Cabinet resource detection and identification method, system, device and medium

Abstract

The application discloses a cabinet resource detection and identification method, system, device and medium, belongs to the technical field of multi-target detection and information identification in the scene of machine room inspection, resource survey and the like, and aims to solve the technical problem of low idle U position utilization rate of a cabinet caused by untimely and low-efficiency traditional cabinet resource inspection and identification, and adopts the technical scheme that a target detection-classification dataset of a cabinet area is constructed, the dataset is preprocessed, a detection-classification fusion model model_fusion is constructed, the fusion model is trained, the detection-classification fusion model model_fusion is trained through a multi-target and multi-size target detection-classification training set, a result is output, the cabinet resource is inspected and identified through the trained detection-classification fusion model model_fusion, and idle U position information is output.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of multi-target detection and information identification in the scene of machine room inspection and resource survey, in particular to a cabinet resource detection and identification method, system, device and medium. BACKGROUND

[0002] With the rapid development of communication and Internet industry, various communication machine rooms, data machine rooms and computing power machine rooms have also ushered in a peak period of construction. Cabinet is the main component in the machine room. One of the important tasks in the management of the machine room is the general survey and change of cabinet resources, including cabinets, equipment on the shelves in the cabinets, and empty U positions in the cabinets. In the early stage of construction, cabinet assets are managed through drawings and asset registration books, but due to frequent changes in equipment in the cabinet, this non-information-based management method leads to many problems such as account confusion and inaccurate data.

[0003] With the continuous improvement of informatization, asset management systems are used to manage machine room resources, including cabinet resources, which avoids the problem of account confusion. However, the entry of information such as cabinet placement, changes and equipment installation on site is still manually entered by on-site construction personnel or relies on machine room planning, so there are still many inconsistencies between system data and on-site data.

[0004] To solve the above problems, the construction management App is used to allow on-site construction personnel to complete the cabinet placement, changes, equipment installation and other work after which the photos are uploaded and then verified and entered into the system by the back-end personnel. However, this method greatly increases the workload of back-end personnel, and because the photos are not standardized, some data is unclear or entered incorrectly. With the development of AI technology, AI models are used to identify information from photos, but due to the non-standardization of photos, the recognition accuracy is low, the process is complicated, and there are problems such as individual personnel using past photos to deceive. Currently, there are still many problems of asset and data inconsistency in many machine rooms, especially the inaccurate data of empty cabinets and empty U positions, which directly affects the external leasing prediction, revenue assessment and other problems of the machine room. SUMMARY

[0005] The technical task of the present application is to provide a cabinet resource detection and identification method, system, device and medium to solve the problem of low utilization rate of idle U positions caused by the low efficiency of traditional cabinet resource inspection and identification.

[0006] The technical task of the present application is achieved in the following way: a cabinet resource detection and identification method, which is specifically as follows:

[0007] Target detection-classification dataset of cabinet area is constructed: image collection is performed on cabinets, servers, switches, BBUs (indoor baseband processing units), RRUs (remote radio frequency modules) and U position numbers in the cabinet area by a fixed camera or a mobile phone camera, and target detection label annotation and U position number classification annotation are performed on the collected images by a labeImg annotation tool to construct a target detection-classification dataset of the cabinet area;

[0008] Dataset preprocessing: data augmentation is performed on various target images in the target detection-classification dataset of the cabinet area to ensure data balance of various target images, and various target images are segmented and normalized to the same size by size normalization to construct a multi-target and multi-size target detection-classification training set;

[0009] Construction of detection-classification fusion model model_fusion: a target detection network with optimized network structure is taken as a backbone, and a lightweight classification subnetwork is connected on the side to construct a detection-classification fusion model model_fusion; wherein the detection-classification fusion model model_fusion is a fusion model with shared features and parallel task heads for detection and classification;

[0010] Fusion model training: the detection-classification fusion model model_fusion is trained by using the multi-target and multi-size target detection-classification training set;

[0011] Result output: the cabinet resources are checked and recognized by using the trained detection-classification fusion model model_fusion, and idle U position information is output.

