Real estate project progress monitoring method and device

Abstract

The invention discloses a real estate project progress monitoring method and device. The method comprises the steps of obtaining basic information of a project and current image data of project construction; performing denoising processing on the current image data to obtain denoised image data; carrying out super-resolution reconstruction on the de-noised image data to obtain enhanced image data; inputting the enhanced image data into a target detection model to obtain a target detection result; wherein the target detection result comprises a constructed area; and calculating the ratio of the constructed area to the current planning construction area to obtain the project progress. According to the method and the device, the coverage of the whole project area is ensured by collecting comprehensive image data, the real situation of the project can be comprehensively known, the real situation can be provided for risk early warning, and effective risk early warning is ensured.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a method and device for monitoring the progress of a real estate project. Background Art

[0002] Since real estate and banks are deeply connected, real estate defaults are very likely to threaten the credit and liquidity of banks. After banks lend money to real estate developers, they face the problem of difficulty in controlling the flow and use of funds. Therefore, real estate project data needs to be monitored in real time to complete risk monitoring.

[0003] At present, risk monitoring is generally carried out by manually collecting data on real estate projects, mainly including manually collecting project construction progress and completion quality, entering them into relevant systems or manually compiling reports, and finally tracking the situation through relevant data to achieve project risk monitoring. However, most of the manually collected real estate project data are point-like and isolated, and it is impossible to fully understand the real situation of the project, which leads to insufficient in-depth risk analysis and judgment, thus affecting the effect of risk warning.

[0004] Therefore, how to ensure effective risk warning has become an urgent problem to be solved in this field. Summary of the invention

[0005] The present application provides a method and device for monitoring the progress of a real estate project, the purpose of which is to ensure effective risk warning.

[0006] In order to achieve the above objectives, this application provides the following technical solutions:

[0007] A method for monitoring progress of a real estate project, comprising:

[0008] Obtaining basic information of the project and current image data of the project construction; wherein the basic information of the project at least includes the current planned construction area;

[0009] Performing denoising processing on the current image data to obtain denoised image data;

[0010] Performing super-resolution reconstruction on the denoised image data to obtain enhanced image data;

[0011] Inputting the enhanced image data into a target detection model to obtain a target detection result; wherein the target detection result includes an area where construction has been completed;

[0012] The ratio between the area that has completed construction and the area that is currently planned to be constructed is calculated to obtain the project progress.

[0013] Optionally, the performing denoising processing on the current image data to obtain denoised image data includes:

[0014] Performing wavelet transform on the current image data to obtain a plurality of high-frequency sub-bands and low-frequency sub-bands;

[0015] For each of the high frequency sub-bands, determining whether the high frequency sub-band is greater than a preset threshold;

[0016] If the high frequency sub-band is greater than the preset threshold, calculating the absolute value of the difference between the high frequency sub-band and the preset threshold, and determining it as a new high frequency sub-band;

[0017] If the high frequency sub-band is not greater than the preset threshold, the high frequency sub-band is determined to be 0 and marked as a new high frequency sub-band;

[0018] Based on all new high-frequency sub-bands and the low-frequency sub-bands, denoised image data is constructed.

[0019] Optionally, performing super-resolution reconstruction on the denoised image data to obtain enhanced image data includes:

[0020] Scaling the denoised image data according to a preset scaling factor to obtain new image data;

[0021] For each target pixel in the new image data, determining a position of the target pixel in the denoised image data according to the preset zoom factor, and marking the position as the target position;

[0022] Searching for a plurality of pixel points closest to the target position in the denoised image data;

[0023] The weight of each pixel is calculated using a bicubic basis function; wherein the weight of the pixel indicates a contribution value of the pixel to the target pixel;

[0024] Performing weighted summation on each pixel using the weight of each pixel to obtain a pixel value of a target pixel;

[0025] Based on the pixel values ​​of all target pixels, the enhanced image data is constructed.

[0026] Optionally, inputting the enhanced image data into a target detection model to obtain a target detection result includes:

[0027] Inputting the enhanced image data into the convolutional layer of the target detection model to obtain a feature map;

[0028] Inputting the feature map into the RPN network of the target detection model to obtain candidate regions of multiple scales;

[0029] Inputting the feature map and the multiple candidate regions of the proportions into the ROI pooling layer of the target detection model to obtain multiple candidate regions of the same proportions;

[0030] The multiple candidate regions of the same proportion are input into the classifier of the target detection model to obtain the target detection result.

[0031] Optionally, the feature map is input into an RPN network of a target detection model to obtain candidate regions of multiple proportions, including:

[0032] Inputting the feature map into the RPN network of the target detection model to obtain anchor frames of multiple scales;

[0033] For each anchor frame of the proportion, a Softmax layer is used to determine whether the anchor frame is the background;

[0034] If the anchor frame is the background, deleting the anchor frame;

[0035] If the anchor frame is not the background, a correction value of the anchor frame is calculated, and the anchor frame is corrected by using the correction value of the anchor frame to obtain a corrected anchor frame;

[0036] The corrected anchor box is determined as a candidate region.

