Architectural drawing surface compliance intelligent inspection method and system

Abstract

The invention relates to the technical field of drawing detection, and discloses an architectural drawing surface compliance intelligent inspection method and system, and the method comprises the steps: obtaining a to-be-detected architectural drawing surface image, inputting the to-be-detected architectural drawing surface image to a drawing surface feature extraction neural network, obtaining architectural drawing surface features, and obtaining a to-be-detected architectural drawing surface image; and inputting the architectural drawing surface features and the to-be-detected architectural drawing surface image into a drawing surface region extraction neural network to obtain a drawing surface region extraction result, and inputting the architectural drawing surface features and the drawing surface region extraction result into a drawing surface target classification determination neural network to obtain an architectural drawing surface detection result. According to the method, image segmentation is carried out through cooperation of the drawing feature extraction neural network, the drawing region extraction neural network and the drawing target classification determination neural network, and blank axis numbers, redundant calibration cloud lines and messy code font regions can be accurately segmented, so that whether the drawing meets specification requirements or not is determined.

Description

Technical Field

[0001] The present invention relates to the technical field of drawing detection, and in particular to a method and system for intelligently checking the compliance of architectural drawings. Background Art

[0002] Before the engineering drawings are produced, in order to ensure the quality and accuracy of the drawings, the technical quality-related departments need to conduct a large number of manual checks on the drawings before they are produced, including errors in the construction unit, project name, sub-item name, design number, drawing name, project code, version number, design stage, drawing number, drawing scale, drawing date, no joint signature, incorrect drawing frame scale, incorrect annotation scale, label column not displayed on the right, line type not divided into thick and thin, missing axis number, missing drawing name, extra review cloud lines, no barcode, the number of sheets does not match the catalog, missing external referenced elements, etc. Manually checking drawings is a time-consuming and tedious task, and is prone to inconsistent audits and missed audits. Automation can greatly improve efficiency and reduce waste of time and human resources. Automatic checking can also provide a higher level of consistency and compliance, ensuring that the use of symbols and marks is consistent and better meets industry and corporate standards.

[0003] Therefore, there is an urgent need for an intelligent drawing compliance inspection method to identify PDF drawings, thereby realizing automatic inspection, assisting reviewers or archivists in inspection, reducing the workload of manual inspection, and improving compliance. Summary of the invention

[0004] The purpose of the present invention is to overcome one or more of the above-mentioned existing technical problems and provide a method for intelligently checking the compliance of architectural drawings.

[0005] To achieve the above-mentioned purpose, the present invention provides an intelligent method for checking the compliance of architectural drawings, comprising:

[0006] Obtaining an image of a building drawing to be detected;

[0007] Inputting the architectural drawing surface image to be detected into a surface feature extraction neural network to obtain architectural drawing surface features, wherein the surface feature extraction neural network includes a first calculation module, a second calculation module, a first enhanced calculation module, a second enhanced calculation module, a third enhanced calculation module and a result integration module;

[0008] Inputting the architectural drawing surface features and the architectural drawing surface images to be detected into a drawing area extraction neural network to obtain a drawing area extraction result, wherein the drawing area extraction neural network includes a drawing area feature integration module, a drawing area basic screening module, a drawing area preliminary prediction module, a drawing area regression prediction module and a drawing target area screening module;

[0009] The architectural drawing surface features and the drawing area extraction results are input into a drawing target classification determination neural network to obtain an architectural drawing surface detection result. The drawing target classification determination neural network includes a drawing target classification module and a drawing target position refinement module.

[0010] According to one aspect of the present invention, the architectural drawing image to be detected is input into the first calculation module to obtain a first calculation result, wherein the formula is:

[0011] R 11 =M1(R1);

[0012] R 11 =MaxPool(ReLU(Conv2d(Conv2d(R1))));

[0013] Among them, R 11 represents the first calculation result;

[0014] M1 represents the first computing module;

[0015] R1 represents the architectural drawing image to be detected;

[0016] MaxPool indicates the maximum pooling layer with a pooling kernel of 2×2;

[0017] ReLU stands for rectified linear unit activation function;

[0018] Conv2d represents a 3×3 convolutional layer;

[0019] The first calculation result is input into the second calculation module to obtain a second calculation result, wherein the formula is:

[0020] R 12 =M2(R 11 );

[0021] R 12 =MaxPool(ReLU(Conv2d(Conv2d(R 11 ))));

[0022] Among them, R 12 represents the second calculation result;

[0023] M2 represents the second computing module;

[0024] The architectural drawing image to be detected and the second calculation result are input into the first enhanced calculation module to obtain a first enhanced calculation result, wherein the formula is:

[0025] R 13 =T1(R 12, R1);

[0026] R 13 =MaxPool(ReLU(Conv2d(Conv2d(Conv2d(R 12 , R1)))));

[0027] Among them, R 13 represents the first enhanced calculation result;

[0028] T1 represents the first enhanced computing module;

[0029] The first enhanced calculation result is continuously input into the second enhanced calculation module and the third enhanced calculation module to obtain the second enhanced calculation result and the third enhanced calculation result, wherein the formula is,

[0030] R 14 =T2(R 13 );

[0031] R 15 =T3(R 14 );

[0032] R 14 =MaxPool(ReLU(Conv2d(Conv2d(Conv2d(R 13 )))));

[0033] R 15 =MaxPool(ReLU(Conv2d(Conv2d(Conv2d(R 14 )))));

[0034] Among them, R 14 represents the second enhanced calculation result;

[0035] R 15 represents the third enhanced calculation result;

[0036] T2 represents the second enhanced computing module;

[0037] T3 represents the third enhanced computing module.

[0038] According to one aspect of the present invention, the first enhanced calculation result and the third enhanced calculation result are input into the result integration module to obtain the architectural drawing surface feature, wherein the formula is,

[0039] X1=F(R 13 , R 15 );

[0040] X1=AvgPool(FC2(FC1(Softmax(R 13 , R15 ))));

[0041] Among them, X1 represents the characteristics of the architectural drawing;

[0042] F represents the result integration module;

[0043] FC1 represents the first fully connected layer;

[0044] FC2 represents the second fully connected layer;

[0045] Softmax represents the output normalized exponential function;

[0046] AvgPool represents an average pooling layer with a pooling kernel of 2×2.

[0047] According to one aspect of the present invention, the architectural drawing surface features and the architectural drawing surface image to be detected are input into the surface area feature integration module to obtain the surface area feature integration result, wherein the formula is,

[0048] R 21 =S1(R1,X1);

[0049] R 21 =Conv2d(R,X1);

[0050] Among them, R 21 Indicates the result of integrating features of the image region;

[0051] S1 represents the image region feature integration module.

[0052] According to one aspect of the present invention, the image region feature integration result is input into the image region basic screening module to obtain the image region basic screening result, wherein the formula is:

[0053] R 22 =S2(R 21 );

[0054] R 22 =Slide(R 21 );

[0055] Among them, R 22 Indicates the basic screening results of the drawing area;

[0056] Slide means sliding window mode;

[0057] S2 represents the basic screening module of the image area.

[0058] According to one aspect of the present invention, the basic screening result of the drawing area is input into the drawing area preliminary prediction module to obtain the drawing area preliminary prediction result, wherein the formula is:

[0059] R 23 =S3(R 22 );

[0060] R 23 =ReShape(Softmax(ReShape(Conv(Conv(Conv(R 22 ))))));

[0061] Among them, R 23 Indicates the preliminary prediction results of the map area;

[0062] ReShape means merging the last two dimensions;

[0063] Conv means convolution with a convolution kernel of 1×1;

[0064] S3 represents the preliminary prediction module of the image region.

[0065] According to one aspect of the present invention, the preliminary prediction result of the image region and the integrated result of the image region features are input into the image region regression prediction module to obtain the image region regression prediction result, wherein the formula is:

[0066] R 24 =S4(R 23 , R 21 );

[0067] S4=SmoothL1(Conv(Conv(Conv(R 23 , R 21 ))));

[0068] Among them, R 24 Indicates the regression prediction result of the image area;

[0069] S4 represents the image region regression prediction module;

[0070] SmoothL1 represents a smooth loss function.

[0071] According to one aspect of the present invention, the image region regression prediction result is input into the image target region screening module to obtain an image region extraction result, wherein the formula is:

[0072] X2=S5(R 24 );

[0073] X2=MaxPool(RoundUp(BilinearInterpolation(R 24 )));

[0074] Among them, X2 represents the surface area extraction result;

[0075] S5 represents the image target area screening module;

[0076] RoundUp means rounding up;

[0077] BilinearInterpolation represents bilinear interpolation.

