Spare part management method and system based on target identification and change detection

Through the target recognition and change detection technology based on the YOLO algorithm, the problem that the existing technology cannot manage complex inventory environments is solved, and the automated identification and efficient management of a variety of commodities are achieved, and the accuracy and efficiency of inventory management are improved.

CN119005857APending Publication Date: 2024-11-22FUJIAN NETPOWER TECH DEV CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411025078.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-29
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The existing technology cannot effectively manage complex inventory environments. A device can only monitor the quantity of one commodity and cannot cope with the inventory of multiple commodities.

Method used

The spare parts management system based on the YOLO algorithm is adopted, including image acquisition, preprocessing, object detection and log management modules, and the object detection is carried out through multi-layer convolutional neural network and reinforcement learning model to achieve multi-target area detection. and confirmation of inbound and outbound information.

Benefits of technology

It realizes automated, accurate identification and efficient management of spare parts, improves the accuracy and efficiency of inventory management, can monitor and record inventory changes in real time, and reduces the possibility of human error.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119005857A_ABST
    Figure CN119005857A_ABST
Patent Text Reader

Abstract

The invention relates to a spare part management method and system based on target identification and change detection, and the method comprises the following steps: collecting warehouse-in and warehouse-out images or videos of spare parts through a camera disposed in a warehouse, and transmitting the image information to an image processing assembly; preprocessing the input image or video, wherein the preprocessing comprises denoising, normalization and enhancement operation; according to the input preprocessed image or video, through an image processing algorithm and a reinforcement learning model pair, warehouse-in and warehouse-out information of the spare parts is confirmed; processing the output warehouse-in and warehouse-out information of the spare parts into a format which can be identified by the log management module, and outputting the processed information to the log management module; the log management module records detailed logs of system operations, including user operations, system events and processing results, provides operation records and problem tracking functions, and helps system maintenance and troubleshooting; and the user interaction module can query and correct the in-out storage history and real-time image and result information of the spare parts.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of information processing technology, and more specifically, to a spare parts management method and system based on target recognition and change detection. Background Art

[0002] With the development of society and the rapid growth of the national economy, the number and types of materials managed by the State Grid Corporation are also increasing. As a result, the difficulty of material management has increased, including the management of procurement, warehousing, outbound delivery, inventory and other links, which require more scientific and efficient methods to manage. At present, the main solution for the intelligent management of the spare parts warehouse of the State Grid Corporation is to introduce RFID technology to identify and manage materials, so as to realize the automation and precision of material management. This technology requires the affixing of RFID labels to each item, and then the asset management is managed through the system file creation. The management of the entry and exit of items is completed by using the RFID reader.

[0003] The present invention realizes high-accuracy real-time intelligent images for target recognition and change detection based on the YOLO algorithm. The YOLO algorithm has excellent real-time performance and accuracy. By fine-tuning and optimizing YOLO, the recognition capability of multiple target areas can be significantly improved. The present invention is applied to the spare parts material management system of the State Grid to realize image-free management of spare parts, automatically identify asset problems and basic information of circulating backup parts, and provide functions such as intelligent and efficient retrieval, high precision, and high-quality inspection.

[0004] Prior art, such as a Chinese patent with patent number "CN117151595A", discloses a commodity inventory management method, device and storage medium. The method includes: obtaining a commodity inventory image in a warehouse taken by a camera; inputting the commodity inventory image into a commodity recognition model for recognition processing to obtain the monitored inventory of the target commodity; determining the current inventory of the target commodity based on the registered inventory of the target commodity and the monitored inventory; in response to the confirmation operation of the current inventory, calling a sales forecast model; determining the predicted sales of the target commodity within a preset time in the future through the sales forecast model; determining the planned outbound quantity or planned inbound quantity of the target commodity within the preset time in the future based on the current inventory and the predicted sales. The present invention can obtain a more accurate and reasonable current inventory, and can accurately predict the future sales of commodities, obtain the planned outbound quantity or planned inbound quantity, and ensure that the inventory can respond to matching commodity sales in a timely manner.

[0005] The problem with the above-mentioned prior art is that when this method recognizes the images of goods in inventory, one device can only monitor the quantity of one kind of goods and cannot cope with complex inventory environments. Summary of the invention

[0006] To solve the above technical problems, the present invention proposes.