[0037] Optionally, inputting the multiple candidate regions of the same proportion into a classifier of a target detection model to obtain a target detection result includes:

[0038] For each candidate region of the same proportion, input the candidate region into a classifier of a target detection model, so as to calculate the probability of the category to which the candidate region belongs by using a fully connected layer and a Softmax activation function in the classifier, and obtain a probability vector, wherein the probability vector includes multiple probabilities;

[0039] For each of the probabilities, determining whether the probability is greater than a probability threshold;

[0040] If the probability is greater than the probability threshold, determining the target detection result as the candidate area belonging to the category;

[0041] If the probability is not greater than the probability threshold, the target detection result is determined as the candidate area not belonging to the category.

[0042] A monitoring device for real estate project progress, comprising:

[0043] An acquisition unit, used to acquire basic information of the project and current image data of the project construction; wherein the basic information of the project at least includes the current planned construction area;

[0044] A denoising unit, used for performing denoising processing on the current image data to obtain denoised image data;

[0045] An enhancement unit, used for performing super-resolution reconstruction on the denoised image data to obtain enhanced image data;

[0046] A detection unit, used to input the enhanced image data into a target detection model to obtain a target detection result; wherein the target detection result includes an area where construction has been completed;

[0047] The calculation unit is used to calculate the ratio between the area that has completed construction and the area that is currently planned to be constructed, so as to obtain the project progress.

[0048] Optionally, the denoising unit is specifically used for:

[0049] Performing wavelet transform on the current image data to obtain a plurality of high-frequency sub-bands and low-frequency sub-bands;

[0050] For each of the high frequency sub-bands, determining whether the high frequency sub-band is greater than a preset threshold;

[0051] If the high frequency sub-band is greater than the preset threshold, calculating the absolute value of the difference between the high frequency sub-band and the preset threshold, and determining it as a new high frequency sub-band;

[0052] If the high frequency sub-band is not greater than the preset threshold, the high frequency sub-band is determined to be 0 and marked as a new high frequency sub-band;

[0053] Based on all new high-frequency sub-bands and the low-frequency sub-bands, denoised image data is constructed.

[0054] Optionally, the enhancement unit is specifically used for:

[0055] Scaling the denoised image data according to a preset scaling factor to obtain new image data;

[0056] For each target pixel in the new image data, determining a position of the target pixel in the denoised image data according to the preset zoom factor, and marking the position as the target position;

[0057] Searching for a plurality of pixel points closest to the target position in the denoised image data;

[0058] The weight of each pixel is calculated using a bicubic basis function; wherein the weight of the pixel indicates a contribution value of the pixel to the target pixel;

[0059] Performing weighted summation on each pixel using the weight of each pixel to obtain a pixel value of a target pixel;

[0060] Based on the pixel values ​​of all target pixels, the enhanced image data is constructed.

[0061] Optionally, the detection unit is specifically used for:

[0062] Inputting the enhanced image data into the convolutional layer of the target detection model to obtain a feature map;

[0063] Inputting the feature map into the RPN network of the target detection model to obtain candidate regions of multiple scales;

[0064] Inputting the feature map and the multiple candidate regions of the proportions into the ROI pooling layer of the target detection model to obtain multiple candidate regions of the same proportions;

[0065] The multiple candidate regions of the same proportion are input into the classifier of the target detection model to obtain the target detection result.

[0066] The technical solution provided by the present application obtains the basic information of the project and the current image data of the project construction; performs denoising on the current image data to obtain denoised image data; performs super-resolution reconstruction on the denoised image data to obtain enhanced image data; inputs the enhanced image data into the target detection model to obtain the target detection result; wherein the target detection result includes the area where the construction has been completed; calculates the ratio between the area where the construction has been completed and the area currently planned for construction to obtain the project progress. In the present application, by collecting comprehensive image data to ensure coverage of the entire project area, the real situation of the project can be fully understood, the real situation can be provided for risk warning, and effective risk warning can be ensured. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0068] Figure 1 A flowchart of a method for monitoring the progress of a real estate project provided in an embodiment of the present application;

[0069] Figure 2 A schematic diagram of data entry provided in an embodiment of the present application;

[0070] Figure 3 A schematic diagram of a remote sensing image in a project image library provided in an embodiment of the present application;

[0071] Figure 4 A flowchart of a method for denoising image data provided in an embodiment of the present application;

[0072] Figure 5 A schematic diagram of an image denoising process provided in an embodiment of the present application;

[0073] Figure 6 A schematic diagram of image enhancement provided in an embodiment of the present application;

[0074] Figure 7 A flowchart of an image data resolution reconstruction method provided in an embodiment of the present application;

[0075] Figure 8 A schematic diagram of a pixel point provided in an embodiment of the present application;

[0076] Fig. 9 A schematic diagram of reconstructing image resolution provided in an embodiment of the present application;

[0077] Fig.10 A flow chart of a target detection method provided in an embodiment of the present application;

[0078] Fig.11 A flowchart of a method for determining a candidate region provided in an embodiment of the present application;

[0079] Fig.12 A schematic diagram of the basic structure of Faster RCNN, a target detection model provided in an embodiment of the present application;

[0080] Fig.13 A flowchart for determining the category to which a detection target belongs provided in an embodiment of the present application;

[0081] Fig.14 A schematic diagram of a target detection effect provided in an embodiment of the present application;

[0082] Fig.15 A schematic diagram of a project progress provided in an embodiment of the present application;

[0083] Fig.16 A schematic diagram of the architecture of a real estate project progress monitoring device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0084] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0085] In this application, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element.