[0078] According to one aspect of the present invention, the architectural drawing surface features and the surface area extraction results are input into the surface object classification module to obtain the surface object classification results, wherein the formula is:

[0079] R 31 =C1(X2,X1);

[0080] R 31 =Reg_head(Cls_head(MaxPool(Feature(X2,X1))));

[0081] Among them, R 31 Indicates the image target classification result;

[0082] Reg _ head represents the bounding box regression subnetwork;

[0083] Cls_head represents the classification subnetwork;

[0084] Feature represents the Conv and Pooling layers of the feature extraction network;

[0085] C1 represents the image object classification module;

[0086] The drawing object classification result and the drawing area extraction result are input into the drawing object position refinement module to obtain the building drawing drawing detection result, where the formula is:

[0087] X3=C2(R 31 , X2);

[0088] X3=Hermite(AvgPool(Scale(R 31 , X2)));

[0089] Among them, X3 represents the inspection result of architectural drawing;

[0090] C2 represents the image target location fine module;

[0091] Hermite represents interpolation calculation;

[0092] Scale represents the scaling ratio calculation of the rectangular area relative to the feature map.

[0093] To achieve the above object, the present invention provides an intelligent inspection system for compliance of architectural drawings, comprising:

[0094] Architectural drawing image acquisition module: acquires the architectural drawing image to be detected;

[0095] Architectural drawing surface feature acquisition module: input the architectural drawing surface image to be detected into the surface feature extraction neural network to obtain the architectural drawing surface features, and the surface feature extraction neural network includes a first calculation module, a second calculation module, a first enhanced calculation module, a second enhanced calculation module, a third enhanced calculation module and a result integration module;

[0096] Drawing area extraction result acquisition module: input the architectural drawing drawing features and the architectural drawing drawing image to be detected into the drawing area extraction neural network to obtain the drawing area extraction result, and the drawing area extraction neural network includes a drawing area feature integration module, a drawing area basic screening module, a drawing area preliminary prediction module, a drawing area regression prediction module and a drawing target area screening module;

[0097] Architectural drawing surface detection result acquisition module: input the architectural drawing surface features and the drawing area extraction results into the drawing target classification determination neural network to obtain the architectural drawing surface detection results. The drawing target classification determination neural network includes a drawing target classification module and a drawing target position refinement module.

[0098] Based on this, the beneficial effect of the present invention is that by adopting deep learning image segmentation technology to extract blank axis numbers, redundant review cloud lines, and garbled font areas in architectural drawings, and by identifying the special features of blank axis numbers, redundant review cloud lines, and garbled fonts, blank axis numbers, redundant review cloud lines, and garbled font areas can be accurately segmented, thereby determining whether the drawings meet the requirements of the specifications. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] Figure 1 A flowchart schematically showing a method for intelligently checking the compliance of architectural drawings according to the present invention;

[0100] Figure 2 A schematic diagram schematically shows a neural network diagram of an intelligent inspection system for compliance of architectural drawings according to the present invention;

[0101] Figure 3 The flowchart of a system for intelligently checking the compliance of architectural drawings according to the present invention is schematically shown. DETAILED DESCRIPTION

[0102] The present invention will now be discussed with reference to exemplary embodiments. It should be understood that the embodiments discussed are only for enabling those skilled in the art to better understand and thereby implement the present invention, rather than implying any limitation on the scope of the present invention.

[0103] As used herein, the term “including” and variations thereof are to be interpreted as open-ended terms meaning “including but not limited to.” The term “based on” is to be interpreted as “based, at least in part, on,” and the terms “one embodiment” and “an embodiment” are to be interpreted as “at least one embodiment.”

