Intelligent detection method and system for property right certificate table based on edge enhancement and large model

Abstract

The application discloses a kind of based on edge enhancement and big model's property right certificate table form intelligent detection method and system, it is related to image recognition technical field, the based on edge enhancement and big model's property right certificate table form intelligent detection method, comprising: obtaining the property right certificate table form image to be detected;The property right certificate table form image to be detected is input to pre-training based on edge enhancement and big model's property right certificate table form intelligent detection model;The feature extraction unit carries out edge guide feature extraction to the property right certificate table form image to be detected, obtains table form image feature;The target detection unit carries out target detection based on the table form image feature, obtains the table cell in the property right certificate table form image to be detected.By edge guide feature extraction, the stability and diversity of feature expression are effectively enhanced;By embedding attention unit, the understanding ability of model to table cell structure information is effectively improved.

Description

Technical Field

[0001] This application relates to the field of image recognition technology, and in particular to a method and system for intelligent detection of property certificate forms based on edge enhancement and large models. Background Technology

[0002] Currently, table detection, as an important task in intelligent document analysis, has been widely studied in scenarios such as invoice recognition, contract parsing, and financial statement digitization. However, in the recognition of property ownership certificates in the real estate sector, existing technologies still have significant shortcomings.

[0003] Prior to 2016, house ownership certificates and land use right certificates recorded ownership information for houses and land respectively, both in structured table format. After 2016, the two certificates were merged into a single real estate ownership certificate, which has a more complex table structure, requiring the recording of key information such as location, real estate unit number, right holder, area, use, and term of use within the same document. These types of certificate forms are often subject to interference factors such as anti-counterfeiting backgrounds, red seals, signature overlays, and scan fading, leading to a significant decrease in the accuracy of existing form detection methods in practical applications. Summary of the Invention

[0004] The main purpose of this application is to provide an intelligent detection method and system for property certificate forms based on edge enhancement and large models, which aims to solve the technical problem of low detection accuracy of existing property certificate forms.

[0005] To achieve the above objectives, this application proposes an intelligent detection method for property ownership certificate forms based on edge enhancement and a large model. The intelligent detection method for property ownership certificate forms based on edge enhancement and a large model includes:

[0006] Obtain the image of the property ownership certificate form to be inspected;

[0007] The image of the property certificate form to be detected is input into a pre-trained intelligent property certificate form detection model based on edge enhancement and large model to detect the form and obtain the target form unit.

[0008] The intelligent detection model for property certificate forms based on edge enhancement and large models includes: a feature extraction unit, an object detection unit, and an attention unit;

[0009] The step of inputting the image of the property certificate form to be detected into a pre-trained intelligent property certificate form detection model based on edge enhancement and large model to perform form detection and obtain the target form unit includes:

[0010] The image of the property certificate form to be detected is input into the feature extraction unit for edge-guided feature extraction to obtain the features of the form image.

[0011] The target detection unit performs target detection based on the features of the table image to obtain the target table unit in the property certificate table image to be detected.

[0012] The attention unit is embedded within the target detection unit to enhance the features of the table image.

[0013] In one embodiment, the feature extraction unit is composed of an EIEStem subunit and a plurality of stacked EIEM subunits;

[0014] The step of inputting the image of the property certificate form to be detected into the feature extraction unit for edge-guided feature extraction to obtain the features of the form image includes:

[0015] The image of the property certificate form to be detected is input into the feature extraction unit;

[0016] The EIEStem subunit extracts features from the image of the property certificate form to be detected through edge branches to obtain edge detection features;

[0017] The EIEStem subunit extracts features from the image of the property certificate form to be detected through spatial branching to obtain spatial detection features;

[0018] The EIEStem subunit fuses the edge detection features and the spatial detection features to obtain a first fused feature; the first fused feature can be used as the input feature of the EIEM subunit.

[0019] The first fused feature is structurally enhanced by several stacked EIEM sub-units to obtain table image features.

[0020] In one embodiment, the step of performing structural feature enhancement on the first fused feature using a plurality of stacked EIEM sub-units to obtain table image features includes:

[0021] The EIEM subunit enhances the input features through an edge information enhancement branch to obtain edge-enhanced features;

[0022] The EIEM subunit performs feature learning on the input features through a spatial feature learning branch to obtain spatial context features;

[0023] The EIEM subunit fuses the edge enhancement features and the spatial context features to obtain a second fused feature; the second fused feature can be used as the input feature of the next EIEM subunit, until the second fused feature output by the last EIEM subunit is output as a tabular image feature.

[0024] In one embodiment, the step of the target detection unit performing target detection based on the table image features to obtain the target table cell in the property certificate table image to be detected includes:

[0025] The target detection unit generates a set of candidate boxes for target detection based on the features of the table image;

[0026] The candidate box with the highest confidence score in the candidate box set is selected as the target candidate box, and the intersection-union ratio (IUU) between the target candidate box and other candidate boxes is calculated.

[0027] Based on the intersection-union comparison, other candidate boxes are filtered to obtain the filtered detection boxes;

[0028] The table image features are detected by the detection box to obtain the table cells in the property certificate table image to be detected.

[0029] In one embodiment, during the pre-training of the intelligent detection model for property certificate forms based on edge enhancement and large models, the method further includes:

[0030] Obtain the original table image sample set;

[0031] The original table image sample set is enhanced by the feature enhancement unit to obtain the enhanced image sample set.

[0032] The enhanced image sample set is input into the initial intelligent detection model of property certificate form based on edge enhancement and large model for training, so as to obtain the pre-trained intelligent detection model of property certificate form based on edge enhancement and large model.

[0033] In one embodiment, the feature enhancement unit includes a first generator and a second generator;

[0034] The step of performing feature enhancement on the original table image sample set based on the feature enhancement unit to obtain an enhanced image sample set includes:

[0035] The first generator maps the original image samples in the original table image sample set to a complex style domain to obtain a pseudo target image;

[0036] The second generator reverse maps the pseudo-target image back to the original domain to obtain a reconstructed enhanced image sample set.

