C2f extraction modules for feature extraction to obtain a feature map output by the fourth C2f extraction module; a fifth processing submodule is used to input the feature map output by the fourth C2f extraction module into the feature enhancement module for feature enhancement processing to obtain a feature map after feature enhancement; a sixth processing submodule is used to determine N first feature maps based on the feature map output by the second C2f extraction module, the feature map output by the third C2f extraction module, and the feature map after feature enhancement.

[0108] Optionally, in the target detection device provided in the embodiment of the present application, the neck network further includes: F splicing layers, H upsampling modules and P CBS modules, wherein F, H and P are positive integers, and the fusion subunit includes: a first sampling processing module, which is used to perform upsampling processing on the feature maps after feature enhancement in the N first feature maps based on the first upsampling module in the H upsampling modules, so as to obtain the feature maps output by the first upsampling module; a first splicing processing module, which is used to perform splicing processing on the feature maps output by the first upsampling module and the feature maps output by the third C2f extraction module in the N first feature maps based on the first splicing layer in the F splicing layers, so as to obtain the feature maps output by the first splicing layer; and a fusion processing module, which is used to perform feature fusion processing on the feature maps output by the first splicing layer based on the first C2f fusion module in the T C2f fusion modules, so as to obtain the first C a second sampling processing module, which is used to perform upsampling processing on the feature map output by the first C2f fusion module based on the second upsampling module in the H upsampling modules, so as to obtain the feature map output by the second upsampling module; a second splicing processing module, which is used to perform splicing processing on the feature map output by the second upsampling module and the feature map output by the second C2f extraction module in the N first feature maps based on the second splicing layer in the F splicing layers, so as to obtain the feature map output by the second splicing layer, and perform feature fusion processing on the feature map output by the second splicing layer based on the second C2f fusion module in the T C2f fusion modules, so as to obtain the feature map output by the second C2f fusion module; a determination module, which is used to determine N second feature maps based on the feature map output by the first C2f fusion module, the feature map output by the second C2f fusion module and the P CBS modules.

[0109] Optionally, in the target detection device provided in the embodiment of the present application, the determination module includes: a seventh processing submodule, which is used to perform feature extraction on the feature map output by the second C2f fusion module based on the first target CBS module among the P CBS modules, to obtain the feature map output by the first target CBS module, and to perform splicing processing on the feature map output by the first C2f fusion module and the feature map output by the first target CBS module based on the third splicing layer among the F splicing layers, to obtain the feature map output by the third splicing layer; an eighth processing submodule, which is used to perform feature fusion processing on the feature map output by the third splicing layer based on the third C2f fusion module among the T C2f fusion modules, to obtain the feature map output by the third C2f fusion module; a ninth processing submodule, which is used to extract the feature map output by the second C2f fusion module based on the first target CBS module among the P CBS modules, to obtain the feature map output by the first target CBS module, and to perform splicing processing on the feature map output by the first C2f fusion module and the feature map output by the first target CBS module based on the third splicing layer among the F splicing layers, to obtain the feature map output by the third splicing layer; The second target CBS module in the F splicing layers performs feature extraction on the feature map output by the third C2f fusion module to obtain the feature map output by the second target CBS module, and based on the fourth splicing layer in the F splicing layers, performs splicing processing on the feature map output by the second target CBS module and the feature map output by the feature enhancement module to obtain the feature map output by the fourth splicing layer; a fusion processing submodule is used to perform feature fusion processing on the feature map output by the fourth splicing layer based on the fourth C2f fusion module in the T C2f fusion modules to obtain the feature map output by the fourth C2f fusion module; a determination submodule is used to determine N second feature maps based on the feature map output by the second C2f fusion module, the feature map output by the third C2f fusion module and the feature map output by the fourth C2f fusion module.

[0110] Optionally, in the target detection device provided in the embodiment of the present application, the target detector is trained by the following units: an acquisition unit, used to acquire training samples, wherein the training samples include: W historical images of the computer room, each historical image includes marking information of the leakage area, the W historical images include: images of the computer room collected in a historical time period, and W is a positive integer; a training unit, used to perform model training on the initial detector based on the training samples, and when the initial detector training converges, determine the converged initial detector as the target detector, wherein the initial detector includes: a target detector that has not undergone model training.