[0104] Figure 1 A flowchart schematically shows a method for intelligently checking the compliance of architectural drawings according to the present invention, as shown in FIG. Figure 1 As shown, a method for intelligently checking the compliance of architectural drawings of the present invention comprises:

[0105] Obtaining an image of a building drawing to be detected;

[0106] Inputting the architectural drawing surface image to be detected into the surface feature extraction neural network to obtain the architectural drawing surface features, the surface feature extraction neural network comprising a first calculation module, a second calculation module, a first enhanced calculation module, a second enhanced calculation module, a third enhanced calculation module and a result integration module;

[0107] Inputting the drawing features of the architectural drawings and the drawing images to be detected into the drawing area extraction neural network to obtain the drawing area extraction results, wherein the drawing area extraction neural network includes a drawing area feature integration module, a drawing area basic screening module, a drawing area preliminary prediction module, a drawing area regression prediction module and a drawing target area screening module;

[0108] The drawing features and drawing area extraction results of the architectural drawings are input into the drawing target classification determination neural network to obtain the architectural drawing detection results. The drawing target classification determination neural network includes a drawing target classification module and a drawing target position refinement module.

[0109] According to one embodiment of the present invention, Figure 2 A schematic diagram of a neural network of an intelligent inspection system for compliance of architectural drawings according to the present invention is shown schematically. Figure 2 As shown, the architectural drawing image to be detected is input into the first calculation module to obtain the first calculation result. The first calculation module consists of two 3×3 convolutional layers (Conv2d), one maximum pooling layer (MaxPool) with a pooling kernel of 2×2, and one ReLU (Rectified Linear unit, rectified linear unit activation function), where the formula is,

[0110] R11 =M1(R1);

[0111] R 11 =MaxPool(ReLU(Conv2d(Conv2d(R1))));

[0112] Among them, R 11 represents the first calculation result;

[0113] M1 represents the first computing module;

[0114] R1 represents the architectural drawing image to be detected;

[0115] MaxPool indicates the maximum pooling layer with a pooling kernel of 2×2;

[0116] ReLU stands for rectified linear unit activation function;

[0117] Conv2d represents a 3×3 convolutional layer;

[0118] The first calculation result is input into the second calculation module to obtain the second calculation result. The second calculation module consists of two 3×3 convolutional layers (Conv2d), one maximum pooling layer (MaxPool) with a pooling kernel of 2×2, and one ReLU (Rectified Linear unit, rectified linear unit activation function), where the formula is,

[0119] R 12 =M2(R 11 );

[0120] R 12 =MaxPool(ReLU(Conv2d(Conv2d(R 11 ))));

[0121] Among them, R 12 represents the second calculation result;

[0122] M2 represents the second computing module;

[0123] The architectural drawing image to be detected and the second calculation result are input into the first enhancement calculation module to obtain the first enhancement calculation result. The first enhancement calculation module consists of three 3×3 convolutional layers (Conv2d), one maximum pooling layer (MaxPool) with a pooling kernel of 2×2, and one rectified linear unit activation function (ReLU), where the formula is,

[0124] R 13 =T1(R 12 , R1);

[0125] R 13=MaxPool(ReLU(Conv2d(Conv2d(Conv2d(R 12 , R1)))));

[0126] Among them, R 13 represents the first enhanced calculation result;

[0127] T1 represents the first enhanced computing module;

[0128] The first enhancement calculation result is continuously input into the second enhancement calculation module and the third enhancement calculation module to obtain the second enhancement calculation result and the third enhancement calculation result. The second enhancement calculation module consists of three 3×3 convolutional layers (Conv2d), one maximum pooling layer (MaxPool) with a pooling kernel of 2×2, and one rectified linear unit activation function (ReLU). The third enhancement calculation module consists of three 3×3 convolutional layers (Conv2d), one maximum pooling layer (MaxPool) with a pooling kernel of 2×2, and one rectified linear unit activation function (ReLU). The formula is,

[0129] R 14 =T2(R 13 );

[0130] R 15 =T3(R 14 );

[0131] R 14 =MaxPool(ReLU(Conv2d(Conv2d(Conv2d(R 13 )))));

[0132] R 15 =MaxPool(ReLU(Conv2d(Conv2d(Conv2d(R 14 )))));

[0133] Among them, R 14 represents the second enhanced calculation result;

[0134] R 15 represents the third enhanced calculation result;

[0135] T2 represents the second enhanced computing module;

[0136] T3 represents the third enhanced computing module.