[0086] like Figure 1 As shown, it is a flow chart of a method for monitoring the progress of a real estate project provided by an embodiment of the present application, comprising the following steps:

[0087] S101: Obtain basic information of the project and current image data of the project construction.

[0088] Among them, the basic information of the project at least includes the current planned construction area.

[0089] Optionally, since the information database pre-stores the customer information and the basic information of the project, the basic information of the project can be obtained from the information database.

[0090] It should be noted that before storing the basic information of the customer and the project in the information database, data needs to be entered. Only when the data is entered accurately can it be processed and analyzed. Data entry mainly involves entering data through materials such as loan contracts and debt information, including the preliminary data entry stage and the data entry stage during construction. Figure 2 .

[0091] The early data entry stage is generally carried out at the beginning of contract signing or before the start of real estate projects, mainly entering project information, contract data, and financing conditions. Project information mainly records customer information and basic project information in order to mark and distinguish different customers and projects. It can include customer name, project number, project name, total project investment amount, credit type, approved available amount, customer name of credit occupied, project planning, handling bank, corresponding account manager, etc. Contract data mainly records the important information involved in the contract. Contract data includes contract status, contract number, business number, customer number, contract effective date, contract start date, contract expiration date, main investment direction when the loan is issued, total contract amount, term, number of planned withdrawals, etc. Debt information records relevant data involving principal and interest, including debt status, whether it is a syndicated loan, interest rate type, interest rate effective date, interest rate determination basis, available credit type, guarantee method, business category, debt amount, current debt balance, loan transfer account, non-performing loan balance, etc.

[0092] During the data entry stage of construction, data is generally entered in stages and multiple times during the project construction process. This is mainly achieved by regularly obtaining the overall situation of the current project and the phased situation of the project construction. The phased situation of the project construction includes the preliminary construction preparation, recent construction progress, completed output value and completion ratio, project fund usage, current major problems and solutions, on-site construction images, etc. Project status information is collected regularly so that management personnel can better monitor the progress of the project throughout the entire process and supervise the legal and compliant use of loan funds.

[0093] It should be emphasized that different methods are used for data entry for paper contracts and electronic contracts. For paper contracts, ORC (Optical Character Recognition) technology is used to identify and then review the entered data. OCR technology is used to detect the text and data printed or handwritten on the contract and convert it into readable data, so that a large amount of text data can be quickly, conveniently, and time-savingly entered, realizing the "electronicization" of information processing. In order to further ensure the accuracy of the entered contract data, manual review is performed after OCR recognition to confirm the entered contract data item by item. For electronic contracts, data is directly extracted from the contract and then manually reviewed to double-guarantee the accuracy of the data.

[0094] In addition, the current image data of the project construction can be obtained from the project image library, because the project image library pre-stores the image data of the project construction at different periods, and the image data of the project construction at different periods can be obtained through the service platform of satellite or aircraft remote sensing technology.

[0095] They are marked as before construction, during construction, and after construction according to the period. Among them, the image data before construction is used to record the initial state before the start of the real estate project. The image data marked as under construction are obtained in batches according to the volume of the project, and the number of acquisitions is set independently. They are marked as the first stage of construction, the second stage of construction, the third stage of construction, and so on. The image data of different stages of the construction process are marked respectively to judge the progress of the project at different periods of construction. Since real estate projects often have a large amount of engineering, the construction period is relatively long, and the engineering conditions during this period are prone to change. Engineering delays or extensions also occur in this period. Therefore, it is necessary to monitor this period in different time periods and obtain image data of this period regularly in order to understand the progress of the project construction in a timely manner. The image data marked as after construction records the visual situation when the project is completed. If the entire project goes smoothly, the image data marked as after construction should be almost the same as the project plan. Overall, the satellite remote sensing image data of the real estate project are observed in order from the imaging time to the end. These image data will show the realization process of the entire project from existence to non-existence, and describe the gradual progress of the project construction from a visual perspective.

[0096] For details, see Figure 3 , shows a schematic diagram of remote sensing images in a project image library, including remote sensing image schematic diagrams corresponding to various periods in the construction process of a real estate project, where H is the building required by the project plan. Figure 1 in the remote sensing image schematic diagram before the project construction shows the situation before the start of construction, at which time there is basically no building H required to be implemented in the project plan. Buildings H begin to appear in Figure 2-1 in the remote sensing image schematic diagram of the first stage of construction, and then more buildings H appear in Figure 2-2 in the remote sensing image schematic diagram of the second stage of construction. Compared with Figure 2-1 and Figure 2-2, the number of buildings H in Figure 2-3 in the remote sensing image schematic diagram of the third stage of construction continues to increase. During the construction process, the number of H appearing in the remote sensing image schematic diagram will gradually increase with the recommendation of the construction stage until the final completion of the project, that is, entering the post-construction period, when all the buildings H required by the project plan should appear in Figure 3 in the remote sensing image schematic diagram.