[0037] Furthermore, to achieve the above objectives, this application also proposes an intelligent detection system for property ownership certificate forms based on edge enhancement and a large model. The intelligent detection system for property ownership certificate forms based on edge enhancement and a large model includes:

[0038] The image acquisition module is used to acquire the image of the property certificate form to be inspected;

[0039] The target detection module is used to input the image of the property certificate form to be detected into a pre-trained intelligent property certificate form detection model based on edge enhancement and large model to detect the form and obtain the target form unit.

[0040] The intelligent detection model for property certificate forms based on edge enhancement and large models includes: a feature extraction unit, an object detection unit, and an attention unit;

[0041] The feature extraction unit is used to input the property certificate table image to be detected into the feature extraction unit for edge-guided feature extraction to obtain table image features;

[0042] The target detection unit is used to perform target detection based on the features of the table image to obtain the target table unit in the property certificate table image to be detected.

[0043] The attention unit is embedded within the target detection unit to enhance the features of the table image.

[0044] Furthermore, to achieve the above objectives, this application also proposes an intelligent detection device for property certificate forms based on edge enhancement and large models. The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. The computer program is configured to implement the steps of the intelligent detection method for property certificate forms based on edge enhancement and large models as described above.

[0045] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the intelligent detection method for property certificate forms based on edge enhancement and large models as described above.

[0046] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the intelligent detection method for property certificate forms based on edge enhancement and large models as described above.

[0047] One or more technical solutions proposed in this application have at least the following technical effects:

[0048] This application obtains an image of a property ownership certificate form to be detected; inputs the image into a pre-trained intelligent property ownership certificate form detection model based on edge enhancement and a large model; the feature extraction unit performs edge-guided feature extraction on the image to be detected to obtain table image features; the target detection unit performs target detection based on the table image features to obtain table cells in the image to be detected. Edge-guided feature extraction effectively enhances the stability and diversity of feature representation; embedding an attention unit effectively improves the model's ability to understand the structural information of table cells. Attached Figure Description

[0049] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0050] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is a flowchart illustrating an embodiment of the intelligent detection method for property certificate forms based on edge enhancement and large models in this application.

[0052] Figure 2 This is a schematic diagram of the multi-head self-attention mechanism in one implementation of the intelligent detection method for property certificate forms based on edge enhancement and large models in this application.

[0053] Figure 3 This is a schematic diagram of the feature extraction module structure in one implementation of the intelligent detection method for property certificate forms based on edge enhancement and large models in this application;

[0054] Figure 4 This is a schematic diagram of the feature enhancement module provided in one implementation of the intelligent detection method for property certificate forms based on edge enhancement and large models in this application;

[0055] Figure 5 This is a schematic diagram of the multi-module collaborative optimization framework of the intelligent detection method for property certificate forms based on edge enhancement and large model in this application, which can be used for intelligent detection of property certificate forms based on edge enhancement and large model in complex scenarios.

[0056] Figure 6 This is a schematic diagram of the module structure of the intelligent detection system for property certificate forms based on edge enhancement and large model according to an embodiment of this application;

[0057] Figure 7 This is a schematic diagram of the hardware operating environment of the intelligent detection method for property certificate forms based on edge enhancement and large models in the embodiments of this application.

[0058] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0059] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0060] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0061] The main solution of this application embodiment is: to acquire an image of a property certificate form to be detected; to input the image of the property certificate form to be detected into a pre-trained intelligent detection model of a property certificate form based on edge enhancement and a large model, the intelligent detection model of a property certificate form based on edge enhancement and a large model includes: a feature extraction unit, an object detection unit, and an attention unit; the feature extraction unit performs edge-guided feature extraction on the image of the property certificate form to be detected to obtain table image features; the object detection unit performs object detection based on the table image features to obtain table units in the image of the property certificate form to be detected.

[0062] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a computer or server, or an electronic device or virtual system capable of performing the above functions. The following description uses an intelligent detection device for property certificate forms based on edge enhancement and a large model (hereinafter referred to as the detection device) as an example to illustrate this embodiment and the following embodiments.

[0063] Based on this, embodiments of this application provide an intelligent detection method for property certificate forms based on edge enhancement and large models, referring to... Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the intelligent detection method for property certificate forms based on edge enhancement and large models in this application.

[0064] In this embodiment, the intelligent detection method for property certificate forms based on edge enhancement and large models includes steps S10 to S40:

[0065] Step S10: Obtain the image of the property certificate form to be inspected;

[0066] Step S20: Input the image of the property certificate form to be detected into the pre-trained intelligent detection model of property certificate form based on edge enhancement and large model to detect the form and obtain the target form unit.

[0067] Understandably, in the intelligent detection of property certificate forms based on edge enhancement and large models, the property certificate form image to be detected is the input property certificate image that needs to be recognized for table structure.

[0068] It should be noted that the aforementioned intelligent detection model for property certificate forms based on edge enhancement and a large model can be a pre-trained model used to perform intelligent detection of property certificate forms based on edge enhancement and a large model on the image to be detected. The intelligent detection model for property certificate forms based on edge enhancement and a large model in this application embodiment can be built based on the YOLO series models, or it can be built in other ways. This application embodiment does not limit this.

[0069] In some embodiments of this application, the intelligent detection model for property certificate forms based on edge enhancement and large models may include a feature extraction unit, an object detection unit, and an attention unit.

[0070] It should be explained that the above-mentioned feature extraction unit can extract table image features from the input property certificate table image to be detected. The feature extraction unit of this application embodiment can be based on convolutional neural network, Transformer encoder, etc., and this application embodiment does not limit it.