[0111] It should be noted that the above-mentioned receiving unit 61, collecting unit 62 and detecting unit 63 correspond to steps S201 to S203 in the first embodiment, and the examples and application scenarios implemented by the two modules and the corresponding steps are the same, but are not limited to the contents disclosed in the first embodiment. It should be noted that the above-mentioned modules or units can be hardware components or software components stored in a memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n), and the above-mentioned modules can also be run in the computer terminal 10 provided in the first embodiment as part of the device.

[0112] Embodiment 3

[0113] An embodiment of the present application may provide an electronic device, Figure 7 is a structural block diagram of an electronic device according to an embodiment of the present application. Figure 7 As shown, the electronic device may include: one or more ( Figure 7 (only one is shown) processor 1002, memory 1004, storage controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0114] Among them, the memory can be used to store software programs and modules, such as program instructions / modules corresponding to the methods and devices in the embodiments of the present application, and the processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, realizing the above-mentioned method. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0115] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: receiving a target detection request, wherein the target detection request includes: a detection request for detecting a target in the computer room, and the target includes: a water leakage location; responding to the target detection request, collecting an image of a target area in the computer room to obtain a target image, wherein the target area includes: the top of the computer room; inputting the target image into the target detector for N scale detections to obtain a target detection result, wherein the network structure of the target detector includes: a backbone network, a neck network and a head network, the backbone network at least includes: M C2f extraction modules for feature extraction, the C2f extraction module at least includes: a deformable convolutional neural network, the neck network at least includes: T C2f fusion modules for feature fusion, the C2f fusion module includes: group convolution and point-by-point convolution, the target detection result is used to indicate whether there is a target in the target image, and M, N and T are positive integers.

[0116] The processor can call the information and application programs stored in the memory through the transmission device to perform the following steps: the backbone network also includes: a feature enhancement module, the feature enhancement module is used to perform feature enhancement processing on the extracted features, and the target image is input into the target detector for N scale detection to obtain the target detection result, including: inputting the target image into the backbone network, extracting features of the target image based on the feature enhancement module and M C2f extraction modules, and obtaining N first feature maps, wherein the N first feature maps carry different amounts of information; inputting the N first feature maps into the neck network, and based on T C2f fusion modules, performing feature fusion on the N first feature maps and the feature maps generated in the neck network to obtain N second feature maps, wherein the N second feature maps carry different amounts of information; inputting the N second feature maps into the head network for N scale detection to obtain the target detection result.

[0117] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: the backbone network also includes: S CBS modules, the CBS module includes: a convolution layer, a batch normalization layer and an activation function, S is a positive integer, based on the feature enhancement module and the M C2f extraction modules, feature extraction is performed on the target image to obtain N first feature maps, including: based on the first CBS module in the S CBS modules and the second CBS module in the S CBS modules, feature extraction and nonlinear transformation are performed on the target image to obtain a target feature map, wherein the second CBS module and the second CBS module are two adjacent CBS modules in the backbone network; based on the S CBS modules and the M C2f extraction modules, feature extraction is performed on the target feature map, and feature enhancement processing is performed on the feature map obtained by feature extraction by the feature enhancement module to obtain N first feature maps.

[0118] The processor can call the information and application program stored in the memory through the transmission device to perform the following steps: extracting features from the target feature map based on S CBS modules and M C2f extraction modules, and performing feature enhancement processing on the feature map obtained by feature extraction by the feature enhancement module to obtain N first feature maps, including: extracting features from the target feature map based on the first C2f extraction module in the M C2f extraction modules, and inputting the extracted feature map into the third CBS module in the S CBS modules for feature extraction to obtain a feature map output by the third CBS module; inputting the feature map output by the third CBS module into the second C2f extraction module in the M C2f extraction modules for feature extraction to obtain a feature map output by the second C2f extraction module; inputting the feature map output by the second C2f extraction module into the S CBS modules The fourth CBS module in the M C2f extraction modules performs feature extraction, and the feature map output by the fourth CBS module is input into the third C2f extraction module in the M C2f extraction modules for feature extraction, so as to obtain the feature map output by the third C2f extraction module; the feature map output by the third C2f extraction module is input into the fifth CBS module in the S CBS modules for feature extraction, and the feature map output by the fifth CBS module is input into the fourth C2f extraction module in the M C2f extraction modules for feature extraction, so as to obtain the feature map output by the fourth C2f extraction module; the feature map output by the fourth C2f extraction module is input into the feature enhancement module for feature enhancement processing, so as to obtain the feature map after feature enhancement; based on the feature map output by the second C2f extraction module, the feature map output by the third C2f extraction module and the feature map after feature enhancement, N first feature maps are determined.