[0137] According to one embodiment of the present invention, the first enhancement calculation result and the third enhancement calculation result are input into the result integration module to obtain the architectural drawing surface features, which are composed of an average pooling layer, two fully connected layers and a final probability calculation layer Softma xComposition, where the formula is,

[0138] X1=F(R 13 , R 15 );

[0139] X1=AvgPool(FC2(FC1(Softmax(R 13 , R 15 ))));

[0140] Among them, X1 represents the characteristics of the architectural drawing;

[0141] F represents the result integration module;

[0142] FC1 represents the first fully connected layer;

[0143] FC2 represents the second fully connected layer;

[0144] Softmax represents the output normalized exponential function;

[0145] AvgPool represents an average pooling layer with a pooling kernel of 2×2.

[0146] According to one embodiment of the present invention, the architectural drawing surface features and the architectural drawing surface images to be detected are input into the surface area feature integration module to obtain the surface area feature integration result. The surface area feature integration module is composed of a 3×3 convolution layer. After the input feature map is convoluted, the shape remains unchanged, and the function is to integrate the features. The formula is:

[0147] R 21 =S1(R1,X1);

[0148] R 21 =Conv2d(R,X1);

[0149] Among them, R 21 Indicates the result of integrating features of the image region;

[0150] S1 represents the image region feature integration module.

[0151] According to one embodiment of the present invention, the image region feature integration result is input into the image region basic screening module to obtain the image region basic screening result. The image region basic screening module uses a sliding window method to determine a basic selection rectangular frame in the input feature map, where the formula is:

[0152] R 22 =S2(R 21 );

[0153] R 22 =Slide(R 21 );

[0154] Among them, R 22 Indicates the basic screening results of the drawing area;

[0155] Slide means sliding window mode;

[0156] S2 represents the basic screening module of the image area.

[0157] According to one embodiment of the present invention, the basic screening result of the image area is input into the image area preliminary prediction module to obtain the image area preliminary prediction result. The image area preliminary prediction module first performs three convolution operations with a convolution kernel of 1×1 on the input, and then merges the last two dimensions of the convolution result, and sets the second dimension to 2, indicating that each picture has two groups of probability values, the first group represents the probability that the rectangular box belongs to the background, and the second group represents the probability that the rectangular box belongs to the foreground, and then performs Softmax processing on the last dimension in the output of the previous step to obtain the third group representing the probability that the rectangular box belongs to the target, and finally performs dimension merging again, and only retains the result with the third dimension index of 1, that is, the probability that each rectangular box contains the target, where the formula is,

[0158] R 23 =S3(R 22 );

[0159] R 23 =ReShape(Softmax(ReShape(Conv(Conv(Conv(R 22 ))))));

[0160] Among them, R 23 Indicates the preliminary prediction results of the map area;

[0161] ReShape means merging the last two dimensions;

[0162] Conv means convolution with a convolution kernel of 1×1;

[0163] S3 represents the preliminary prediction module of the image region.

[0164] According to one embodiment of the present invention, the preliminary prediction result of the image area and the image area feature integration result are input into the image area regression prediction module to obtain the image area regression prediction result. The image area regression prediction module first performs three convolution operations with a convolution kernel of 1×1 on the above result to obtain four groups of offsets, which respectively represent the x-coordinate offset of the center point of all rectangular boxes, the y-coordinate offset of the center point, the width offset, and the height offset. The rectangular box obtained by sliding the picture is recorded as A, and (A X , A y , A w ,Ah ) represents the actual rectangular box, and (G X , G y , G w , G h ), the offset is denoted as d, and (d X , d y , d w , d h ) indicates that the four weight parameter matrices W are obtained by training and learning using the SmoothL1 loss function, which are respectively represented by (W X , W y , W w , W h ), and finally the offset of the rectangular box can be predicted according to W, where the formula is:

[0165] R 24 =S4(R 23 , R 21 );

[0166] S4=SmoothL1(Conv(Conv(Conv(R 23 , R 21 ))));

[0167] Among them, R 24 Indicates the regression prediction result of the image area;

[0168] S4 represents the image region regression prediction module;

[0169] SmoothL1 represents a smooth loss function.