[0071] It should be noted that the attention unit mentioned above can be a functional unit used to enhance the features of a table image. By filtering and focusing information on the features of the table image, the key features are enhanced and long-distance dependencies are modeled. The attention unit in this application embodiment can be implemented based on self-attention, spatial attention, channel attention, etc., and this application embodiment does not limit it.

[0072] It should be explained that the aforementioned target detection unit can be a functional unit that can be used to locate and classify targets based on input features.

[0073] It should be noted that the target table unit mentioned above is the table unit detected from the image of the property certificate table to be detected.

[0074] In some embodiments of this application, step S100 involves inputting the image of the property certificate form to be detected into the feature extraction unit for edge-guided feature extraction to obtain the form image features;

[0075] In step S200, the target detection unit performs target detection based on the table image features to obtain the target table units in the property certificate table image to be detected. The attention unit is embedded in the target detection unit to enhance the table image features.

[0076] It should be noted that the feature extraction unit in this embodiment can be an extraction unit with image edge / structure information extraction function. By capturing structurally sensitive areas such as grid lines and intersections, it can effectively improve the model's perception accuracy of boundaries, broken lines and fine structure areas.

[0077] In some embodiments of this application, the target detection unit can be a detection unit built based on YOLOv11. In this application, an attention unit can be introduced before each scale detection head of YOLOv11. This attention unit can be a functional unit based on Multi-Head Self-Attention (MHSA), which guides the target detection unit to automatically capture structural constraints between tables, such as alignment, repetition, and spacing, by constructing long spatial dependencies and global interaction modeling capabilities. This solves interference problems such as tight table structures, blurred boundaries, and uneven scales.

[0078] Understandably, multi-head self-attention is a core component of the Transformer encoder. By capturing information from different subspaces within a sequence through multiple parallel attention pointers, it enhances the model's ability to model contextual dependencies. Multi-head self-attention enhances the model's ability to handle complex relationships by learning multiple sets of different attention representations in parallel. In this embodiment, the multi-head self-attention mechanism significantly strengthens the model's ability to perceive the overall layout and semantic consistency of the entire table during recognition, exhibiting higher stability and accuracy, especially in scenarios with distorted table structures and irregular cell shapes.

[0079] In some embodiments of this application, the multi-head self-attention mechanism of this application can be as follows: Figure 2 As shown, Figure 2 This is a schematic diagram of the multi-head self-attention mechanism in one implementation of the intelligent detection method for property certificate forms based on edge enhancement and large models in this application.

[0080] like Figure 2 As shown, the input feature map x (with dimensions H×W×d) can be processed by three 1×1 convolution operations (W... Q W K and W V ), generating the query matrix q, the key matrix k, and the value matrix v, which can be specifically represented as:

[0081] ;

[0082] Where Wq is the convolution operation W Q The convolution kernel parameter matrix, Wk is the convolution operation W KThe convolution kernel parameter matrix, Wv is the convolution operation W V The convolution kernel parameter matrix.

[0083] It should be noted that, in order to enhance the spatial awareness of the model, this application embodiment also introduces a lateral position encoding R. h (Dimensions are H×1×d) and vertical position code R w (Size is 1×W×d). By adding the horizontal and vertical position codes, the position offset tensor r can be constructed, which can be specifically represented as:

[0084] ;

[0085] Furthermore, by adding the query matrix and the position bias, and then performing a dot product operation with the key matrix, the attention weights can be calculated, which can be specifically expressed as:

[0086]

[0087] Where T represents transpose and softmax is the normalized exponential function.

[0088] It should be noted that the output feature z can be obtained by multiplying the attention weights by the value matrix, which can be specifically expressed as:

[0089] .

[0090] It should be noted that by embedding attention units in front of the three detectors of the YOLOv11 model, the model's ability to understand the structural information of table cells can be effectively improved. In scenarios with complex background interference, weak boundary overlap, and multi-scale table targets, the proposed solution can enhance the perceptual representation ability of the detectors, improve recall and localization accuracy, and demonstrates high robustness and practical application value.

[0091] In some embodiments of this application, when performing object detection, the traditional Non-Maximum Suppression (NMS) algorithm in YOLOv11 can be used for candidate box selection. NMS is mainly used to remove redundant detection boxes in object detection tasks. Its basic principle is to sort all candidate boxes according to their confidence scores, retain the candidate box A with the highest score, and calculate the Intersection over Union (IoU) with each of the other candidate boxes Bi. If the IoU exceeds a set threshold Nt, the confidence score Si corresponding to Bi is directly set to 0, and the box is considered redundant and discarded.

[0092] This traditional strategy can be expressed by the following function:

[0093] ;

[0094] Where iou(A,B) i ) is used to represent target candidate box A and candidate box B. i The intersection-union ratio between them.

[0095] It should be noted that this "hard suppression" mechanism has obvious drawbacks in table detection scenarios, especially when the cells are densely arranged, the boundaries overlap severely, or the image has background interference. It can easily lead to the false deletion of detection boxes that actually contain the target, thereby reducing the model's recall rate and structure restoration effect.

[0096] In some embodiments of this application, a candidate box selection optimization method based on Soft-NMS (Soft Non-Maximum Suppression) is introduced into the inference stage of YOLOv11 to replace the traditional non-maximum suppression algorithm in YOLOv11. This addresses the problem of falsely deleting target boxes in complex document images where table cells are dense, boundaries overlap, and structures are blurred. Specifically, the basic idea of ​​Soft-NMS is to apply a decay weight to the score of candidate boxes based on the degree of overlap between them and the highest-scoring box. This gradually reduces the score of highly overlapping candidate boxes but retains them, thereby reducing the false deletion of true targets. That is, the step of the target detection unit performing target detection based on the table image features to obtain table cells in the property certificate table image to be detected includes: the target detection unit generating a set of candidate boxes for target detection based on the table image features; selecting the candidate box with the highest confidence score in the candidate box set as the target candidate box, and calculating the intersection-union ratio (IUU) between the target candidate box and other candidate boxes; filtering other candidate boxes based on the IUU to obtain the filtered detection boxes; and performing target detection on the table image features through the detection boxes to obtain table cells in the property certificate table image to be detected.