[0012] As an optimization, the detection-classification fusion model model_fusion is constructed as follows:

[0013] VGG16 is selected as the target detection network, the first convolution kernel of any two convolution layers in VGG16 is changed to a 2*2 convolution kernel, and the details of small targets of U position numbers are retained;

[0014] A lightweight classification subnetwork is added after the fourth convolution layer of VGG16, the input layer dimension restriction is removed, and the output nodes of the fully connected layer are changed to "U position number category number + 1";

[0015] The fusion model loss function is designed, the detection loss of VGG16 and the classification loss of LeNet are added, and the joint loss during training of the fusion model is obtained;

[0016] A fusion model inference mode is designed, and linear additivity is used to equivalently combine the convolution weights of LeNet into the corresponding layers of VGG16 to obtain a single-path network. The single-path network retains the accuracy of the trained detection-classification fusion model model_fusion.

[0017] More preferably, the lightweight classification sub-network uses LeNet, which has only 5 convolutional layers and can meet the classification of U-bit numbers with very small computational overhead.

[0018] More preferably, when the joint loss is less than the parameter p during the training of the detection-classification fusion model model_fusion, the training of the detection-classification fusion model model_fusion is completed, and the trained detection-classification fusion model model_fusion is obtained.

[0019] As a preferred embodiment, the trained detection-classification fusion model model_fusion is used to inspect and identify the cabinet resources, and the idle U-bit information is output as follows:

[0020] A single cabinet area is shot into an independent video, and the video data is frame-extracted at a time interval t to obtain a set of detection pictures;

[0021] A resource survey data list is constructed, and all contents in the resource survey data list are initialized to null values. The contents of the resource survey data list include cabinet number, cabinet purpose, installed equipment list in the cabinet, U-bit number list, and idle U-bit number.

[0022] The trained detection-classification fusion model model_fusion is used to detect the cabinet label target in the set of detection pictures, and the target area is intercepted and temporarily saved in the memory;

[0023] The trained detection-classification fusion model model_fusion is used to detect the multiple targets of servers, switches, BBUs, RRUs, and U-bit numbers in the cabinet in the set of detection pictures, and the detection results are aggregated and saved to the resource survey data list. The U-bit number target area is screenshot and temporarily saved in the memory;

[0024] The cabinet number and cabinet purpose information in the temporarily saved cabinet label target in the memory are identified, and the identification results are de-duplicated and aggregated. The de-duplicated and aggregated results are saved to the resource survey data list;

[0025] The U-bit number information in the temporarily saved U-bit number target in the memory is identified, and the identification results are de-duplicated and aggregated. The de-duplicated and aggregated results are saved to the resource survey data list;

[0026] Read the total U-bit number of the corresponding cabinet from the original resource library using the cabinet number information, and traverse the U-bit number list from number 1 according to the U-bit number list in the resource survey data list. When encountering a missing U-bit number, add 1 to the empty U-bit number. After traversing the U-bit number list in the resource survey data list, update the empty U-bit number, which is the total number of empty U-bits.

[0027] Read the resource information of the corresponding cabinet from the original resource library using the cabinet number information, and compare it with the cabinet resource information in the resource survey data list. Return the inconsistent survey items and output the corresponding resource survey results and empty U-bit information.

[0028] A cabinet resource detection and identification system, comprising:

[0029] A dataset construction module for image acquisition of multiple targets of cabinets, servers, switches, BBU (indoor baseband processing unit), RRU (remote radio frequency module) and U-bit numbers in the cabinet area by fixed cameras or mobile phone cameras, and target detection label annotation and U-bit number classification annotation of the collected images by labeImg annotation tool, to construct a target detection-classification dataset of the cabinet area.

[0030] A dataset preprocessing module for data enhancement of various target images in the target detection-classification dataset of the cabinet area to ensure balanced data of various target images, and segmentation and normalization of various target images to the same size by size normalization to construct a multi-target, multi-size target detection-classification training set.

[0031] A fusion model construction module for constructing a detection-classification fusion model model_fusion with a target detection network with optimized network structure as the backbone and a lightweight classification subnetwork connected on the side. The detection-classification fusion model model_fusion is a fusion model with shared features and parallel task heads for detection and classification.

[0032] A fusion model training module for training the detection-classification fusion model model_fusion through the multi-target, multi-size target detection-classification training set.