[0119] The processor can call the information and application program stored in the memory through the transmission device to perform the following steps: the neck network also includes: F splicing layers, H upsampling modules and P CBS modules, where F, H and P are positive integers, and based on T C2f fusion modules, feature fusion is performed on N first feature maps and feature maps generated in the neck network to obtain N second feature maps, including: based on the first upsampling module in the H upsampling modules, upsampling the feature maps after feature enhancement in the N first feature maps to obtain the feature maps output by the first upsampling module; based on the first splicing layer in the F splicing layers, splicing the feature maps output by the first upsampling module and the feature maps output by the third C2f extraction module in the N first feature maps to obtain the feature maps output by the first splicing layer; based on the first C2f fusion module in the T C2f fusion modules block, perform feature fusion processing on the feature map output by the first splicing layer to obtain the feature map output by the first C2f fusion module; up-sample the feature map output by the first C2f fusion module based on the second up-sampling module in the H up-sampling modules to obtain the feature map output by the second up-sampling module; based on the second splicing layer in the F splicing layers, perform splicing processing on the feature map output by the second up-sampling module and the feature map output by the second C2f extraction module in the N first feature maps to obtain the feature map output by the second splicing layer, and perform feature fusion processing on the feature map output by the second splicing layer based on the second C2f fusion module in the T C2f fusion modules to obtain the feature map output by the second C2f fusion module; determine N second feature maps based on the feature map output by the first C2f fusion module, the feature map output by the second C2f fusion module and the P CBS modules.

[0120] The processor can call the information and application program stored in the memory through the transmission device to perform the following steps: based on the feature map output by the first C2f fusion module, the feature map output by the second C2f fusion module and the P CBS modules, determine N second feature maps, including: based on the first target CBS module in the P CBS modules, extract the feature map output by the second C2f fusion module to obtain the feature map output by the first target CBS module, and based on the third splicing layer in the F splicing layers, splice the feature map output by the first C2f fusion module and the feature map output by the first target CBS module to obtain the feature map output by the third splicing layer; based on the third C2f fusion module in the T C2f fusion modules, perform feature fusion processing on the feature map output by the third splicing layer to obtain The feature map output by the third C2f fusion module; based on the second target CBS module among the P CBS modules, feature extraction is performed on the feature map output by the third C2f fusion module to obtain the feature map output by the second target CBS module, and based on the fourth splicing layer among the F splicing layers, the feature map output by the second target CBS module and the feature map output by the feature enhancement module are spliced ​​to obtain the feature map output by the fourth splicing layer; based on the fourth C2f fusion module among the T C2f fusion modules, feature fusion processing is performed on the feature map output by the fourth splicing layer to obtain the feature map output by the fourth C2f fusion module; based on the feature map output by the second C2f fusion module, the feature map output by the third C2f fusion module and the feature map output by the fourth C2f fusion module, N second feature maps are determined.

[0121] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: the target detector is trained in the following manner: obtaining training samples, wherein the training samples include: W historical images of the computer room, each historical image includes marking information of the leaking area, the W historical images include: images of the computer room collected in the historical time period, and W is a positive integer; based on the training samples, the initial detector is model trained, and when the initial detector training converges, the converged initial detector is determined as the target detector, wherein the initial detector includes: a target detector that has not been model trained.

[0122] By adopting the embodiment of the present application, based on the image in the computer room, a target detector including a C2f extraction module and a C2f fusion module is used to detect the target (leakage position) of the computer room at multiple scales, thereby avoiding the situation that the leakage area in the computer room cannot be identified by placing a leakage sensing line in the related art, thereby achieving the technical effect of improving the accuracy of the detection of the leakage area in the computer room. In addition, the technical problem that the detection effect of the leakage detection in the computer room by placing a leakage sensing line in the related art is poor is solved.

[0123] It can be understood by those skilled in the art that Figure 7 The structure shown is for illustration only, and the electronic device may also be a smart phone, a tablet computer, a PDA, a mobile Internet device (MID), a PAD or other terminal device. Figure 7 The structure of the electronic device is not limited. Figure 7 More or fewer components (such as network interfaces, display devices, etc.) shown in, or having Figure 7 Different configurations are shown.