[0097] It should be noted that during object detection, a set of candidate boxes can be generated. This set can include several candidate boxes, each with a corresponding confidence score used to evaluate the probability of a true object within the candidate box. The candidate box with the highest confidence score in the set is selected as the target candidate box, and the intersection-union ratio (IUR) between the target candidate box and other candidate boxes in the set is determined. This IUR can be used to characterize the degree of overlap between the target candidate box and other candidate boxes.

[0098] In this embodiment, candidate box filtering can be performed using a linear filtering method. Specifically, the linear decay form of Soft-NMS is as follows:

[0099]

[0100] In some embodiments of this application, considering the complexity of boundary blurring and target overlap issues in the table structure, a Gaussian decay function can also be used in this application to obtain a smoother and more continuous score adjustment mechanism, as follows:

[0101]

[0102] Where Si is the updated confidence score of the i-th candidate box, IoU(A, bi) represents the intersection-union ratio (IoU) between the current candidate box bi and the target candidate box A, σ is a parameter controlling the decay rate, and D represents the candidate box set. Compared with linear functions, the Gaussian decay function has a gentler decay trend in the high IoU range, avoiding abrupt score changes, and is especially suitable for dense detection scenarios with intersecting table lines or close boundaries.

[0103] Understandably, the above method allows for a smooth decay of candidate box confidence. For any candidate box, when its intersection-union ratio (IU) with the target candidate box exceeds a set threshold, its confidence is reduced using an exponential or linear function to update the confidence, rather than completely eliminating it. Ultimately, all candidate boxes can be sorted based on their updated confidence, and an appropriate number of candidate boxes are selected as the detection results (target table cells).

[0104] This application embodiment acquires an image of a property ownership certificate form to be detected; inputs the image into a pre-trained intelligent detection model for property ownership certificates based on edge enhancement and a large model; the feature extraction unit performs edge-guided feature extraction on the image to be detected to obtain table image features; the target detection unit performs target detection based on the table image features to obtain table cells in the image to be detected. Edge-guided feature extraction effectively enhances the stability and diversity of feature representation; embedding an attention unit effectively improves the model's ability to understand the structural information of table cells.

[0105] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in Embodiment 1 above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 3 , Figure 3 This is a schematic diagram of the feature extraction unit structure in one implementation of the intelligent detection method for property certificate forms based on edge enhancement and large models in this application.

[0106] like Figure 3 As shown in this embodiment, the feature extraction unit consists of an EIEStem subunit and several stacked EIEM subunits; the step of inputting the property certificate table image to be detected into the feature extraction unit for edge-guided feature extraction to obtain the table image features includes:

[0107] The image of the property certificate form to be detected is input into the feature extraction unit;

[0108] The EIEStem subunit extracts features from the image of the property certificate form to be detected through edge branches to obtain edge detection features;

[0109] The EIEStem subunit extracts features from the image of the property certificate form to be detected through spatial branching to obtain spatial detection features;

[0110] The EIEStem subunit fuses the edge detection features and the spatial detection features to obtain a first fused feature; the first fused feature can be used as the input feature of the EIEM subunit.

[0111] The first fused feature is structurally enhanced by several stacked EIEM sub-units to obtain table image features.

[0112] It should be noted that this application proposes a lightweight feature extraction unit that integrates edge structure awareness and spatial feature learning to replace the traditional convolutional structure at the front of the backbone in a table detection network. This achieves deep fusion of edge and spatial features, improving structural modeling and boundary awareness capabilities in complex document backgrounds, and is particularly suitable for image environments with blurred table lines and strong background interference. Specifically, this feature extraction unit consists of an EIEStem subunit and several stacked C3k2_EIEM subunits, forming a complete front-end edge-space co-modeling path.

[0113] It should be explained that the EIEStem subunit in this application embodiment integrates an edge extraction branch based on the Sobel edge operator, a spatial pooling branch with no parameter overhead, and a convolution branch, and has good structure awareness and information preservation capabilities.

[0114] Specifically, the image input to the EIEStem subunit (such as the image of the property certificate table to be detected) first undergoes a standard convolution operation (Conv) to reduce its dimensionality, and then enters two branches: the first is the edge branch (SobelConv), which improves the ability to perceive display-enhanced structures by combining the classic Sobel edge operators (Sobel-X operator and Sobel-Y operator) with trainable convolutions, and achieves fast extraction of horizontal and water quality-oriented edges through three-dimensional group convolutions; the second is the spatial branch (Maxpool), which performs downsampling through max pooling in the spatial branch, which can expand the receptive field and enhance the model's ability to perceive the global layout of the table while preserving significant structural features, and extract response information of key spatial locations without increasing the number of parameters.

[0115] It should be noted that the feature maps output by the edge branch and the spatial branch can be concatenated (Concat) along the channel dimension and then initially fused through subsequent convolution (Conv) to form the final first fused feature as the front-end feature representation.

[0116] It should be explained that, by introducing the EIEStem sub-unit, the embodiments of this application effectively compensate for the shortcomings of traditional downsampling convolution in modeling edge structures, and improve the response intensity of table boundary lines in complex backgrounds.

[0117] It should be noted that, in order to further enhance the expressive power of structural features, the embodiments of this application may also introduce a C3k2_EIEM module into the Backbone structure. The C3k2_EIEM module is based on the standard C3 structure (C3 means three convolution operations, k2 means two convolution kernels). The C3k2_EIEM module uses an improved EIEM sub-unit to replace the original Bottleneck module, realizing the deep fusion of edge features and spatial features.