[0170] According to one embodiment of the present invention, the image region regression prediction result is input into the image target region screening module to obtain the image region extraction result. The image target region screening module first uses bilinear interpolation to perform the first quantization on the input, aligns the network, and then uses the rounding up method to perform the second quantization, divides the feature grid into sub-regions, and finally performs the maximum pooling on each sub-region, where the formula is,

[0171] X2=S5(R 24 );

[0172] X2=MaxPool(RoundUp(BilinearInterpolation(R 24 )));

[0173] Among them, X2 represents the surface area extraction result;

[0174] S5 represents the image target area screening module;

[0175] RoundUp means rounding up;

[0176] BilinearInterpolation represents bilinear interpolation.

[0177] According to one embodiment of the present invention, the drawing features and drawing area extraction results of the architectural drawings are input into the drawing object classification module to obtain the drawing object classification results. The drawing object classification module first extracts the Conv and Pooling layers of the input feature network to generate a feature map, and then uses ROIPooling to scale the candidate area to a size of 7×7. After that, each scaled rectangular area is expanded and input into the classification and regression subnetwork for classification and bounding box regression, where the formula is,

[0178] R 31 =C1(X2,X1);

[0179] R 31 =Reg_head(Cls_head(MaxPool(Feature(X2,X1))));

[0180] Among them, R 31 Indicates the image target classification result;

[0181] Reg_head represents the bounding box regression subnetwork;

[0182] Cls_head represents the classification subnetwork;

[0183] Feature represents the Conv and Pooling layers of the feature extraction network;

[0184] C1 represents the image object classification module;

[0185] The drawing target classification results and drawing area extraction results are input into the drawing target position refinement module to obtain the architectural drawing drawing detection results. The drawing target position refinement module first calculates the position of the input rectangular area relative to the feature map, and then calculates the size and position of the Pool area corresponding to each rectangular area, and interpolates each rectangular area to obtain the final feature value and splices each rectangular area. The formula is,

[0186] X3=C2(R 31 , X2);

[0187] X3=Hermite(AvgPool(Scale(R 31 , X2)));

[0188] Among them, X3 represents the inspection result of architectural drawing;

[0189] C2 represents the image target location fine module;

[0190] Hermite represents interpolation calculation;

[0191] Scale represents the scaling ratio calculation of the rectangular area relative to the feature map.

[0192] Furthermore, in order to achieve the above-mentioned purpose of the invention, the present invention also provides an intelligent inspection system for compliance of architectural drawings. Figure 3 A flowchart schematically shows a system for intelligently checking the compliance of architectural drawings according to the present invention, as shown in FIG. Figure 3 As shown, according to an intelligent system for checking compliance of architectural drawings and surfaces of the present invention, the system comprises:

[0193] Architectural drawing image acquisition module: acquires the architectural drawing image to be detected;

[0194] Architectural drawing surface feature acquisition module: input the architectural drawing surface image to be detected into the surface feature extraction neural network to obtain the architectural drawing surface features, and the surface feature extraction neural network includes a first calculation module, a second calculation module, a first enhanced calculation module, a second enhanced calculation module, a third enhanced calculation module and a result integration module;

[0195] Drawing area extraction result acquisition module: inputting the drawing features of the architectural drawings and the drawing images to be detected into the drawing area extraction neural network to obtain the drawing area extraction results. The drawing area extraction neural network includes a drawing area feature integration module, a drawing area basic screening module, a drawing area preliminary prediction module, a drawing area regression prediction module and a drawing target area screening module;

[0196] Architectural drawing surface detection result acquisition module: input the architectural drawing surface features and drawing area extraction results into the drawing target classification determination neural network to obtain the architectural drawing surface detection results. The drawing target classification determination neural network includes a drawing target classification module and a drawing target position refinement module.

[0197] In order to achieve the above-mentioned purpose of the invention, the present invention also provides an electronic device, which includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, and when the computer program is executed by the processor, the above-mentioned intelligent method for checking the compliance of architectural drawings is implemented.

[0198] In order to achieve the above-mentioned purpose of the invention, the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned intelligent inspection method for compliance of architectural drawings is implemented.

[0199] Based on this, the beneficial effect of the present invention lies in that by adopting deep learning image segmentation technology to extract blank axis numbers, redundant review cloud lines, and garbled font areas in architectural drawings, and by identifying the special features of blank axis numbers, redundant review cloud lines, and garbled fonts, blank axis numbers, redundant review cloud lines, and garbled font areas can be accurately segmented, thereby determining whether the drawings meet the requirements of the specifications.