[0124] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, and the storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0125] Embodiment 4

[0126] The embodiment of the present application further provides a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the target detection method provided in the first embodiment.

[0127] Optionally, in this embodiment, the above storage medium may be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group.

[0128] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing the program steps of the target detection method.

[0129] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0130] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0131] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0132] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0133] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0134] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0135] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A target detection method, characterized in that: include: receiving a target detection request, wherein the target detection request includes: a detection request for detecting a target in a computer room, the target including: a water leakage location; In response to the target detection request, an image of a target area in the computer room is collected to obtain a target image, wherein the target area includes: a top of the computer room; The target image is input into a target detector for N scale detections to obtain a target detection result, wherein the network structure of the target detector includes: a backbone network, a neck network and a head network, the backbone network includes at least: M C2f extraction modules for feature extraction, the C2f extraction module includes at least: a deformable convolutional neural network, the neck network includes at least: T C2f fusion modules for feature fusion, the C2f fusion module includes: group convolution and point-by-point convolution, the target detection result is used to indicate whether the target exists in the target image, and M, N and T are positive integers.

2. The target detection method according to claim 1, characterized in that: The backbone network further includes: a feature enhancement module, which is used to perform feature enhancement processing on the extracted features, input the target image into the target detector for N scale detection, and obtain the target detection result, including: Inputting the target image into the backbone network, performing feature extraction on the target image based on the feature enhancement module and the M C2f extraction modules, and obtaining N first feature maps, wherein the N first feature maps carry different amounts of information; Inputting the N first feature maps into the neck network, and performing feature fusion on the N first feature maps and the feature map generated in the neck network based on T C2f fusion modules to obtain N second feature maps, wherein the N second feature maps carry different amounts of information; Inputting the N second feature maps into the head network to perform N scale detections to obtain the target detection result.

3. The target detection method according to claim 2, characterized in that: The backbone network further includes: S CBS modules, each of which includes: a convolution layer, a batch normalization layer, and an activation function, where S is a positive integer, and features of the target image are extracted based on the feature enhancement module and the M C2f extraction modules to obtain N first feature maps, including: Based on a first CBS module among the S CBS modules and a second CBS module among the S CBS modules, performing feature extraction and nonlinear transformation on the target image to obtain a target feature map, wherein the first CBS module and the second CBS module are two adjacent CBS modules in the backbone network; Based on the S CBS modules and the M C2f extraction modules, feature extraction is performed on the target feature map, and feature enhancement processing is performed on the feature map obtained by feature extraction by the feature enhancement module to obtain N first feature maps.

4. The target detection method according to claim 3, characterized in that: The target feature map is subjected to feature extraction based on the S CBS modules and the M C2f extraction modules, and feature enhancement processing is performed on the feature map obtained by feature extraction by the feature enhancement module to obtain N first feature maps, including: Performing feature extraction on the target feature map based on a first C2f extraction module among the M C2f extraction modules, and inputting the extracted feature map into a third CBS module among the S CBS modules for feature extraction, so as to obtain a feature map output by the third CBS module; Inputting the feature map output by the third CBS module into a second C2f extraction module among the M C2f extraction modules for feature extraction, thereby obtaining a feature map output by the second C2f extraction module; Inputting the feature map output by the second C2f extraction module into a fourth CBS module among the S CBS modules for feature extraction, and inputting the feature map output by the fourth CBS module into a third C2f extraction module among the M C2f extraction modules for feature extraction, to obtain a feature map output by the third C2f extraction module; Inputting the feature map output by the third C2f extraction module into a fifth CBS module among the S CBS modules for feature extraction, and inputting the feature map output by the fifth CBS module into a fourth C2f extraction module among the M C2f extraction modules for feature extraction, to obtain a feature map output by the fourth C2f extraction module; Inputting the feature map output by the fourth C2f extraction module into the feature enhancement module for feature enhancement processing to obtain a feature map after feature enhancement; Determine N first feature maps based on the feature map output by the second C2f extraction module, the feature map output by the third C2f extraction module, and the feature map after feature enhancement.