[0118] In this embodiment, the EIEM subunit may include two parallel paths: one is an edge feature enhancement branch (Sobel Conv), used to explicitly extract edge information and obtain edge enhancement features; the other is a spatial feature learning branch (Conv), used to preserve spatial context features. By concatenating the features from the two parallel paths along the channel dimension (Concat), and then fusing them through convolution (Conv) and residual superimposing them with the original input features, the diversity and stability of feature representation can be effectively enhanced. Compared with traditional residual structures, the EIEM subunit is more sensitive to changes in local image structure and contours, and can better capture key information such as row and column dividing lines and cell boundaries in table detection. That is, the step of performing structural feature enhancement on the first fused feature through several stacked EIEM sub-units to obtain table image features includes: the EIEM sub-unit performing feature enhancement on the input feature through an edge information enhancement branch to obtain edge enhancement features; the EIEM sub-unit performing feature learning on the input feature through a spatial feature learning branch to obtain spatial context features; the EIEM sub-unit fusing the edge enhancement features and the spatial context features to obtain a second fused feature; the second fused feature can be used as the input feature of the next EIEM sub-unit, until the second fused feature output by the last EIEM sub-unit is output as the table image feature.

[0119] In some embodiments of this application, within the C3k2_EIEM module, when C3k is set to True, the module internally splits (Splits) and stacks two (i.e., N=2) C3k_EIEM structures. Each C3k_EIEM structure contains several cascaded EIEM sub-units, enabling deeper edge structure modeling capabilities. When C3k is False, a standard Bottleneck stacking structure is adopted to balance efficiency. This flexible structural configuration allows for an effective trade-off between performance and efficiency.

[0120] It should be noted that the feature extraction module proposed in this application addresses the problems of blurred table boundaries and severe structural interference in complex document images by constructing an efficient front-end structure that integrates Sobel edge operators and spatial convolution features. Through the self-designed EIEStem sub-unit and EIEM unit, the model's ability to perceive key areas such as table structure lines and cell outlines is significantly enhanced, improving the overall detection system's structural modeling effect in complex backgrounds. This module features novel structure, strong feature expression capabilities, and low parameter overhead, providing core support for constructing an intelligent detection system for property certificate tables based on edge enhancement and large models for complex backgrounds.

[0121] In this embodiment, the feature extraction unit consists of an EIEStem subunit and several stacked EIEM subunits. The EIEStem subunit extracts features from the property certificate table image to be detected through edge branches, obtaining edge detection features. The EIEStem subunit also extracts features from the property certificate table image to be detected through spatial branches, obtaining spatial detection features. The EIEStem subunit fuses the edge detection features and spatial detection features to obtain a first fused feature. The front-end features can be used as input features for the EIEM subunits. The first fused feature is then enhanced with structural features by several stacked EIEM subunits to obtain the table image features. Because the front-end design uses EIEStem, the input property certificate table image to be detected is extracted and initially fused through edge and spatial branches to form the front-end feature representation. This compensates for the shortcomings of traditional downsampling convolution in modeling edge structures and improves the response strength of table boundary lines in complex backgrounds. The introduction of the C3k2_EIEM module further enhances the expressive power of structural features, achieving an effective trade-off between performance and efficiency.

[0122] Based on the first and / or second embodiments of this application, in the third embodiment of this application, the content that is the same as or similar to the first and / or second embodiments described above can be referred to the above description and will not be repeated hereafter. Based on this, please refer to... Figure 4 , Figure 4 This is a schematic diagram of the feature enhancement module provided in one implementation of the intelligent detection method for property certificate forms based on edge enhancement and large models in this application.

[0123] like Figure 4 As shown, in order to pre-train the intelligent detection model of property certificate forms based on edge enhancement and large model, the method further includes the following steps during the pre-training process: obtaining an original set of table image samples; performing feature enhancement on the original set of table image samples based on feature enhancement units to obtain an enhanced image sample set; and inputting the enhanced image sample set into the initial intelligent detection model of property certificate forms based on edge enhancement and large model for training to obtain a pre-trained intelligent detection model of property certificate forms based on edge enhancement and large model.

[0124] It should be noted that this application also proposes a CycleGAN-based style transfer enhancement method for table images. This method is suitable for expanding data samples and enhancing model training in intelligent detection tasks of property certificate tables based on edge enhancement and large models in complex document images. It is particularly effective in addressing issues such as uneven lighting, background interference, and table structure distortion in naturally captured, historical, and scanned images, thereby improving model robustness and generalization ability. In this application, the CycleGAN unsupervised generative adversarial network is used to generate a large number of pseudo-realistic complex samples by performing cross-domain image transformation between standard-style and complex-style table images. This expands the distribution of training data and enhances the model's adaptability to various complex environments.

[0125] Specifically, the CycleGAN model in this application embodiment consists of two generators G(x→y) and F(y→x) and two discriminators D. x With D The system comprises: a first generator G that maps the original image x from the original domain X to the complex style domain Y in the original table image samples, generating a pseudo-target image G(x); and a second generator F that reverse-maps the pseudo-target image G(x) back to the original domain X, obtaining the reconstructed enhanced sample image F(G(x)). That is, the feature enhancement unit includes a first generator and a second generator; the step of performing feature enhancement on the original table image sample set based on the feature enhancement unit to obtain an enhanced image sample set includes: the first generator mapping the original image samples in the original table image sample set to the complex style domain to obtain the pseudo-target image; and the second generator reverse-mapping the pseudo-target image back to the original domain to obtain the reconstructed enhanced image sample set.

[0126] It should be noted that the generator processing described above constitutes a closed-loop transformation path of "standard style → complex style → standard style". To ensure that the image maintains structural consistency during the transformation process, a cycle consistency loss function is introduced, which is expressed as:

[0127] ;

[0128] Where E_x and E_y represent the desired operations on the original domain X and the complex style domain Y, respectively. This represents the L1 norm.