[0200] Those of ordinary skill in the art will appreciate that the modules and algorithm steps described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of 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 the present invention.

[0201] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and equipment can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0202] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0203] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the embodiments of the present invention.

[0204] In addition, each functional module in the embodiment of the present invention may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0205] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for enabling 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 energy-saving signal sending / receiving method of each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a ROM, a RAM, a magnetic disk, or an optical disk.

[0206] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the present application.

[0207] It should be understood that the size of the serial numbers of each step in the content of the invention and the embodiments of the present invention does not absolutely mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

Claims

1. An intelligent method for checking the compliance of architectural drawings, characterized in that: include: Obtaining an image of a building drawing to be detected; Inputting the architectural drawing surface image to be detected into a surface feature extraction neural network to obtain architectural drawing surface features, wherein the surface feature extraction neural network includes a first calculation module, a second calculation module, a first enhanced calculation module, a second enhanced calculation module, a third enhanced calculation module and a result integration module; Inputting the architectural drawing surface features and the architectural drawing surface images to be detected into a drawing area extraction neural network to obtain a drawing area extraction result, wherein the drawing area extraction neural network includes a drawing area feature integration module, a drawing area basic screening module, a drawing area preliminary prediction module, a drawing area regression prediction module and a drawing target area screening module; The architectural drawing surface features and the drawing area extraction results are input into a drawing target classification determination neural network to obtain an architectural drawing surface detection result. The drawing target classification determination neural network includes a drawing target classification module and a drawing target position refinement module.

2. According to claim 1, a method for intelligently checking the compliance of architectural drawings is characterized in that: The architectural drawing image to be detected is input into the first calculation module to obtain a first calculation result, where the formula is: R 11 =M1(R1); R 11 =MzxPool(ReLU(Conv2d(Conv2d(R1)))); Among them, R 11 represents the first calculation result; M1 represents the first computing module; R1 represents the architectural drawing image to be detected; MaxPool indicates the maximum pooling layer with a pooling kernel of 2×2; ReLU stands for rectified linear unit activation function; Conv2d represents a 3×3 convolutional layer; The first calculation result is input into the second calculation module to obtain a second calculation result, wherein the formula is: R 12 =M2(R 11 ); R 12 =MaxPool(ReLU(Conv2d(Conv2d(R 11 )))); Among them, R 12 represents the second calculation result; M2 represents the second computing module; The architectural drawing image to be detected and the second calculation result are input into the first enhanced calculation module to obtain a first enhanced calculation result, wherein the formula is: R 13 =T1(R 12 ,R1); R 13 =MaxPool(ReLU(Conv2d(Conv2d(Conv2d(R 12 ,R1))))); Among them, R 13 represents the first enhanced calculation result; T1 represents the first enhanced computing module; The first enhanced calculation result is continuously input into the second enhanced calculation module and the third enhanced calculation module to obtain the second enhanced calculation result and the third enhanced calculation result, wherein the formula is, R 14 =T2(R 13 ); R 15 =T3(R 14 ); R 14 =MaxPool(ReLU(Conv2d(Conv2d(Conv2d(R 13 ))))); R 15 =MaxPool(ReLU(Conv2d(Conv2d(Conv2d(R 14 ))))); Among them, R 14 represents the second enhanced calculation result; R 15 represents the third enhanced calculation result; T2 represents the second enhanced computing module; T3 represents the third enhanced computing module.

3. The method for intelligently checking the compliance of architectural drawings according to claim 2 is characterized in that: The first enhanced calculation result and the third enhanced calculation result are input into the result integration module to obtain the architectural drawing surface feature, where the formula is: X1=F(R 13 ,R 15 ); X1=AvgPool(FC2(FC1(Softmax(R 13 ,R 15 )))); Among them, X1 represents the characteristics of the architectural drawing; F represents the result integration module; FC1 represents the first fully connected layer; FC2 represents the second fully connected layer; Softmax represents the output normalized exponential function; AvgPool represents an average pooling layer with a pooling kernel of 2×2.

4. The method for intelligently checking the compliance of architectural drawings according to claim 3 is characterized in that: The architectural drawing surface features and the architectural drawing surface image to be detected are input into the surface area feature integration module to obtain the surface area feature integration result, where the formula is: R 21 =S1(R1,X1); R 21 =Conv2d(R,X1); Among them, R 21 Indicates the result of integrating features of the image region; S1 represents the image region feature integration module.