[0097] S102: Perform denoising processing on the current image data to obtain denoised image data.

[0098] It is understandable that since the current image data is obtained through remote sensing WeChat or high-altitude payload equipment such as aircraft, it is affected by noise such as meteorology, atmosphere, and surface clutter, resulting in image quality damage and blurred imaging. Therefore, in order to ensure the accuracy of subsequent image recognition, it is necessary to first denoise the current image data to remove the noise in the current image data to ensure the image's detailed information and good visual effects.

[0099] Optionally, the current image data may be denoised by domain averaging, median filtering, low-pass filtering, Fourier transform, wavelet transform, and other methods.

[0100] Optionally, in another embodiment of the present application, the specific implementation of step S102 is as follows: Figure 4 As shown, the following steps are included:

[0101] S401: Perform wavelet transform on current image data to obtain multiple high-frequency sub-bands and low-frequency sub-bands.

[0102] It can be understood that the wavelet transform is performed on the current image data so that the current image data is converted from the spatial domain to the wavelet transform domain to obtain multiple high-frequency sub-bands and low-frequency sub-bands (eg, four groups of frequency coefficients).

[0103] It should be noted that the high-frequency sub-band and the low-frequency sub-band refer to high-frequency coefficients and low-frequency coefficients.

[0104] Among them, the low-frequency sub-band usually contains the main structure of the image, and the high-frequency sub-band contains noise. Therefore, it is not necessary to perform noise processing on the low-frequency sub-band, but only on the high-frequency sub-band.

[0105] S402: For each high frequency sub-band, determine whether the high frequency sub-band is greater than a preset threshold.

[0106] If the high frequency sub-band is greater than the preset threshold, step S403 is executed; if the high frequency sub-band is not greater than the preset threshold, step S404 is executed.

[0107] Specifically, assuming that the high frequency sub-band is 5 and the preset threshold is 4, it is determined whether the high frequency sub-band is greater than the preset threshold. Obviously, the high frequency sub-band is greater than the preset threshold, so step S403 is continued.

[0108] S403: Calculate the absolute value of the difference between the high frequency sub-band and a preset threshold, and determine it as a new high frequency sub-band.

[0109] It can be understood that by calculating the absolute value of the difference between the high frequency sub-band and the preset threshold and determining it as a new high frequency sub-band, the noise in the high frequency sub-band is removed so that there is no noise in the new high frequency sub-band.

[0110] S404: Determine the high frequency sub-band as 0 and mark it as a new high frequency sub-band.

[0111] It can be understood that the high frequency sub-band is determined to be 0 to remove the noise in the high frequency sub-band, so that no noise exists in the new high frequency sub-band.

[0112] S405: Construct denoised image data based on all new high-frequency sub-bands and low-frequency sub-bands.

[0113] It can be understood that since there is no noise in all new high-frequency sub-bands and the low-frequency sub-bands contain the main structure of the image, the denoised image data is constructed based on all new high-frequency sub-bands and low-frequency sub-bands.

[0114] based on Figure 4 From the content, we can see that the signal and noise are separated in the transform domain, and then the noise is filtered out, and the remaining is the signal. Separating the signal and noise in the transform domain means first converting the image from the spatial domain to the transform domain according to the prescribed transform operation, then removing the noise in different frequency components in the transform domain, and finally reconstructing the image to the original spatial domain using the inverse transform operation corresponding to the previous transform operation to obtain the denoised image.

[0115] For details, see Figure 5 , a schematic diagram of an image denoising process is shown. Figure 5In the figure, Figure_p1 is the current image data, and Figure_p2 is the image data after denoising. Comparing Figure_p1 and Figure_p2, it is obvious that the denoised image of Figure_p2 is clearer.

[0116] S103: Perform super-resolution reconstruction on the denoised image data to obtain enhanced image data.

[0117] It is understandable that due to the high altitude advantage of remote sensing satellites, remote sensing technology can cover a wide area and easily obtain image data over a large area. However, long-distance imaging inevitably brings about the problem of low image resolution. Low-resolution remote sensing images will affect the subsequent use of remote sensing images to judge the progress of construction and will also affect the visual experience. Therefore, in order to better use remote sensing images, it is necessary to reconstruct their resolution.

[0118] Optionally, the resolution of the denoised image data may be reconstructed using a super-resolution reconstruction method.

[0119] Combining step S102 and step S103, it can be seen that the image data is enhanced by image denoising and resolution reconstruction, so that the enhanced image data can be used to judge the construction progress later. Figure 6 , a schematic diagram of image enhancement is shown. Figure 6 In the figure, Figure_p1 is the current image data. By denoising the current image data, Figure_p2, i.e. the denoised image data, is obtained. Then, the super-resolution reconstruction method is used to reconstruct the resolution of the denoised image data to obtain Figure_p3, i.e. the enhanced image data.