[0129] In some embodiments of this application, to enhance the realism of the generated image in the target style domain, two adversarial loss functions can be introduced into CycleGAN. The first adversarial loss is:

[0130] ;

[0131] The second type of combat loss is:

[0132] ;

[0133] Finally, CycleGAN's total loss function integrates the above three loss terms and is expressed as follows:

[0134] ;

[0135] Wherein, λ is a balancing factor used to adjust the ratio between cycle consistency loss and adversarial loss.

[0136] It should be noted that, in practical applications, the first generator G, after training, can convert standard-style table images into pseudo-complex images with features such as complex backgrounds, lighting perturbations, and structural distortions. This expands the number of training samples and improves the performance of the detection model in complex document scenarios. Experimental results show that this method can expand the training sample size to three times that of the original dataset, effectively alleviating the problems of sample scarcity and uneven distribution, and providing stable data support for the subsequent training of intelligent property certificate table detection models based on edge enhancement and large models.

[0137] In some embodiments of this application, when training the intelligent detection model for property certificate forms based on the above-mentioned enhanced image sample set, this application embodiment can also perform large-scale semantic reasoning based on the training detection results obtained by the intelligent detection model for property certificate forms based on the enhanced image sample set, thereby improving the semantic understanding ability of the lightweight detection model in complex property certificate scenarios.

[0138] Specifically, candidate boxes (i.e. target table cells in the training detection results) in the training detection results can be input into the multimodal large model, which will then perform global logical reasoning, including row and column alignment, cross-cell merging judgment, and field attribution determination, thereby generating table detection results that are more in line with semantic constraints.

[0139] In some embodiments of this application, the inference results of the above-mentioned multimodal large model can be converted into weak labels or pseudo labels, and the weak labels or pseudo labels can be combined with the original annotation results corresponding to the enhanced image sample set to construct new supervision signals.

[0140] In this embodiment, the training method and inference process of the above-mentioned multimodal large model are not limited, and can be selected based on the actual application.

[0141] In some embodiments of this application, the semantic reasoning capabilities of a multimodal large model can be transferred to a property certificate form intelligent detection model through a knowledge distillation mechanism. This enables the property certificate form intelligent detection model to possess a certain degree of global semantic understanding during reasoning. In the final practical deployment, the property certificate form intelligent detection model can independently complete the detection of table cells, boundary detection, and semantic reconstruction of the property certificate form, balancing detection efficiency and logical consistency.

[0142] In some embodiments of this application, such as Figure 5 As shown, Figure 5 This is a schematic diagram of the multi-module collaborative optimization framework of the intelligent detection method for property certificate forms based on edge enhancement and large models in this application, which can be used for intelligent detection of property certificate forms based on edge enhancement and large models in complex scenarios.

[0143] Reference Figure 5 The embodiments of this application may include a data augmentation module and a backbone network, a neck network, and a head for target detection.

[0144] It should be noted that the aforementioned backbone network can be composed of the feature extraction units described in this application. Specifically, it can include EIEStem sub-units and several stacked C3k2_EIEM sub-units. Spatial pyramid pooling (SPPF) is used to perform pooling operations on the feature maps at different scales and then concatenate the pooling results to obtain a fixed-length feature representation. C2PSA is used to enhance the feature representation and improve the multi-scale feature extraction capability.

[0145] It should be explained that the features obtained after processing by each C3k2_EIEM subunit and the features output by C2PSA can be input into the neck network for downsampling and concatenation. Furthermore, a multi-head self-attention unit is introduced and embedded before the three Detect modules of the detection head, thereby improving the accuracy of target detection.

[0146] This application's embodiments are based on the YOLOv11 backbone detection network, integrating image style enhancement, edge-guided feature extraction, multi-scale target preservation strategies, and attention-aware mechanisms to construct a high-precision intelligent detection method for property certificate forms based on edge enhancement and a large model, suitable for complex document backgrounds. First, a CycleGAN network is used to simulate various printing styles and image interference, generating training samples with cross-format and cross-style characteristics, improving the detection model's generalization ability in diverse document format scenarios. Then, an edge-guided feature extraction module (EIEStem) is proposed, utilizing image gradient and spatial response fusion to enhance the representation of boundary features such as table wireframes, improving structure perception capabilities without introducing additional parameters. Simultaneously, a Soft-NMS target preservation strategy is introduced, replacing the traditional direct elimination mechanism with fractional decay, effectively suppressing target loss in densely populated small-unit regions. Finally, a multi-head self-attention mechanism (MHSA) module is placed before the detection head to enhance the model's ability to model the long-range dependency structure of tables, achieving accurate extraction of implicit table structures in complex backgrounds. The overall system exhibits good robustness and practicality, and can be widely adapted to tasks such as document scanning, invoice recognition, and form archiving.

[0147] This application embodiment obtains an original table image sample set; performs feature enhancement on the original table image sample set based on a feature enhancement unit to obtain an enhanced image sample set; inputs the enhanced image sample set into an initial intelligent detection model for property certificate forms based on edge enhancement and a large model for training, resulting in a pre-trained intelligent detection model for property certificate forms based on edge enhancement and a large model. Feature enhancement of the original table image sample set is performed through joint loss optimization of adversarial loss and cycle consistency, achieving high-fidelity style transfer between the source and target domain images. This mechanism can simulate real-world interference such as different resolutions, printing textures, image fading, and artifact occlusion, providing rich training sample support for table detection.

[0148] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the intelligent detection method for property certificate forms based on edge enhancement and large models in this application. Any simple transformations based on this technical concept are within the protection scope of this application.