[0033] A result output module for checking and identifying cabinet resources through the trained detection-classification fusion model model_fusion and outputting idle U-bit information.

[0034] As a preferred approach, the fusion model construction module uses VGG16 as the object detection network. The first convolutional kernel of any two convolutional layers in VGG16 is changed to a 2x2 convolutional kernel to preserve the details of small targets with U-digit numbers. A lightweight classification sub-network is added after the fourth convolutional layer of VGG16, and the input layer dimension limitation is removed. The output node of the fully connected layer is changed to "number of U-digit number categories + 1". Simultaneously, a fusion model loss function is designed, adding the detection loss of VGG16 and the classification loss of LeNet as the joint loss during fusion model training. A fusion model inference mode is designed, utilizing linear additive property to equivalently merge the convolutional weights of LeNet back into the corresponding layers of VGG16, obtaining a single-path network. This single-path network preserves the accuracy of the detection-classification fusion model_fusion after training.

[0035] More preferably, the result output module includes:

[0036] The image set acquisition submodule is used to capture a single cabinet area as an independent video, and extract frames from the video data according to the time interval t to acquire the image set to be detected;

[0037] The Resource Census Data List Construction Submodule is used to construct the resource census data list and initialize all contents in the resource census data list to null values. The contents of the resource census data list include rack number, rack purpose, list of equipment installed in the rack, list of U-position numbers and number of vacant U-positions.

[0038] The cabinet target inspection submodule is used to detect cabinet label targets in the image set to be detected using the trained detection-classification fusion model_fusion, extract the target area, and temporarily store all of them in memory;

[0039] The U-position number target detection submodule is used to detect various targets in the rack, such as servers, switches, BBUs, RRUs and U-position numbers, in the image set to be inspected using the trained detection-classification fusion model_fusion. The detection results are aggregated and saved to the resource survey data list, and the U-position number target area is captured and temporarily stored in memory.

[0040] The cabinet label target recognition submodule is used to identify the cabinet number and cabinet usage information in the cabinet label targets temporarily stored in memory, deduplicate and aggregate the recognition results, and save the deduplicated and aggregated results to the resource survey data list.

[0041] The U-position number target identification submodule is used to identify the U-position number information in the U-position number targets temporarily stored in memory, deduplicate and aggregate the identification results, and save the deduplicated and aggregated results to the resource survey data list.

[0042] The U-bit vacancy total number statistics submodule is configured to read the total number of U-bits of the corresponding cabinet from the original resource library by using the cabinet number information, and traverse the U-bit number list from No. 1 according to the U-bit number list in the resource survey data list, and add 1 to the vacancy U-bit number when encountering a missing U-bit number; after traversing the U-bit number list in the resource survey data list, the vacancy U-bit number is updated to be the total number of U-bit vacancies;

[0043] The cabinet resource information correction submodule is configured to read the resource information of the corresponding cabinet from the original resource library by using the cabinet number information, and correct the cabinet resource information with the resource information in the resource survey data list, return the survey items with inconsistent data, and output the corresponding resource survey result and vacancy U-bit information.

[0044] An electronic device, comprising a memory and at least one processor;

[0045] The memory stores a computer program;

[0046] The at least one processor executes the computer program stored in the memory, so that the at least one processor executes the cabinet resource detection and identification method as described above.

[0047] A computer readable storage medium, the computer readable storage medium stores a computer program, the computer program can be executed by a processor to realize the cabinet resource detection and identification method as described above.

[0048] The cabinet resource detection and identification method, system, device and medium of the application have the following advantages:

[0049] (1) The application uses a network structure optimized target detection network as the backbone, and connects a lightweight classification subnetwork, designs a detection-classification fusion model sharing features and task heads in parallel, realizes fast and accurate detection and identification of various targets and U-bit numbers of the cabinet, judges how many vacancy U-bits exist through the U-bit number identification result, finally reads the related data of the computer room resource table to conduct resource survey, solves the problems of low efficiency of traditional cabinet resource survey and low utilization rate of cabinet idle U-bits, provides support for U-bit capacity early warning and rack planning, has the advantages of reducing labor cost, high precision, improving cabinet vacancy resource utilization rate and wide application, etc.