[0007] The technical solution of the present invention is as follows:

[0008] On the one hand, the present invention proposes a spare parts management system based on target recognition and change detection, comprising:

[0009] The spare parts detection module includes an image acquisition unit, an image preprocessing unit, a target detection unit, and a post-processing unit, wherein:

[0010] Image acquisition unit: collects images or videos of spare parts entering and leaving the warehouse through the image acquisition equipment installed in the warehouse, and transmits the image or video information to the image preprocessing unit through the communication interface;

[0011] Image preprocessing unit: preprocess the input image or video information, including denoising, normalization and enhancement operations;

[0012] Target detection unit: Based on the input pre-processed image or video information, the image processing algorithm and reinforcement learning model are used to confirm the in and out information of spare parts;

[0013] Post-processing unit: processes the outputted storage and in-and-out information of spare parts into a format that can be recognized by the log management module, and outputs it to the log management module;

[0014] Log management module: records detailed logs of system operations, including user operations, system events, and processing results;

[0015] User interaction module: used to query the history of spare parts in and out of the warehouse and real-time images and result information; to select different image acquisition devices; to display the current system status, receive user input, and configure system parameters.

[0016] As a preferred implementation, the image preprocessing unit preprocesses the input image or video information, specifically comprising the following steps:

[0017] The specific process of denoising is as follows:

[0018] Mean filtering: For each pixel in the image, its new pixel value is the average value of its neighborhood pixel values; if the neighborhood size is m×n, then the new value of the pixel (x, y) I new (x,y) is:

[0019]

[0020] Where I(i,j) represents the original pixel value at pixel point (i,j);

[0021] Median filtering: sort the pixel values ​​in the neighborhood and take the middle value as the new pixel value;

[0022] The specific formula of the normalization process is as follows:

[0023]

[0024] Where I(x,y) is the original pixel value, I max and I min are the minimum and maximum pixel values ​​in the image, respectively;

[0025] The specific formula of the enhancement process is as follows:

[0026] Brightness adjustment: I brightened (x,y)=I(x,y)+b; where b is the brightness adjustment value;

[0027] Contrast adjustment: I contrasted (x,y)=a×I(x,y)+b; where a controls contrast and b controls brightness.

[0028] As a preferred implementation, the target detection unit confirms the in-and-out information of spare parts through image processing algorithms and reinforcement learning models, including the following steps:

[0029] Network forward propagation: Process the preprocessed image through a multi-layer convolutional neural network to extract multi-layer features; use the Backbone deep neural network to extract the basic features of the image, and use the pyramid pooling module PPM to capture multi-scale information, integrate the feature pyramid network FPN to fuse feature maps at different levels, and combine the adaptive spatial feature network ASFF to improve the feature fusion effect; finally, use the multi-scale detection head Multi-scale Detection Head to perform target detection at multiple scales to adapt to targets of different sizes; during the training process, optimize and improve the Mosaic data enhancement strategy, enhance the generalization ability of the model, and dynamically adjust the learning rate to improve training efficiency and model performance;

[0030] Prediction output: The output of the network is a fixed-size grid, where each grid cell predicts the bounding box, confidence score, and class probability of the object;

[0031] Decoding output: Decode the prediction results output by the network into the actual target bounding box and category; convert the relative coordinates into the actual coordinates of the image, filter out low-confidence bounding boxes according to the confidence threshold, remove repeated overlapping bounding boxes, retain the highest confidence bounding box, and avoid multiple detections of the same target;

[0032] Post-processing: Perform subsequent processing on the final target detection results; draw the detected target bounding box and category label on the original image, and output the detection results to the log management module for storage.

[0033] As a preferred implementation, the network forward propagation process specifically includes the following steps:

[0034] The specific formula for the convolution operation is as follows:

[0035]

[0036] Where: I in and I out are the input and output images respectively, is the convolution kernel, k depends on the size of the convolution kernel;

[0037] The specific formula of the pyramid pooling module PPM is as follows:

[0038]

[0039] Where: F pooled is the feature map obtained under different pooling scales d, s is the pooling window size;

[0040] The feature map F after feature pyramid network FPN fusion is expressed as:

[0041] F=H+L

[0042] Where: H is the high-level feature map, L is the upsampled low-level feature map;

[0043] The specific formula of adaptive spatial feature network ASFF is as follows:

[0044] F ASFF =w1F1+w2F2+,…,+w c F c

[0045] In the formula, F c is the cth fused feature map, w c corresponds to the adaptive weights obtained through network learning;

[0046] The specific formula of Multi-scale Detection Heads is as follows:

[0047] x=x a +Δx,y=y a +Δy,z=z a +Δz,h=h a +Δh

[0048] Where: x a ,ya 、z a 、h a are the original anchor box coordinates, Δx, Δy, Δz, Δh are the predicted bounding box offsets.