[0149] This application also provides an intelligent detection system for property certificate forms based on edge enhancement and large models. Please refer to [reference needed]. Figure 6 , Figure 6 This is a schematic diagram of the module structure of the intelligent property certificate form detection system based on edge enhancement and large model according to an embodiment of this application. The intelligent property certificate form detection system based on edge enhancement and large model includes:

[0150] Image acquisition module 10 is used to acquire the image of the property certificate form to be detected;

[0151] The target detection module 20 is used to input the image of the property certificate form to be detected into a pre-trained intelligent detection model of property certificate form based on edge enhancement and large model to detect the form and obtain the target form unit.

[0152] The intelligent detection model for property certificate forms based on edge enhancement and large models includes: a feature extraction unit, an object detection unit, and an attention unit;

[0153] The feature extraction unit is used to input the property certificate table image to be detected into the feature extraction unit for edge-guided feature extraction to obtain table image features;

[0154] The target detection unit is used to perform target detection based on the features of the table image to obtain the target table unit in the property certificate table image to be detected.

[0155] The attention unit is embedded within the target detection unit to enhance the features of the table image.

[0156] The intelligent property certificate form detection system based on edge enhancement and large model provided in this application adopts the intelligent property certificate form detection method based on edge enhancement and large model in the above embodiments, which can solve the technical problem of low detection accuracy of existing property certificate forms. Compared with the prior art, the beneficial effects of the intelligent property certificate form detection system based on edge enhancement and large model provided in this application are the same as the beneficial effects of the intelligent property certificate form detection method based on edge enhancement and large model provided in the above embodiments, and other technical features of the intelligent property certificate form detection system based on edge enhancement and large model are the same as the features disclosed in the methods of the above embodiments, and will not be repeated here.

[0157] This application provides an intelligent detection device for property ownership certificates based on edge enhancement and large models. The intelligent detection device for property ownership certificates based on edge enhancement and large models includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the intelligent detection method for property ownership certificates based on edge enhancement and large models in the first embodiment described above.

[0158] The following is for reference. Figure 7This document illustrates a structural schematic diagram of a smart property certificate form detection device based on edge enhancement and large model, suitable for implementing embodiments of this application. The smart property certificate form detection device based on edge enhancement and large model in this application embodiment may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 7 The intelligent detection device for property certificate forms based on edge enhancement and large models shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0159] like Figure 7 As shown, the intelligent property certificate form detection device based on edge enhancement and large model can include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 1002 or a program loaded from storage device 1003 into random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the intelligent property certificate form detection device based on edge enhancement and large model. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the edge-enhanced and large-model-based smart certificate form detection device to wirelessly or wiredly communicate with other devices to exchange data. Although the figure shows an edge-enhanced and large-model-based smart certificate form detection device with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0160] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0161] The intelligent property certificate form detection device based on edge enhancement and large model provided in this application adopts the intelligent property certificate form detection method based on edge enhancement and large model in the above embodiments, which can solve the technical problem of low detection accuracy of existing property certificate forms. Compared with the prior art, the beneficial effects of the intelligent property certificate form detection device based on edge enhancement and large model provided in this application are the same as the beneficial effects of the intelligent property certificate form detection method based on edge enhancement and large model provided in the above embodiments, and other technical features in the intelligent property certificate form detection device based on edge enhancement and large model are the same as the features disclosed in the previous embodiment method, and will not be repeated here.

[0162] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0163] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0164] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the intelligent detection method for property certificate forms based on edge enhancement and large models in the above embodiments.

[0165] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), or flash memory, optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0166] The aforementioned computer-readable storage medium may be included in the intelligent detection device for property certificate forms based on edge enhancement and large models; or it may exist independently and not assembled into the intelligent detection device for property certificate forms based on edge enhancement and large models.

[0167] The aforementioned computer-readable storage medium carries one or more programs that, when executed by the intelligent property certificate form detection device based on edge enhancement and large model, cause the intelligent property certificate form detection device based on edge enhancement and large model to:

[0168] Obtain the image of the property ownership certificate form to be inspected;

[0169] The image of the property certificate form to be detected is input into a pre-trained intelligent property certificate form detection model based on edge enhancement and large model to detect the form and obtain the target form unit.

[0170] The intelligent detection model for property certificate forms based on edge enhancement and large models includes: a feature extraction unit, an object detection unit, and an attention unit;

[0171] The step of inputting the image of the property certificate form to be detected into a pre-trained intelligent property certificate form detection model based on edge enhancement and large model to perform form detection and obtain the target form unit includes:

[0172] The image of the property certificate form to be detected is input into the feature extraction unit for edge-guided feature extraction to obtain the features of the form image.

[0173] The target detection unit performs target detection based on the features of the table image to obtain the target table unit in the property certificate table image to be detected.

[0174] The attention unit is embedded within the target detection unit to enhance the features of the table image.

[0175] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0176] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0177] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0178] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described intelligent detection method for property ownership certificates based on edge enhancement and a large model. This method can solve the technical problem of low detection accuracy in existing property ownership certificates. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the intelligent detection method for property ownership certificates based on edge enhancement and a large model provided in the above embodiments, and will not be repeated here.

[0179] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described intelligent detection method for property certificate forms based on edge enhancement and large models.

[0180] The computer program product provided in this application can solve the technical problem of low detection accuracy of existing property certificate forms. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the intelligent detection method for property certificate forms based on edge enhancement and large models provided in the above embodiments, and will not be repeated here.

[0181] The above description is only a part of the embodiments of this application and does not limit the scope of protection of this application. All equivalent structural transformations made under the technical concept of this application and using the content of this application specification and drawings, or direct / indirect applications in other related technical fields, are included in the scope of protection of this application.