[0050] (II) This invention achieves fast and accurate detection and identification of various targets in the cabinet by designing a detection-classification fusion model_fusion, which improves the detection and identification accuracy of U-position numbered targets. Moreover, only one forward propagation calculation is required during inference, which greatly reduces the time consumption. The backbone network of the detection-classification fusion model_fusion can also be replaced with mainstream networks such as ResNet and MobileNet, and other lightweight networks can also be selected for the lightweight classification sub-network, which have similar improvement effects in recognition accuracy and inference speed.

[0051] (iii) This invention can be deployed on GPU devices. The detection-classification fusion model_fusion has low requirements for computing power and video memory resources, and can achieve low-cost and efficient resource survey. Attached Figure Description

[0052] The invention will be further described below with reference to the accompanying drawings.

[0053] Appendix Figure 1 This is a flowchart of a method for detecting and identifying server rack resources. Detailed Implementation

[0054] The cabinet resource detection and identification method, system, equipment and medium of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0055] Example 1: As shown in the attached document Figure 1 As shown in the figure, this embodiment provides a method for detecting and identifying rack resources, which is as follows:

[0056] S1. Constructing a target detection-classification dataset for the server rack area: Images of multiple targets in the server rack area, including server racks, servers, switches, BBU (Indoor Baseband Processing Unit), RRU (Remote Radio Unit), and U-position numbers, are acquired using a fixed camera or a mobile phone camera. The acquired images are then labeled with target detection tags and U-position number classification tags using the labeImg annotation tool to construct a target detection-classification dataset for the server rack area.

[0057] S2. Dataset preprocessing: Data augmentation is performed on various target images in the target detection-classification dataset of the rack area to ensure data balance of various target images. The target images of various types are segmented and normalized to the same size of 640*640 by size normalization to construct a multi-target, multi-size target detection-classification training set.

[0058] S3, build a detection-classification fusion model model_fusion: with the network structure optimized target detection network as the backbone, side joint lightweight classification subnetwork, build detection-classification fusion model model_fusion; wherein, detection-classification fusion model model_fusion is a detection and classification sharing features, task head parallel fusion model;

[0059] S4, fusion model training: training detection-classification fusion model model_fusion through multi-target, multi-size target detection-classification training set;

[0060] S5, result output: through the trained detection-classification fusion model model_fusion, check and identify the cabinet resources, and output idle U bit information.

[0061] The detection-classification fusion model model_fusion in step S3 of the embodiment is built as follows:

[0062] S301, select VGG16 as the target detection network, change the first convolution kernel of any two convolution layers in VGG16 to 2*2 convolution kernel, and keep the details of small targets with U bit number;

[0063] S302, add a lightweight classification subnetwork after the fourth convolution layer of VGG16, and remove the input layer dimension restriction, change the full connection layer output node to "U bit number category number +1";

[0064] S303, design fusion model loss function, add VGG16 detection loss and LeNet classification loss as joint loss during fusion model training;

[0065] S304, design fusion model inference mode, use linear additivity to equivalently combine LeNet convolution weight back to VGG16 corresponding layer, get single path network, and the single path network retains the precision of the trained detection-classification fusion model model_fusion.

[0066] The lightweight classification subnetwork in step S302 of the embodiment selects LeNet, LeNet only has 5 convolution layers, which can meet the classification of U bit number, and the calculation cost is very small.

[0067] When the joint loss is less than the parameter p during training of the detection-classification fusion model model_fusion in step S4 of the embodiment, the training of the detection-classification fusion model model_fusion is completed, and the trained detection-classification fusion model model_fusion is obtained.