[0049] As a preferred implementation, the prediction output process specifically includes the following steps:

[0050] Coordinate transformation:

[0051] x=x grid ×stride+x offset

[0052] y=y grid ×stride+y offset

[0053] Where: x grid ,y grid is the index of the grid cell; stride is the step size of the cell; x offset ,y offset is the predicted offset;

[0054] The confidence Softmax calculation formula is as follows:

[0055]

[0056] Where: p o is the probability category; S o is the confidence score of the o-th category; o = 1, 2, 3, …, b.

[0057] As a preferred implementation, in the decoding output process, low-confidence bounding boxes are filtered out according to the confidence threshold; this is achieved by a non-maximum suppression algorithm, and the specific formula is as follows:

[0058]

[0059] Where: S o is the confidence score; M is the bounding box with the highest score; b v is the bounding box to be processed; IoU(M,b v ) is the bounding box M and b v The intersection-and-union ratio of t is the set intersection-over-union ratio threshold;

[0060] Then we have:

[0061]

[0062] Where: area(M∩b v ) is the intersection area of ​​the two bounding boxes; area(M∪bv ) is the area of ​​the union of the two bounding boxes.

[0063] As a preferred implementation, the user interaction module can query the spare parts in and out of the warehouse history and real-time images and result information. By manually comparing the output results with the images, error information can be discovered in a timely manner, and machine reinforcement learning can be fed back through the user interaction module.

[0064] On the other hand, the present invention also provides a spare parts management method based on target recognition and change detection, comprising the following steps:

[0065] Step S1: Collect images or videos of spare parts entering and leaving the warehouse through a camera installed in the warehouse, and transmit the image information to an image processing component through a communication interface;

[0066] Step S2: preprocessing the input image or video, including denoising, normalization and enhancement operations;

[0067] Step S3: According to the input pre-processed image or video, the in-and-out information of spare parts is confirmed through the image processing algorithm and the reinforcement learning model;

[0068] Step S4: Processing the outputted storage and in-and-out information of spare parts into a format recognizable by the log management module, and outputting it to the log management module;

[0069] Step S5: The log management module records detailed logs of system operations, including user operations, system events, and processing results; provides operation records and problem tracking functions to help system maintenance and troubleshooting;

[0070] Step S6: Through the user interaction module, the spare parts in and out history and real-time images and result information can be queried; different image devices can be selected; the current status can be displayed, user input can be accepted, and system parameters can be configured.

[0071] On the other hand, the present invention further provides an electronic device having a computer program stored thereon, wherein when the computer program is executed by a processor, the method for managing spare parts based on target recognition and change detection as described in any embodiment of the present invention is implemented.

[0072] On the other hand, the present invention also provides a computer-readable medium for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement a spare parts management method based on target recognition and change detection as described in any embodiment of the present invention.

[0073] The present invention has the following beneficial effects:

[0074] Automation and efficiency:

[0075] The use of cameras to automatically capture images or videos of spare parts entering and leaving the warehouse reduces the tedious work of manual recording and inventory, greatly improving management efficiency.

[0076] The entire process from image acquisition, pre-processing, detection to post-processing and logging is fully automated, reducing the possibility of human error.

[0077] Accurate object detection:

[0078] Through image processing algorithms and reinforcement learning models, the inventory information of spare parts can be confirmed more accurately, which improves the accuracy and reliability of detection compared to traditional methods that rely on manual judgment or simple recognition methods.

[0079] Complete log management:

[0080] Logs that record system operations in detail, including user operations, system events, and processing results, provide strong support for system maintenance and troubleshooting. Existing technologies may be insufficient in terms of the completeness and detail of log records.

[0081] Good user interactivity:

[0082] The user interaction module allows users to query the historical and real-time information of spare parts entering and leaving the warehouse, select different image devices, and configure system parameters. This allows users to use the system more flexibly and conveniently to meet personalized needs, while some existing systems may not be friendly and flexible enough in terms of user interaction.

[0083] Real-time and dynamic monitoring:

[0084] It can collect and process images in real time, obtain the inventory status of spare parts in time, realize dynamic monitoring and management, and reflect inventory changes more timely than the traditional regular inventory method.