5. The target detection method according to claim 4, characterized in that: The neck network further includes: F concatenation layers, H upsampling modules and P CBS modules, wherein F, H and P are positive integers, and based on T C2f fusion modules, feature fusion is performed on N first feature maps and feature maps generated in the neck network to obtain N second feature maps, including: Based on a first upsampling module among the H upsampling modules, upsampling the feature maps after feature enhancement in the N first feature maps is performed to obtain a feature map output by the first upsampling module; Based on a first splicing layer among the F splicing layers, splicing the feature map output by the first upsampling module and the feature map output by the third C2f extraction module among the N first feature maps to obtain a feature map output by the first splicing layer; Based on a first C2f fusion module among the T C2f fusion modules, perform feature fusion processing on the feature map output by the first splicing layer to obtain a feature map output by the first C2f fusion module; Performing upsampling processing on the feature map output by the first C2f fusion module based on a second upsampling module among the H upsampling modules, to obtain a feature map output by the second upsampling module; Based on the second splicing layer in the F splicing layers, the feature map output by the second up-sampling module and the feature map output by the second C2f extraction module in the N first feature maps are spliced ​​to obtain the feature map output by the second splicing layer, and based on the second C2f fusion module in the T C2f fusion modules, the feature map output by the second splicing layer is subjected to feature fusion processing to obtain the feature map output by the second C2f fusion module; Based on the feature map output by the first C2f fusion module, the feature map output by the second C2f fusion module, and the P CBS modules, N second feature maps are determined.

6. The target detection method according to claim 5, characterized in that: Determining N second feature maps based on the feature map output by the first C2f fusion module, the feature map output by the second C2f fusion module, and the P CBS modules includes: Based on the first target CBS module among the P CBS modules, feature extraction is performed on the feature map output by the second C2f fusion module to obtain the feature map output by the first target CBS module, and based on the third splicing layer among the F splicing layers, the feature map output by the first C2f fusion module and the feature map output by the first target CBS module are spliced ​​to obtain the feature map output by the third splicing layer; Based on the third C2f fusion module among the T C2f fusion modules, a feature fusion process is performed on the feature map output by the third splicing layer to obtain the feature map output by the third C2f fusion module; Based on the second target CBS module among the P CBS modules, feature extraction is performed on the feature map output by the third C2f fusion module to obtain the feature map output by the second target CBS module, and based on the fourth splicing layer among the F splicing layers, the feature map output by the second target CBS module and the feature map output by the feature enhancement module are spliced ​​to obtain the feature map output by the fourth splicing layer; Based on a fourth C2f fusion module among the T C2f fusion modules, perform feature fusion processing on the feature map output by the fourth splicing layer to obtain a feature map output by the fourth C2f fusion module; Based on the feature map output by the second C2f fusion module, the feature map output by the third C2f fusion module, and the feature map output by the fourth C2f fusion module, N second feature maps are determined.

7. The target detection method according to claim 1, characterized in that: The target detector is trained in the following way: Acquire training samples, wherein the training samples include: W historical images of the computer room, each of the historical images includes marking information of the water leakage area, the W historical images include: images of the computer room collected in a historical time period, and W is a positive integer; Model training is performed on the initial detector based on the training samples, and when the initial detector training converges, the converged initial detector is determined as the target detector, wherein the initial detector includes: a target detector that has not undergone model training.

8. A target detection device, characterized in that: include: A receiving unit, configured to receive a target detection request, wherein the target detection request includes: a detection request for detecting a target in a computer room, wherein the target includes: a water leakage location; A collection unit, configured to respond to the target detection request, collect an image of a target area in the computer room, and obtain a target image, wherein the target area includes: a top of the computer room; A detection unit is used to input the target image into a target detector for N scale detections to obtain a target detection result, wherein the network structure of the target detector includes: a backbone network, a neck network and a head network, the backbone network includes at least: M C2f extraction modules for feature extraction, the C2f extraction module includes at least: a deformable convolutional neural network, the neck network includes at least: T C2f fusion modules for feature fusion, the C2f fusion module includes: grouped convolution and point-by-point convolution, the target detection result is used to indicate whether the target exists in the target image, and M, N and T are positive integers.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored executable program, wherein when the executable program is executed, the device where the computer-readable storage medium is located is controlled to execute the target detection method according to any one of claims 1 to 7.

10. A computer program product comprising computer instructions, characterized in that: When the computer instructions are executed by a processor, the steps of the target detection method described in any one of claims 1 to 7 are implemented.

Images