5. The method for intelligently checking the compliance of architectural drawings according to claim 4 is characterized in that: The image area feature integration result is input into the image area basic screening module to obtain the image area basic screening result, where the formula is: R 22 =S2(R 21 ); R 22 =Slide(R 21 ); Among them, R 22 Indicates the basic screening results of the drawing area; Slide means sliding window mode; RS2 represents the surface area basic screening module.

6. The intelligent inspection method for compliance of architectural drawings according to claim 5 is characterized in that: The basic screening result of the drawing area is input into the preliminary prediction module of the drawing area to obtain the preliminary prediction result of the drawing area, wherein the formula is: R 23 =S3(R 22 ); R 23 =ReShape(Softmax(ReShape(Conv(Conv(Conv(R 22 )))))); Among them, R 23 Indicates the preliminary prediction result of the drawing area; ReShape means merging the last two dimensions; Conv means convolution with a convolution kernel of 1×1; S3 represents the preliminary prediction module of the image region.

7. The method for intelligently checking the compliance of architectural drawings according to claim 6, characterized in that: The preliminary prediction result of the drawing area and the integrated result of the drawing area features are input into the drawing area regression prediction module to obtain the drawing area regression prediction result, where the formula is: R 24 =S4(R 23 ,R 21 ); S4=SmoothL1(Conv(Conv(Conv(R 23 ,R 21 )))); Among them, R 24 Indicates the regression prediction result of the image area; S4 represents the image region regression prediction module; SmoothL1 represents a smooth loss function.

8. The intelligent inspection method for compliance of architectural drawings according to claim 7 is characterized in that: The image area regression prediction result is input into the image target area screening module to obtain the image area extraction result, where the formula is: X2=S5(R 24 ); X2=MaxPool(RoundUp(BilinearInterpolation(R 24 ))); Among them, X2 represents the surface area extraction result; S5 represents the image target area screening module; RoundUp means rounding up; BilinearInterpolation represents bilinear interpolation.

9. The method for intelligently checking the compliance of architectural drawings according to claim 8, characterized in that: The architectural drawing features and the drawing area extraction results are input into the drawing object classification module to obtain the drawing object classification results, where the formula is: R 31 =C1(X2,X1); R 31 =Reg_head(Cls_head(MaxPool(Feature(X2,X1)))); Among them, R 31 Indicates the image target classification result; Reg_head represents the bounding box regression subnetwork; Cls_head represents the classification subnetwork; Feature represents the Conv and Pooling layers of the feature extraction network; C1 represents the image object classification module; The drawing object classification result and the drawing area extraction result are input into the drawing object position refinement module to obtain the building drawing drawing detection result, where the formula is: X3=C2(R 31 ,X2); X3=Hermite(AvgPool(Scale(R 31 ,X2))); Among them, X3 represents the inspection result of architectural drawing; C2 represents the image target location fine module; Hermite represents interpolation calculation; Scale represents the scaling ratio calculation of the rectangular area relative to the feature map.

10. An intelligent inspection system for compliance of architectural drawings, characterized in that: include: Architectural drawing image acquisition module: acquires the architectural drawing image to be detected; Architectural drawing surface feature acquisition module: input the architectural drawing surface image to be detected into the surface feature extraction neural network to obtain the architectural drawing surface features, and the surface feature extraction neural network includes a first calculation module, a second calculation module, a first enhanced calculation module, a second enhanced calculation module, a third enhanced calculation module and a result integration module; Drawing area extraction result acquisition module: input the architectural drawing drawing features and the architectural drawing drawing image to be detected into the drawing area extraction neural network to obtain the drawing area extraction result, and the drawing area extraction neural network includes a drawing area feature integration module, a drawing area basic screening module, a drawing area preliminary prediction module, a drawing area regression prediction module and a drawing target area screening module; Architectural drawing surface detection result acquisition module: input the architectural drawing surface features and the drawing area extraction results into the drawing target classification determination neural network to obtain the architectural drawing surface detection results. The drawing target classification determination neural network includes a drawing target classification module and a drawing target position refinement module.

Images