[0085] Data traceability:

[0086] Complete log records and clear processing of inbound and outbound information ensure data traceability and facilitate review and analysis of the process and results of inventory management, which may not be prominent enough in existing technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0088] Figure 1 It is a schematic diagram of the system flow of the present invention;

[0089] Figure 2 The figure is a flow chart of the spare parts detection module in the first embodiment. DETAILED DESCRIPTION

[0090] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0091] It should be understood that the step numbers used in this document are only for convenience of description and are not intended to limit the order in which the steps are executed.

[0092] It should be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0093] The terms “include” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0094] The term "and / or" means and includes any and all possible combinations of one or more of the associated listed items.

[0095] Embodiment 1:

[0096] In order to make the purpose, technical solutions and advantages of the present invention more clear, the following will be combined with the specific embodiments of the present application and refer to the attached Figure 1 , clearly and completely describe the technical solution of the present invention.

[0097] In order to solve the problems in the prior art, the present invention provides a spare parts management system based on target recognition and change detection, comprising:

[0098] The spare parts detection module includes an image acquisition unit, an image preprocessing unit, a target detection unit, and a post-processing unit, wherein:

[0099] Image acquisition unit: collects images or videos of spare parts entering and leaving the warehouse through the image acquisition equipment installed in the warehouse, and transmits the image or video information to the image preprocessing unit through the communication interface;

[0100] The collected spare parts need to be stored in smart cabinets and laid out in layers. There are no blind spots in identification caused by stacking and no clear identification due to dense placement.

[0101] Image preprocessing unit: preprocess the input image or video information, including denoising, normalization and enhancement operations; the specific operation process is as follows:

[0102] Denoising:

[0103] Mean filtering: For each pixel in the image, its new pixel value is the average value of its neighborhood pixel values; if the neighborhood size is m×n, then the new value of the pixel (x, y) I new (x,y) is:

[0104]

[0105] Where I(i,j) represents the original pixel value at pixel point (i,j);

[0106] Median filtering: sort the pixel values ​​in the neighborhood and take the middle value as the new pixel value;

[0107] The specific formula of the normalization process is as follows:

[0108]

[0109] Where I(x,y) is the original pixel value, I max and I min are the minimum and maximum pixel values ​​in the image, respectively;

[0110] The specific formula of the enhancement process is as follows:

[0111] Brightness adjustment: I brightened (x,y)=I(x,y)+b; where b is the brightness adjustment value;

[0112] Contrast adjustment: I contrasted (x,y)=a×I(x,y)+b; where a controls contrast and b controls brightness.

[0113] Target detection unit: Based on the input pre-processed image or video, the image processing algorithm and reinforcement learning model are used to confirm the in and out information of spare parts. The specific process is as follows:

[0114] Network forward propagation: Process the preprocessed image through a multi-layer convolutional neural network to extract multi-layer features; use the Backbone deep neural network to extract the basic features of the image, and capture multi-scale information by using the pyramid pooling module PPM, integrate the feature pyramid network FPN to fuse feature maps at different levels, and combine the adaptive spatial feature network ASFF to improve the feature fusion effect; finally, use the multi-scale detection head Multi-scale Detection Head to perform target detection at multiple scales to adapt to targets of different sizes; during the training process, optimize and improve the Mosaic data enhancement strategy, enhance the generalization ability of the model, and dynamically adjust the learning rate to improve training efficiency and model performance; specifically include the following steps:

[0115] The specific formula for the convolution operation is as follows:

[0116]

[0117] Where: I in and I out are the input and output images respectively, is the convolution kernel, k depends on the size of the convolution kernel;

[0118] The specific formula of the pyramid pooling module PPM is as follows:

[0119]

[0120] Where: F pooled is the feature map obtained under different pooling scales d, s is the pooling window size;

[0121] The feature map F after feature pyramid network FPN fusion is expressed as:

[0122] F=H+L

[0123] Where: H is the high-level feature map, L is the upsampled low-level feature map;

[0124] The specific formula of adaptive spatial feature network ASFF is as follows:

[0125] F ASFF =w1F1+w2F2+,…,+w c F c

[0126] In the formula, F c is the cth fused feature map, w c corresponds to the adaptive weights obtained through network learning;

[0127] The specific formula of Multi-scale Detection Heads is as follows:

[0128] x=x a +Δx,y=y a +Δy,z=z a +Δz,h=h a +Δh

[0129] Where: x a ,y a 、z a 、h a are the original anchor box coordinates, Δx, Δy, Δz, Δh are the predicted bounding box offsets.