[0068] The checking and identification of the cabinet resources in step S5 of the embodiment by the trained detection-classification fusion model model_fusion, and the output of the idle U-bit information are as follows:

[0069] S501, a single cabinet area is shot into an independent video, and the video data is frame-extracted according to a time interval t to obtain a set of to-be-detected pictures;

[0070] S502, a resource survey data list is constructed, and all contents in the resource survey data list are initialized as null values; wherein the contents of the resource survey data list include cabinet number, cabinet purpose, installed equipment list in the cabinet, U-bit number list, and idle U-bit number;

[0071] S503, the cabinet label target in the set of to-be-detected pictures is detected by the trained detection-classification fusion model model_fusion, the target area is intercepted, and all are temporarily saved in the memory;

[0072] S504, the multiple targets of the servers, switches, BBUs, RRUs, and U-bit numbers in the cabinet in the set of to-be-detected pictures are detected by the trained detection-classification fusion model model_fusion, the detection results are aggregated and saved to the resource survey data list, the U-bit number target area is intercepted, and all are temporarily saved in the memory;

[0073] S505, the cabinet number and cabinet purpose information in the cabinet label target temporarily saved in the memory are identified, the identification results are de-duplicated and aggregated, and the de-duplicated and aggregated results are saved to the resource survey data list;

[0074] S506, the U-bit number information in the U-bit number target temporarily saved in the memory is identified, the identification results are de-duplicated and aggregated, and the de-duplicated and aggregated results are saved to the resource survey data list;

[0075] S507, the total U-bit number of the corresponding cabinet is read from the original resource library by using the cabinet number information, and the U-bit number list is traversed from number 1 according to the U-bit number list in the resource survey data list, when a missing U-bit number is encountered, the idle U-bit number is increased by 1; after traversing the U-bit number list in the resource survey data list, the idle U-bit number is updated to be the total number of idle U-bits;

[0076] S508, the resource information of the corresponding cabinet is read from the original resource library by using the cabinet number information, and is collated with the cabinet resource information in the resource survey data list, the survey items with inconsistent data are returned, and the corresponding resource survey result and idle U-bit information are output.

[0077] Embodiment 2: The embodiment provides a cabinet resource detection and identification system, which comprises:

[0078] a dataset construction module configured to collect images of cabinets, servers, switches, BBUs (indoor baseband processing units), RRUs (remote radio frequency modules), and U-position numbers in a cabinet area through a fixed camera or a mobile phone camera, and to perform target detection label annotation and U-position number classification annotation on the collected images through a labeImg annotation tool, so as to construct a target detection-classification dataset of the cabinet area;

[0079] a dataset preprocessing module configured to perform data enhancement on target images of each type in the target detection-classification dataset of the cabinet area, to ensure data balance of the target images of each type, and to segment and normalize the target images of each type into the same size through size normalization, so as to construct a multi-target and multi-size target detection-classification training set;

[0080] a fusion model construction module configured to use a target detection network with optimized network structure as a backbone, and to connect a lightweight classification subnetwork to the side, so as to construct a detection-classification fusion model model_fusion; wherein the detection-classification fusion model model_fusion is a fusion model in which detection and classification share features and task heads in parallel;

[0081] a fusion model training module configured to train the detection-classification fusion model model_fusion through the multi-target and multi-size target detection-classification training set;

[0082] a result output module configured to check and identify cabinet resources through the trained detection-classification fusion model model_fusion, and to output idle U-position information.

[0083] In the fusion model construction module in this embodiment, VGG16 is selected as the target detection network, the first convolution kernel of any two convolution layers in VGG16 is changed to a 2*2 convolution kernel, and the details of small targets of U-position numbers are retained; a lightweight classification subnetwork is added after the fourth convolution layer of VGG16, and the input layer dimension restriction is removed, the output nodes of the fully connected layer are changed to “U-position number category number + 1”; a fusion model loss function is designed, the detection loss of VGG16 and the classification loss of LeNet are added, and the sum is used as the joint loss during training of the fusion model; a fusion model inference mode is designed, the convolution weights of LeNet are equivalently combined back to the corresponding layers of VGG16 by using linear additivity, a single path network is obtained, and the single path network retains the accuracy of the detection-classification fusion model model_fusion after training.

[0084] The result output module in this embodiment includes:

[0085] The to-be-detected picture set obtaining submodule is configured to shoot a single cabinet area into an independent video, and frame the video data according to a time interval t to obtain a to-be-detected picture set.

[0086] The resource survey data list constructing submodule is configured to construct a resource survey data list, and initialize all contents in the resource survey data list as null values; wherein the contents of the resource survey data list include cabinet number, cabinet purpose, installed equipment list in the cabinet, U position number list and vacant U position number.