[0120] Optionally, in another embodiment of the present application, the specific implementation of step S103 is as follows: Figure 7 As shown, the following steps are included:

[0121] S701: Scaling the denoised image data according to a preset scaling factor to obtain new image data.

[0122] If the size of the denoised image data is m×n and the preset scaling factor is k, the denoised image data is scaled according to the preset scaling factor to obtain new image data of M×N, that is, k=M / n.

[0123] S702: For each target pixel in the new image data, determine the position of the target pixel in the denoised image data according to a preset zoom factor, and mark it as the target position.

[0124] It can be understood that since the value of each pixel in the denoised image data is known, and the value of the target pixel in the new image data is unknown, in order to determine the value of the target pixel, the position of the target pixel (ie, X, Y) in the denoised image data is first determined according to a preset zoom factor and marked as the target position (ie, A (x, y)).

[0125] S703: Searching for a plurality of pixel points closest to the target position in the denoised image data.

[0126] The plurality of pixels may be 16 pixels. For details, see Figure 8 , a schematic diagram of a pixel point is shown. Figure 8 In the figure, A is the target position, and the adjacent 16 pixels are the four points a in the first column. 00 ,a 10 ,a 20 ,a 30 , the four points a in the second column 01 ,a 11 ,a 21 ,a 31 , the four points a in the third column 02 ,a 12 ,a 22 ,a 32 , and the four points a in the fourth column 03 ,a 13 ,a 23 ,a 33 .

[0127] S704: Calculate the weight of each pixel using the bicubic basis function.

[0128] The weight of a pixel indicates the contribution value of the pixel to the target pixel.

[0129] In step S704, the specific expression form of the bicubic basis function is as shown in formula (1).

[0130] (1)

[0131] In formula (1), W(r) is the weight corresponding to the pixel, a is a constant, generally a=0.5, and r is the distance from each point in the neighborhood to the target pixel (X, Y) in the denoised image.

[0132] S705: Perform weighted summation on each pixel using the weight of each pixel to obtain a pixel value of the target pixel.

[0133] The specific form of weighted summing of each pixel using the weight of each pixel is shown in formula (2).

[0134] (2)

[0135] In formula (2), r ij The coordinates in the field are (x ij ,y ij ) ij The distance to the target location A, W(x-xi) and W(y-yi) are the weights calculated by formula (1). If the point in the 4×4 neighborhood of the target location A is denoted by a ij (x i ,y j ), i=0,1,2,3 and j=0,1,2,3, substitute them into formula (2) to obtain the pixel value of the target pixel.

[0136] S706: Construct enhanced image data based on the pixel values ​​of all target pixels.

[0137] It can be understood that after the pixel values ​​of all target pixels are calculated, higher-resolution image data (ie, enhanced image data) can be constructed based on the pixel values ​​of all pixels.

[0138] In order to better Figure 7 The following is a detailed description of the content shown in the figure. Fig. 9 A schematic diagram of the reconstructed image resolution is shown in FIG. Fig. 9 In FIG. 1 , FIG_p2 is the denoised image data. After super-resolution reconstruction (eg, bicubic interpolation), that is, after steps S701 to S706 , enhanced image data FIG_p3 is obtained.

[0139] S104: Input the enhanced image data into the target detection model to obtain the target detection result.

[0140] Among them, the target detection results include areas where construction has been completed.

[0141] It should be noted that the target detection model is a pre-trained model, which is pre-trained based on sample image data as input and the target detection results manually annotated on the sample image data as training targets. The target detection model includes but is not limited to a single-stage target detection model and a two-stage target detection model.

[0142] The single-stage target detection model is a model that can realize target detection by extracting features only once, that is, it can determine the target position and category in the image in a single stage. Compared with the two-stage method, it is slightly faster but slightly less accurate. Specifically, the main algorithms for single-stage target detection include YOLO series, SSD, RetinaNet, and key point-based detection methods.

[0143] The two-stage target detection model divides the target detection task into two stages. The first stage proposes candidate boxes that may contain targets and preliminarily selects the target area. The second stage determines the target category of the candidate boxes proposed in the previous stage and refines the target area in order to accurately select the target. Compared with the single-stage target detection model, the accuracy is slightly higher but the speed is slightly slower. Specifically, the main algorithms for two-stage target detection include RCNN, Fast RCNN, Faster RCNN, Cascade RCNN, SPPNet, RFCN, etc.

[0144] It should be emphasized that the two-stage target detection model Faster RCNN is adopted in the embodiment of the present application. Faster RCNN is a deep learning detection algorithm with close real-time performance in detection speed, and at the same time has excellent performance in detection accuracy.

[0145] Optionally, in another embodiment of the present application, the specific implementation of step S104 is as follows: Fig.10 As shown, the following steps are included:

[0146] S1001: Input the enhanced image data into the convolutional layer of the target detection model to obtain a feature map.

[0147] It can be understood that the enhanced image data is input into the convolution layer of the target detection model, and the feature map in the image is extracted through the combination of the convolution layer, the activation function ReLU, and the pooling layer to obtain the feature map.

[0148] S1002: Input the feature map into the RPN network of the target detection model to obtain candidate regions of multiple scales.