[0130] Optimize and improve the Mosaic data augmentation strategy to enhance the generalization ability of the model. The specific methods are as follows:

[0131] The Mosaic data enhancement algorithm combines multiple images into one image in a certain proportion, allowing the model to identify targets in a smaller range. First, four images are randomly selected according to the coordinates of the reference points, and are resized and scaled to the upper left, upper right, lower left, and lower right of the specified large image. Secondly, according to the size transformation method of each image, the mapping relationship is mapped to the image label. Finally, according to the specified horizontal and vertical coordinates, the large image is spliced ​​and the detection box coordinates that exceed the boundary are processed. The Mosaic data enhancement algorithm can effectively enhance data diversity and model robustness, which is conducive to improving the performance of small target detection.

[0132] During the training process, the Mosaic data enhancement strategy is continuously improved by optimizing and adjusting hyperparameters. First, setting img-size to 980*980 is conducive to improving training results under ideal training time and hardware conditions. Secondly, for batch-size, use the maximum value of the hardware to avoid the statistical result offset caused by too small batch-size value. Then, adjust the number of epochs iterations based on the loss curve after the initial training structure to avoid underfitting caused by too small data and overfitting caused by too large values. Next, choose to turn on cache to speed up training and reduce training time. Finally, set random seeds to make the training process repeatable and controllable.

[0133] Prediction output: The output of the network is a fixed-size grid, where each grid cell predicts the bounding box, confidence score, and class probability of the target; the following steps are involved:

[0134] Coordinate transformation:

[0135] x=x grid ×stride+x offset

[0136] y=y grid×stride+y offset

[0137] Where: x grid ,y grid is the index of the grid cell; stride is the step size of the cell; x offset ,y offset is the predicted offset;

[0138] The confidence Softmax calculation formula is as follows:

[0139]

[0140] Where: p o is the probability category; S o is the confidence score of the o-th category; o = 1, 2, 3, …, b.

[0141] Filter out low-confidence bounding boxes based on the confidence threshold; this is achieved through the non-maximum suppression algorithm. The specific formula is as follows:

[0142]

[0143] Where: S o is the confidence score; M is the bounding box with the highest score; b v is the bounding box to be processed; IoU(M,b v ) is the bounding box M and b v The intersection-and-union ratio of t is the set intersection-over-union ratio threshold;

[0144] Then we have:

[0145]

[0146] Where: area(M∩b v ) is the intersection area of ​​the two bounding boxes; area(M∪b v ) is the area of ​​the union of the two bounding boxes.

[0147] Post-processing unit: processes the outputted storage and in-and-out information of spare parts into a format that can be recognized by the log management module, and outputs it to the log management module;

[0148] Log management module: records detailed logs of system operations, including user operations, system events, and processing results;

[0149] User interaction module: used to query the history of spare parts in and out of the warehouse and real-time images and result information; to select different image acquisition devices; to display the current system status, receive user input, and configure system parameters.

[0150] The user interaction module is the front-end part of the system, responsible for interacting with users and providing an operation interface. Its main functions include displaying the current status of the system, accepting user input, and configuring system parameters. It includes the following specific functions:

[0151] 1. Display simulation switching: allows users to switch between different simulation views, helping users to view, understand and use the various functional modules of the system and their status in order to view the different functions or status of the system.

[0152] 2. Web page construction: used to create and manage the system's web interface, including layout design, content display, etc., to provide users with an intuitive and easy-to-operate interface, ensuring that users can easily access and use system functions.

[0153] 3. Configuration panel: It allows users to adjust system parameters and settings. This function helps users configure the system's working mode and parameters according to actual needs to adapt to different application scenarios and operating requirements.

[0154] 4. Input source selection: allows users to select and specify data input sources, such as selecting different image acquisition devices or data streams. This function supports the management and switching of multiple input sources, ensuring that the system can obtain information from different data sources and perform corresponding processing and analysis.

[0155] 5. Historical or real-time data feedback and export: Users can compare historical and real-time output results and images through the user interaction module, find error information in time, and provide feedback to machine reinforcement learning through the user interaction module. The log management module data can also be exported according to user needs.