[0087] The cabinet target checking submodule is configured to detect cabinet label targets in the to-be-detected picture set by using the trained detection-classification fusion model model_fusion, intercept target regions, and temporarily save all the target regions in the memory.

[0088] The U position number target detecting submodule is configured to detect multiple targets including servers, switches, BBUs, RRUs and U position numbers in the cabinet in the to-be-detected picture set by using the trained detection-classification fusion model model_fusion, save the detection results to the resource survey data list after aggregation, intercept U position number target regions, and temporarily save all the target regions in the memory.

[0089] The cabinet label target recognizing submodule is configured to recognize cabinet number and cabinet purpose information in the cabinet label targets temporarily saved in the memory, remove and aggregate the recognition results, and save the removed and aggregated results to the resource survey data list.

[0090] The U position number target recognizing submodule is configured to recognize U position number information in the U position number targets temporarily saved in the memory, remove and aggregate the recognition results, and save the removed and aggregated results to the resource survey data list.

[0091] The U position vacant total number counting submodule is configured to read total U position number of a corresponding cabinet from an original resource library by using cabinet number information, and traverse the U position number list from the U position number list in the resource survey data list starting from number 1, and add 1 to the vacant U position number when a missing U position number is encountered; after traversing the U position number list in the resource survey data list, the vacant U position number is updated to be the U position vacant total number.

[0092] The cabinet resource information checking submodule is configured to read resource information of a corresponding cabinet from an original resource library by using cabinet number information, and check the resource information with cabinet resource information in the resource survey data list, return survey items with inconsistent data, and output corresponding resource survey results and vacant U position information.

[0093] Embodiment 3: The embodiment also provides an electronic device, including a memory and a processor.

[0094] wherein the memory stores computer-executable instructions;

[0095] The processor executes the computer-executable instructions stored in the memory, so that the processor executes the cabinet resource detection and identification method in any embodiment of the present application.

[0096] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), ready programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The processor can be a microprocessor or any conventional processor.

[0097] The memory can be used to store computer programs and / or modules, and the processor realizes various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory, and calling data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store operating systems, application programs required by at least one function, etc.; the data storage area can store data created according to the use of the terminal, etc. In addition, the memory can also include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a secure digital (SD) card, a flash memory card, at least one magnetic disk storage period, a flash memory device, or other volatile solid-state memory devices.

[0098] Embodiment 4: The embodiment also provides a computer-readable storage medium, wherein a plurality of instructions are stored, the instructions are loaded by a processor, so that the processor executes the cabinet resource detection and identification method in any embodiment of the present application. Specifically, a system or device provided with a storage medium can be provided, and software program code for realizing the functions of any one of the above embodiments is stored on the storage medium, and the computer (or CPU or MPU) of the system or device reads out and executes the program code stored in the storage medium.

[0099] In this case, the program code read from the storage medium can realize the functions of any one of the above embodiments, and therefore the program code and the storage medium storing the program code constitute a part of the present application.

[0100] The storage medium for providing the program code includes a floppy disk, a hard disk, a magneto-optical disk, an optical disk (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RYM, DVD-RW, DVD+RW), a magnetic tape, a non-volatile memory card, and a ROM. Alternatively, the program code can be downloaded from a server computer through a communication network.

[0101] Furthermore, it should be understood that the functions of the above-described embodiments can be achieved by not only the execution of the program code read by a computer, but also by the operation of an operating system or the like on the computer under the control of the instructions of the program code to perform part or all of the actual processing, and thus the functions of the above-described embodiments are realized.

[0102] Furthermore, it should be understood that the functions of the above-described embodiments can be achieved by not only the execution of the program code read by a computer, but also by the operation of an operating system or the like on the computer under the control of the instructions of the program code to perform part or all of the actual processing, and thus the functions of the above-described embodiments are realized.