[0149] Among them, the feature map is input into the RPN network of the target detection model to generate reference frames of various proportions and sizes, and the candidate areas are screened out from the reference frames.

[0150] Optionally, in another embodiment of the present application, the specific implementation of step S1002 is as follows: Fig.11 As shown, the following steps are included:

[0151] S1101: Input the feature map into the RPN network of the target detection model to obtain anchor frames of multiple scales.

[0152] Among them, the feature map is input into the RPN network of the target detection model to obtain anchor frames of multiple scales, that is, reference frames of various scales and sizes.

[0153] S1102: For each anchor frame of each ratio, use the Softmax layer to determine whether the anchor frame is the background.

[0154] If the anchor frame is the background, execute step S1103; if the anchor frame is not the background, execute step S1104.

[0155] It can be understood that when the anchor frame is not the background, it means that the anchor frame at this time includes the area where the construction has been completed, and step S1104 is executed at this time.

[0156] S1103: Delete the anchor box.

[0157] If the anchor frame is the background, it means that the anchor frame does not contain the constructed area and needs to be deleted.

[0158] S1104: Calculate a correction value of the anchor frame, and use the correction value of the anchor frame to correct the anchor frame to obtain a corrected anchor frame.

[0159] Optionally, the anchor box may be corrected using a bounding box regression strategy, that is, a correction value of the anchor box is calculated, and the anchor box is corrected using the correction value of the anchor box to obtain a corrected anchor box.

[0160] S1105: Determine the corrected anchor box as a candidate region.

[0161] Among them, the rectangle can more accurately frame the area in the image that may contain the target (that is, the area where the construction has been completed).

[0162] S1003: Input the feature map and candidate regions of multiple proportions into the ROI pooling layer of the target detection model to obtain multiple candidate regions of the same proportion.

[0163] It can be understood that the feature map and candidate regions of multiple scales are input into the ROI pooling layer of the target detection model. The role of the ROI pooling layer is to map the candidate regions to the feature map and normalize the feature maps corresponding to candidate regions of different sizes and shapes to a fixed size.

[0164] S1004: Input multiple candidate regions of the same proportion into the classifier of the target detection model to obtain the target detection result.

[0165] It should be noted that after obtaining the target detection results, it is necessary to use the bounding box regression strategy to obtain the position offset of each candidate area, and correct the bounding box of the candidate area according to the position offset of each candidate area, so as to finally achieve more accurate target positioning.

[0166] Specifically, Fig.10 The content shown can be combined with Fig.12 To explain, Fig.12 In the method, the target detection model includes a convolutional layer, an RPN network, a RIO pooling layer and a classifier. The feature map is input into the RPN network to obtain the feature map. The feature map is input into the RPN network to obtain multiple candidate regions of different sizes. The multiple candidate regions of different sizes and the feature map are input into the RIO pooling layer to obtain multiple candidate regions of the same proportion. The multiple candidate regions of the same proportion are input into the classifier of the target detection model to obtain the target detection result.

[0167] Optionally, in another embodiment of the present application, the specific implementation of step S1004 is as follows: Fig.13 As shown, the following steps are included:

[0168] S1301: For each candidate region of the same proportion, the candidate region is input into the classifier of the target detection model to calculate the probability of the category to which the candidate region belongs using the fully connected layer and the Softmax activation function in the classifier to obtain a probability vector.

[0169] The probability vector includes multiple probabilities.

[0170] For example, when detecting completed areas of a real estate project, the candidate areas are input into the classifier of the target detection model, and the probability of the category to which the candidate areas belong is calculated using the fully connected layer and the Softmax activation function in the classifier to obtain a probability vector, which includes the probability of completed areas and the probability of unfinished areas.

[0171] S1302: For each probability, determine whether the probability is greater than a probability threshold.

[0172] If the probability is greater than the probability threshold, execute step S1303; if the probability is not greater than the probability threshold, execute step S1304.

[0173] Specifically, assuming that the probability is 90% and the probability threshold is 80%, it is determined whether the probability is greater than the probability threshold. Obviously, the probability is greater than the probability threshold, so step S1303 is continued.

[0174] It should be noted that the probability threshold can be set according to actual conditions and is not specifically limited here.

[0175] S1303: Determine the target detection result as a candidate region belonging to a category.

[0176] S1304: Determine the target detection result as a candidate region that does not belong to a category.

[0177] In order to better Fig.10 The following is a detailed description of the content shown in the figure. Fig.14 A schematic diagram of target detection is shown in FIG. Fig.14 In the process, the enhanced image data is input into FasterRCNN for target detection to obtain the target detection result. The target detection result is the area enclosed by the box in the output, that is, the area where construction has been completed.

[0178] S105: Calculate the ratio between the completed construction area and the currently planned construction area to obtain the project progress.

[0179] For details, see Fig.15 , Fig.15 Figure 1 is the currently planned construction area, and Figure 2 is the completed construction area. By calculating the ratio between the completed construction area and the currently planned construction area, the project progress is 50%.