Claims

1. A method for intelligent detection of property ownership certificate forms based on edge enhancement and large model, characterized in that, The method includes: Obtain the image of the property ownership certificate form to be inspected; The image of the property certificate form to be detected is input into a pre-trained intelligent property certificate form detection model based on edge enhancement and large model to detect the form and obtain the target form unit. The intelligent detection model for property certificate forms based on edge enhancement and large models includes: a feature extraction unit, an object detection unit, and an attention unit; The step of inputting the image of the property certificate form to be detected into a pre-trained intelligent property certificate form detection model based on edge enhancement and large model to perform form detection and obtain the target form unit includes: The image of the property certificate form to be detected is input into the feature extraction unit for edge-guided feature extraction to obtain the features of the form image. The target detection unit performs target detection based on the features of the table image to obtain the target table unit in the property certificate table image to be detected. The attention unit is embedded within the target detection unit to enhance the features of the table image; The feature extraction unit consists of an EIEStem subunit and several stacked EIEM subunits; The step of inputting the image of the property certificate form to be detected into the feature extraction unit for edge-guided feature extraction to obtain the features of the form image includes: The image of the property certificate form to be detected is input into the feature extraction unit; The EIEStem subunit extracts features from the image of the property certificate form to be detected through edge branches to obtain edge detection features; The EIEStem subunit extracts features from the image of the property certificate form to be detected through spatial branching to obtain spatial detection features; The EIEStem subunit fuses the edge detection features and the spatial detection features to obtain a first fused feature; the first fused feature can be used as the input feature of the EIEM subunit. The first fused feature is structurally enhanced by several stacked EIEM sub-units to obtain table image features; The step of enhancing the structural features of the first fused feature using several stacked EIEM sub-units to obtain the table image features includes: The EIEM subunit enhances the input features through an edge information enhancement branch to obtain edge-enhanced features; The EIEM subunit performs feature learning on the input features through a spatial feature learning branch to obtain spatial context features; The EIEM subunit fuses the edge enhancement features and the spatial context features to obtain a second fused feature; the second fused feature can be used as the input feature of the next EIEM subunit, until the second fused feature output by the last EIEM subunit is output as a tabular image feature.

2. The intelligent detection method for property certificate forms based on edge enhancement and large model as described in claim 1, characterized in that, The step of the target detection unit performing target detection based on the table image features to obtain the target table unit in the property certificate table image to be detected includes: The target detection unit generates a set of candidate boxes for target detection based on the features of the table image; The candidate box with the highest confidence score in the candidate box set is selected as the target candidate box, and the intersection-union ratio (IUU) between the target candidate box and other candidate boxes is calculated. Based on the intersection-union comparison, other candidate boxes are filtered to obtain the filtered detection boxes; The table image features are detected by the detection box to obtain the table cells in the property certificate table image to be detected.

3. The intelligent detection method for property certificate forms based on edge enhancement and large model as described in claim 1, characterized in that, When pre-training the intelligent detection model for property certificate forms based on edge enhancement and large models, the method further includes: Obtain the original table image sample set; The original table image sample set is enhanced by the feature enhancement unit to obtain the enhanced image sample set. The enhanced image sample set is input into the initial intelligent detection model of property certificate form based on edge enhancement and large model for training, so as to obtain the pre-trained intelligent detection model of property certificate form based on edge enhancement and large model.

4. The intelligent detection method for property certificate forms based on edge enhancement and large model as described in claim 3, characterized in that, The feature enhancement unit includes a first generator and a second generator; The step of performing feature enhancement on the original table image sample set based on the feature enhancement unit to obtain an enhanced image sample set includes: The first generator maps the original image samples in the original table image sample set to a complex style domain to obtain a pseudo target image; The second generator reverse maps the pseudo-target image back to the original domain to obtain a reconstructed enhanced image sample set.

5. A smart detection system for property ownership certificate forms based on edge enhancement and large models, characterized in that, The intelligent detection system for property certificate forms based on edge enhancement and large models includes: The image acquisition module is used to acquire the image of the property certificate form to be inspected; The target detection module is used to input the image of the property certificate form to be detected into a pre-trained intelligent property certificate form detection model based on edge enhancement and large model to detect the form and obtain the target form unit. The intelligent detection model for property certificate forms based on edge enhancement and large models includes: a feature extraction unit, an object detection unit, and an attention unit; The feature extraction unit is used to input the property certificate table image to be detected into the feature extraction unit for edge-guided feature extraction to obtain table image features; The target detection unit is used to perform target detection based on the features of the table image to obtain the target table unit in the property certificate table image to be detected. The attention unit is embedded within the target detection unit to enhance the features of the table image; The feature extraction unit consists of an EIEStem subunit and several stacked EIEM subunits; The feature extraction unit is also used for: The image of the property certificate form to be detected is input into the feature extraction unit; The EIEStem subunit extracts features from the image of the property certificate form to be detected through edge branches to obtain edge detection features; The EIEStem subunit extracts features from the image of the property certificate form to be detected through spatial branching to obtain spatial detection features; The EIEStem subunit fuses the edge detection features and the spatial detection features to obtain a first fused feature; the first fused feature can be used as the input feature of the EIEM subunit. The first fused feature is structurally enhanced by several stacked EIEM sub-units to obtain table image features; The step of enhancing the structural features of the first fused feature through several stacked EIEM sub-units to obtain table image features includes: The EIEM subunit enhances the input features through an edge information enhancement branch to obtain edge-enhanced features; The EIEM subunit performs feature learning on the input features through a spatial feature learning branch to obtain spatial context features; The EIEM subunit fuses the edge enhancement features and the spatial context features to obtain a second fused feature; the second fused feature can be used as the input feature of the next EIEM subunit, until the second fused feature output by the last EIEM subunit is output as a tabular image feature.

6. A smart detection device for property ownership certificate forms based on edge enhancement and large model, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the intelligent detection method for property certificate forms based on edge enhancement and large models as described in any one of claims 1 to 4.

7. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the intelligent detection method for property certificate forms based on edge enhancement and large models as described in any one of claims 1 to 4.

8. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the intelligent detection method for property certificate forms based on edge enhancement and large models as described in any one of claims 1 to 4.

Images