[0103] Finally, it should be noted that the above-described embodiments are merely used to illustrate the technical solutions of the present application, rather than limit the present application; even though the present application has been described in detail with reference to the above embodiments, those ordinarily skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or equivalently replace part or all of the technical features thereof; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for detecting and identifying cabinet resources, characterized in that, The method is as follows: Constructing a target detection-classification dataset of the cabinet area: through a fixed camera or a mobile phone camera, image collection is performed on the cabinets, servers, switches, BBUs, RRUs and U position numbers in the cabinet area, and through the labeImg labeling tool, target detection label labeling and U position number classification labeling are performed on the collected images to construct a target detection-classification dataset of the cabinet area; Data set preprocessing: data augmentation is performed on the target detection-classification dataset of the cabinet area to ensure the data balance of each type of target image, and through size normalization, each type of target image is segmented and normalized to the same size to construct a multi-target, multi-size target detection-classification training set; Constructing a detection-classification fusion model model_fusion: taking a target detection network with optimized network structure as the backbone, a lightweight classification subnetwork is connected on the side to construct a detection-classification fusion model model_fusion; Fusion model training: the detection-classification fusion model model_fusion is trained through the multi-target, multi-size target detection-classification training set; Result output: through the trained detection-classification fusion model model_fusion, the cabinet resources are checked and recognized, and the idle U position information is outputted; Wherein, the construction of the detection-classification fusion model model_fusion is as follows: Select VGG16 as the target detection network, change the first convolution kernel of any two convolution layers in VGG16 to 2 2 convolution kernel, keep the details of small targets with U-bit number A lightweight classification subnetwork is added after the fourth convolutional layer of VGG16, and the input layer dimension restriction is removed, and the output nodes of the fully connected layer are changed to "U position number category number + 1"; Designing a fusion model loss function, adding the detection loss of VGG16 and the classification loss of LeNet as the joint loss during the training of the fusion model; Designing a fusion model inference mode, using linear additivity to equivalently combine the convolutional weights of LeNet back to the corresponding layers of VGG16 to obtain a single path network, which retains the accuracy of the detection-classification fusion model model_fusion after training; Through the trained detection-classification fusion model model_fusion, the cabinet resources are checked and recognized, and the idle U position information is outputted as follows: A single cabinet area is shot into an independent video, and the video data is frame extracted according to the time interval t to obtain a set of detection pictures; Constructing a resource survey data list and initializing all contents in the resource survey data list to null; wherein, the contents of the resource survey data list include cabinet number, cabinet purpose, cabinet installed equipment list, U position number list and idle U position number; Through the trained detection-classification fusion model model_fusion, the cabinet label targets in the detection picture set are detected, the target area is intercepted, and all are temporarily saved in the memory; The trained detection-classification fusion model model_fusion is used to detect the multiple targets of the servers, switches, BBUs, RRUs and U-bit numbers in the cabinet in the to-be-inspected picture set, and the detection results are aggregated and saved into the resource census data list, the U-bit number target area is screenshot, and all are temporarily saved in the memory; The cabinet number and cabinet purpose information in the cabinet label target temporarily saved in the memory are recognized, the recognition results are de-duplicated and aggregated, and the de-duplicated and aggregated results are saved into the resource census data list; The U-bit number information in the U-bit number target temporarily saved in the memory is recognized, the recognition results are de-duplicated and aggregated, and the de-duplicated and aggregated results are saved into the resource census data list; The total U-bit number of the corresponding cabinet is read from the original resource library by using the cabinet number information, and the U-bit number list in the resource census data list is traversed from number 1, and when a missing U-bit number is encountered, the empty U-bit number is increased by 1; after traversing the U-bit number list in the resource census data list, the empty U-bit number is updated to be the total number of empty U-bits. The resource information of the corresponding cabinet is read from the original resource library by using the cabinet number information, and is compared with the cabinet resource information in the resource census data list, and the census items with inconsistent data are returned and the corresponding resource census results and empty U-bit information are output.

2. The cabinet resource detection and identification method of claim 1, wherein, The LeNet is selected for the lightweight classification sub-network, and the LeNet only has 5 convolutional layers, which can meet the classification of the U-bit number.

3. The cabinet resource detection and identification method according to claim 1 or 2, characterized in that, When the joint loss is less than the parameter p during the training of the detection-classification fusion model model_fusion, the training of the detection-classification fusion model model_fusion is completed, and the trained detection-classification fusion model model_fusion is obtained.