[0180] To summarize, by acquiring image data of the project construction, performing image processing on the image data, and inputting the enhanced image data into the target detection model, the area where construction has been completed is obtained, and the area where construction has been completed is compared with the currently planned construction area to obtain the project progress. In the embodiment of the present application, by collecting comprehensive image data to ensure coverage of the entire project area, the real situation of the project can be fully understood, the real situation can be provided for risk warning, and effective risk warning can be ensured.

[0181] like Fig.16 As shown, it is a schematic diagram of the architecture of a real estate project progress monitoring device provided in an embodiment of the present application, and the monitoring device includes: an acquisition unit 100, a denoising unit 200, an enhancement unit 300, a detection unit 400 and a calculation unit 500.

[0182] The acquisition unit 100 is used to acquire basic information of the project and current image data of the project construction; wherein the basic information of the project at least includes the current planned construction area.

[0183] The denoising unit 200 is used to perform denoising processing on the current image data to obtain denoised image data.

[0184] The denoising unit 200 is specifically used for: performing wavelet transform on the current image data to obtain multiple high-frequency sub-bands and low-frequency sub-bands; for each high-frequency sub-band, determining whether the high-frequency sub-band is greater than a preset threshold; if the high-frequency sub-band is greater than the preset threshold, calculating the absolute value of the difference between the high-frequency sub-band and the preset threshold, and determining it as a new high-frequency sub-band; if the high-frequency sub-band is not greater than the preset threshold, determining the high-frequency sub-band as 0, and identifying it as a new high-frequency sub-band; based on all new high-frequency sub-bands and low-frequency sub-bands, constructing denoised image data.

[0185] The enhancement unit 300 is used to perform super-resolution reconstruction on the denoised image data to obtain enhanced image data.

[0186] The enhancement unit 300 is specifically used to: scale the denoised image data according to a preset zoom factor to obtain new image data; for each target pixel in the new image data, determine the position of the target pixel in the denoised image data according to the preset zoom factor, and mark it as the target position; search for multiple pixels closest to the target position in the denoised image data; calculate the weight of each pixel using a bicubic basis function; wherein the weight of the pixel indicates the contribution value of the pixel to the target pixel; perform weighted summation on each pixel using the weight of each pixel to obtain the pixel value of the target pixel; and construct enhanced image data based on the pixel values ​​of all target pixels.

[0187] The detection unit 400 is used to input the enhanced image data into the target detection model to obtain the target detection result; wherein the target detection result includes the area where the construction has been completed.

[0188] The detection unit 400 is specifically used to: input the enhanced image data into the convolution layer of the target detection model to obtain a feature map; input the feature map into the RPN network of the target detection model to obtain candidate regions of multiple proportions; input the feature map and the candidate regions of multiple proportions into the ROI pooling layer of the target detection model to obtain multiple candidate regions of the same proportion; input the multiple candidate regions of the same proportion into the classifier of the target detection model to obtain a target detection result.

[0189] The detection unit 400 is specifically used to: input the feature map into the RPN network of the target detection model to obtain anchor frames of multiple proportions; for each proportion of the anchor frame, use the Softmax layer to determine whether the anchor frame is the background; if the anchor frame is the background, delete the anchor frame; if the anchor frame is not the background, calculate the correction value of the anchor frame, and use the correction value of the anchor frame to correct the anchor frame to obtain the corrected anchor frame; determine the corrected anchor frame as the candidate area.

[0190] The detection unit 400 is specifically used for: for each candidate area of ​​the same proportion, inputting the candidate area into the classifier of the target detection model, so as to use the fully connected layer and the Softmax activation function in the classifier to calculate the probability of the category to which the candidate area belongs, and obtain a probability vector, which contains multiple probabilities; for each probability, judging whether the probability is greater than the probability threshold; if the probability is greater than the probability threshold, determining the target detection result as that the candidate area belongs to the category; if the probability is not greater than the probability threshold, determining the target detection result as that the candidate area does not belong to the category.

[0191] The calculation unit 500 is used to calculate the ratio between the area that has been completed and the area that is currently planned to be constructed, so as to obtain the project progress.

[0192] To summarize, by acquiring image data of the project construction, performing image processing on the image data, and inputting the enhanced image data into the target detection model, the area where construction has been completed is obtained, and the area where construction has been completed is compared with the currently planned construction area to obtain the project progress. In the embodiment of the present application, by collecting comprehensive image data to ensure coverage of the entire project area, the real situation of the project can be fully understood, the real situation can be provided for risk warning, and effective risk warning can be ensured.

[0193] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0194] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may 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.

[0195] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for monitoring the progress of a real estate project, characterized in that: include: Obtaining basic information of the project and current image data of the project construction; wherein the basic information of the project at least includes the current planned construction area; Performing denoising processing on the current image data to obtain denoised image data; Performing super-resolution reconstruction on the denoised image data to obtain enhanced image data; Inputting the enhanced image data into a target detection model to obtain a target detection result; wherein the target detection result includes an area where construction has been completed; The ratio between the area that has completed construction and the area that is currently planned to be constructed is calculated to obtain the project progress.