[0156] Embodiment 2:

[0157] This embodiment provides a spare parts management method based on target recognition and change detection, comprising the following steps:

[0158] Step S1: Collect images or videos of spare parts entering and leaving the warehouse through a camera installed in the warehouse, and transmit the image information to an image processing component through a communication interface;

[0159] Step S2: preprocessing the input image or video, including denoising, normalization and enhancement operations;

[0160] Step S3: According to the input pre-processed image or video, the in-and-out information of spare parts is confirmed through the image processing algorithm and the reinforcement learning model;

[0161] Step S4: Processing the outputted storage and in-and-out information of spare parts into a format recognizable by the log management module, and outputting it to the log management module;

[0162] Step S5: The log management module records detailed logs of system operations, including user operations, system events, and processing results; provides operation records and problem tracking functions to help system maintenance and troubleshooting;

[0163] Step S6: Through the user interaction module, the spare parts in and out history and real-time images and result information can be queried; different image devices can be selected; the current status can be displayed, user input can be accepted, and system parameters can be configured.

[0164] Embodiment three:

[0165] This embodiment provides an electronic device having a computer program stored thereon. When the computer program is executed by a processor, a spare parts management method based on target recognition and change detection as described in any embodiment of the present invention is implemented.

[0166] Embodiment 4:

[0167] This embodiment provides a computer-readable medium for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement a spare parts management method based on target recognition and change detection as described in any embodiment of the present invention.

[0168] In the embodiments of the present application, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can be represented by: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, c can be single or multiple.

[0169] Those of ordinary skill in the art will appreciate that the various units and algorithm steps described in the embodiments disclosed herein can be implemented in a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0170] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0171] In several embodiments provided in the present application, if any function 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, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for 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, mobile hard disk, read-only memory (Read-Only Memory; hereinafter referred to as: ROM), random access memory (Random Access Memory; hereinafter referred to as: RAM), disk or optical disk, and other media that can store program codes.

[0172] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A spare parts management system based on target recognition and change detection, characterized in that: include: The spare parts detection module includes an image acquisition unit, an image preprocessing unit, a target detection unit, and a post-processing unit, wherein: Image acquisition unit: collects images or videos of spare parts entering and leaving the warehouse through the image acquisition equipment installed in the warehouse, and transmits the image or video information to the image preprocessing unit through the communication interface; Image preprocessing unit: preprocess the input image or video information, including denoising, normalization and enhancement operations; Target detection unit: Based on the input pre-processed image or video information, the image processing algorithm and reinforcement learning model are used to confirm the in and out information of spare parts; Post-processing unit: processes the outputted storage and in-and-out information of spare parts into a format that can be recognized by the log management module, and outputs it to the log management module; Log management module: records detailed logs of system operations, including user operations, system events, and processing results; User interaction module: used to query the history of spare parts in and out of the warehouse and real-time images and result information; to select different image acquisition devices; to display the current system status, receive user input, and configure system parameters.

2. A spare parts management system based on target recognition and change detection according to claim 1, characterized in that: The image preprocessing unit preprocesses the input image or video information, specifically including the following steps: The specific process of denoising is as follows: Mean filtering: For each pixel in the image, its new pixel value is the average value of its neighborhood pixel values; if the neighborhood size is m×n, then the new value of the pixel (x, y) I new (x,y) is: Where I(i,j) represents the original pixel value at pixel point (i,j); Median filtering: sort the pixel values ​​in the neighborhood and take the middle value as the new pixel value; The specific formula of the normalization process is as follows: Where I(x,y) is the original pixel value, I max and I min are the minimum and maximum pixel values ​​in the image, respectively; The specific formula of the enhancement process is as follows: Brightness adjustment: I brightened (x,y)=I(x,y)+b; where b is the brightness adjustment value; Contrast adjustment: I contrasted (x,y)=a×I(x,y)+b; where a controls contrast and b controls brightness.