4. A cabinet resource detection and identification system, characterized by, The system comprises: A data set construction module is configured to perform image acquisition on multiple targets of cabinets, servers, switches, BBUs, RRUs and U-bit numbers in a cabinet area by using a fixed camera or a mobile phone camera, and perform target detection label annotation and U-bit number classification annotation on the acquired images by using a labeImg annotation tool, so as to construct a target detection-classification data set of the cabinet area. A data set preprocessing module is configured to perform data enhancement on images of various targets in the target detection-classification data set of the cabinet area, to ensure data balance of the images of various targets, and to segment and normalize the images of various targets into the same size by size normalization, so as to construct a multi-target and multi-size target detection-classification training set. A fusion model construction module is configured to use a target detection network with optimized network structure as a backbone, and connect a lightweight classification sub-network thereto, to construct a detection-classification fusion model model_fusion. A fusion model training module is configured to train the detection-classification fusion model model_fusion by using the multi-target and multi-size target detection-classification training set. A result output module is configured to perform inspection and recognition on cabinet resources by using the trained detection-classification fusion model model_fusion, and output idle U-bit information. Wherein, the fusion model construction module selects VGG16 as the target detection network, changes the first convolution kernel of any two convolution layers in VGG16 to 2 2Convolution kernel, keep the details of small targets with U-bit number; and add a lightweight classification sub-network after the fourth convolution layer of VGG16, remove the input layer dimension restriction, change the output node of the fully connected layer to "U-bit number of categories + 1"; meanwhile, design the fusion model loss function, add the detection loss of VGG16 and the classification loss of LeNet as the joint loss during the training of the fusion model; design the inference mode of the fusion model, use linear additivity to equivalently combine the convolution weight of LeNet back to the corresponding layer of VGG16, obtain a single path network, and the single path network retains the precision of the detection-classification fusion model model_fusion after training; The result output module comprises: The to-be-detected picture set obtaining submodule is configured to shoot a single cabinet area into an independent video, and frame the video data at a time interval t to obtain a to-be-detected picture set; The resource survey data list construction submodule is configured to construct a resource survey data list, and initialize all contents in the resource survey data list as null values; wherein the contents of the resource survey data list include cabinet number, cabinet purpose, cabinet-mounted equipment list, U-bit number list, and empty U-bit number; The cabinet target checking submodule is configured to detect cabinet label targets in the to-be-detected picture set by using the trained detection-classification fusion model model_fusion, intercept target regions, and temporarily save all the target regions in the memory; The U-bit number target detection submodule is configured to detect multiple targets including cabinet servers, switches, BBUs, RRUs, and U-bit numbers in the cabinet in the to-be-detected picture set by using the trained detection-classification fusion model model_fusion, save the detection results to the resource survey data list after aggregation, and temporarily save all the U-bit number target regions in the memory; The cabinet label target recognition submodule is configured to recognize cabinet number and cabinet purpose information in the cabinet label targets temporarily saved in the memory, remove and aggregate the recognition results, and save the removed and aggregated results to the resource survey data list; The U-bit number target recognition submodule is configured to recognize U-bit number information in the U-bit number targets temporarily saved in the memory, remove and aggregate the recognition results, and save the removed and aggregated results to the resource survey data list; The U-bit empty total number statistics submodule is configured to read total U-bit numbers of corresponding cabinets from an original resource library by using cabinet number information, and traverse the U-bit number list from the U-bit number list in the resource survey data list starting from number 1, add 1 to the empty U-bit number when a missing U-bit number is encountered, and update the empty U-bit number as the U-bit empty total number after traversing the U-bit number list in the resource survey data list. The cabinet resource information correction submodule is configured to read resource information of corresponding cabinets from the original resource library by using cabinet number information, correct the cabinet resource information in the resource survey data list, return survey items with inconsistent data, and output corresponding resource survey results and empty U-bit information.

5. An electronic device, comprising: comprise: a memory and at least one processor; wherein the memory has stored thereon a computer program; the at least one processor executes the computer program stored in the memory, so that the at least one processor executes the cabinet resource detection and identification method according to any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, The computer program stored in the computer readable storage medium can be executed by the processor to implement the cabinet resource detection and identification method according to any one of claims 1 to 3.

Images