3. The spare parts management system based on target recognition and change detection according to claim 1 is characterized in that: The target detection unit confirms the in-and-out information of spare parts through image processing algorithms and reinforcement learning models, including the following steps: Network forward propagation: Process the preprocessed image through a multi-layer convolutional neural network to extract multi-layer features; use the Backbone deep neural network to extract the basic features of the image, and use the pyramid pooling module PPM to capture multi-scale information, integrate the feature pyramid network FPN to fuse feature maps at different levels, and combine the adaptive spatial feature network ASFF to improve the feature fusion effect; finally, use the multi-scale detection head Multi-scale Detection Head to perform target detection at multiple scales to adapt to targets of different sizes; during the training process, optimize and improve the Mosaic data enhancement strategy, enhance the generalization ability of the model, and dynamically adjust the learning rate to improve training efficiency and model performance; Prediction output: The output of the network is a fixed-size grid, where each grid cell predicts the bounding box, confidence score, and class probability of the object; Decoding output: Decode the prediction results output by the network into the actual target bounding box and category; convert the relative coordinates into the actual coordinates of the image, filter out low-confidence bounding boxes according to the confidence threshold, remove repeated overlapping bounding boxes, retain the highest confidence bounding box, and avoid multiple detections of the same target; Post-processing: Perform subsequent processing on the final target detection results; draw the detected target bounding box and category label on the original image, and output the detection results to the log management module for storage.

4. The spare parts management system based on target recognition and change detection according to claim 2 is characterized in that: The network forward propagation process specifically includes the following steps: The specific formula for the convolution operation is as follows: Where: I in and I out are the input and output images respectively, is the convolution kernel, k depends on the size of the convolution kernel; The specific formula of the pyramid pooling module PPM is as follows: Where: F pooled is the feature map obtained under different pooling scales d, s is the pooling window size; The feature map F after feature pyramid network FPN fusion is expressed as: F=H+L Where: H is the high-level feature map, L is the upsampled low-level feature map; The specific formula of adaptive spatial feature network ASFF is as follows: F ASFF =w1F1+w2F2+,…,+w c F c In the formula, F c is the cth fused feature map, w c corresponds to the adaptive weights obtained through network learning; The specific formula of Multi-scale Detection Heads is as follows: x=x a +Δx,y=y a +Δy,z=z a +Δz,h=h a +Δh Where: x a ,y a 、z a 、h a are the original anchor box coordinates, Δx, Δy, Δz, Δh are the predicted bounding box offsets.

5. The spare parts management system based on target recognition and change detection according to claim 2 is characterized in that: The prediction output process specifically includes the following steps: Coordinate transformation: x=x grid ×stride+x offset y=y grid ×stride+y offset Where: x grid ,y grid is the index of the grid cell; stride is the step size of the cell; x offset ,y offset is the predicted offset; The confidence Softmax calculation formula is as follows: Where: p o is the probability category; S o is the confidence score of the o-th category; o = 1, 2, 3, …, b.

6. The spare parts management system based on target recognition and change detection according to claim 2 is characterized in that: In the decoding output process, low-confidence bounding boxes are filtered out according to the confidence threshold; this is achieved through a non-maximum suppression algorithm, and the specific formula is as follows: Where: S o is the confidence score; M is the bounding box with the highest score; b v is the bounding box to be processed; IoU(M,b v ) is the bounding box M and b v The intersection-and-union ratio of t is the set intersection-over-union ratio threshold; Then we have: Where: area(M∩b v ) is the intersection area of ​​the two bounding boxes; area(M∪b v ) is the area of ​​the union of the two bounding boxes.

7. The spare parts management system based on target recognition and change detection according to claim 1 is characterized in that: The user interaction module can query the spare parts in and out history and real-time images and result information. By manually comparing the output results with the images, error information can be discovered in time, and the user interaction module can be used to feed back machine reinforcement learning.

8. A spare parts management method based on target recognition and change detection, characterized in that: The following steps are involved: Step S1: Collect images or videos of spare parts entering and leaving the warehouse through a camera installed in the warehouse, and transmit the image information to an image processing component through a communication interface; Step S2: preprocessing the input image or video, including denoising, normalization and enhancement operations; Step S3: According to the input pre-processed image or video, the in-and-out information of spare parts is confirmed through the image processing algorithm and the reinforcement learning model; Step S4: Processing the outputted storage and in-and-out information of spare parts into a format recognizable by the log management module, and outputting it to the log management module; Step S5: The log management module records detailed logs of system operations, including user operations, system events, and processing results; provides operation records and problem tracking functions to help system maintenance and troubleshooting; Step S6: Through the user interaction module, the spare parts in and out history and real-time images and result information can be queried; different image devices can be selected; the current status can be displayed, user input can be accepted, and system parameters can be configured.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the spare parts management method based on target recognition and change detection as described in claim 8 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, the method for managing spare parts based on target recognition and change detection as described in any one of claim 8 is implemented.