2. The method according to claim 1, characterized in that The denoising process is performed on the current image data to obtain denoised image data, including: Performing wavelet transform on the current image data to obtain a plurality of high-frequency sub-bands and low-frequency sub-bands; For each of the high frequency sub-bands, determining whether the high frequency sub-band is greater than a preset threshold; If the high frequency sub-band is greater than the preset threshold, calculating the absolute value of the difference between the high frequency sub-band and the preset threshold, and determining it as a new high frequency sub-band; If the high frequency sub-band is not greater than the preset threshold, the high frequency sub-band is determined to be 0 and marked as a new high frequency sub-band; Based on all new high-frequency sub-bands and the low-frequency sub-bands, denoised image data is constructed.

3. The method according to claim 1, characterized in that The step of reconstructing the resolution of the denoised image data to obtain enhanced image data includes: Scaling the denoised image data according to a preset scaling factor to obtain new image data; For each target pixel in the new image data, determining a position of the target pixel in the denoised image data according to the preset zoom factor, and marking the position as the target position; Searching for a plurality of pixel points closest to the target position in the denoised image data; The weight of each pixel is calculated using a bicubic basis function; wherein the weight of the pixel indicates a contribution value of the pixel to the target pixel; Performing weighted summation on each pixel using the weight of each pixel to obtain a pixel value of a target pixel; Based on the pixel values ​​of all target pixels, the enhanced image data is constructed.

4. The method according to claim 1, characterized in that: The step of inputting the enhanced image data into a target detection model to obtain a target detection result includes: Inputting the enhanced image data into the convolutional layer of the target detection model to obtain a feature map; Inputting the feature map into the RPN network of the target detection model to obtain candidate regions of multiple scales; Inputting the feature map and the multiple candidate regions of the proportions into the ROI pooling layer of the target detection model to obtain multiple candidate regions of the same proportions; The multiple candidate regions of the same proportion are input into the classifier of the target detection model to obtain the target detection result.

5. The method according to claim 4, characterized in that The feature map is input into the RPN network of the target detection model to obtain candidate regions of multiple proportions, including: Inputting the feature map into the RPN network of the target detection model to obtain anchor frames of multiple scales; For each anchor frame of the proportion, a Softmax layer is used to determine whether the anchor frame is the background; If the anchor frame is the background, deleting the anchor frame; If the anchor frame is not the background, a correction value of the anchor frame is calculated, and the anchor frame is corrected by using the correction value of the anchor frame to obtain a corrected anchor frame; The corrected anchor box is determined as a candidate region.

6. The method according to claim 4, characterized in that The step of inputting the plurality of candidate regions of the same proportion into a classifier of a target detection model to obtain a target detection result includes: For each candidate region of the same proportion, input the candidate region into a classifier of a target detection model, so as to calculate the probability of the category to which the candidate region belongs by using a fully connected layer and a Softmax activation function in the classifier, and obtain a probability vector, wherein the probability vector includes multiple probabilities; For each of the probabilities, determining whether the probability is greater than a probability threshold; If the probability is greater than the probability threshold, determining the target detection result as the candidate area belonging to the category; If the probability is not greater than the probability threshold, the target detection result is determined as the candidate area not belonging to the category.

7. A real estate project progress monitoring device, characterized in that: include: An acquisition unit, used to acquire basic information of the project and current image data of the project construction; wherein the basic information of the project at least includes the current planned construction area; A denoising unit, used for performing denoising processing on the current image data to obtain denoised image data; An enhancement unit, used for performing super-resolution reconstruction on the denoised image data to obtain enhanced image data; A detection unit, used to input the enhanced image data into a target detection model to obtain a target detection result; wherein the target detection result includes an area where construction has been completed; The calculation unit is used to calculate the ratio between the area that has completed construction and the area that is currently planned to be constructed, so as to obtain the project progress.

8. The device according to claim 7, characterized in that The denoising unit is specifically used for: Performing wavelet transform on the current image data to obtain a plurality of high-frequency sub-bands and low-frequency sub-bands; For each of the high frequency sub-bands, determining whether the high frequency sub-band is greater than a preset threshold; If the high frequency sub-band is greater than the preset threshold, calculating the absolute value of the difference between the high frequency sub-band and the preset threshold, and determining it as a new high frequency sub-band; If the high frequency sub-band is not greater than the preset threshold, the high frequency sub-band is determined to be 0 and marked as a new high frequency sub-band; Based on all new high-frequency sub-bands and the low-frequency sub-bands, denoised image data is constructed.

9. The device according to claim 7, characterized in that The enhancement unit is specifically used for: Scaling the denoised image data according to a preset scaling factor to obtain new image data; For each target pixel in the new image data, determining a position of the target pixel in the denoised image data according to the preset zoom factor, and marking the position as the target position; Searching for a plurality of pixel points closest to the target position in the denoised image data; The weight of each pixel is calculated using a bicubic basis function; wherein the weight of the pixel indicates a contribution value of the pixel to the target pixel; Performing weighted summation on each pixel using the weight of each pixel to obtain a pixel value of a target pixel; Based on the pixel values ​​of all target pixels, the enhanced image data is constructed.

10. The device according to claim 7, characterized in that The detection unit is specifically used for: Inputting the enhanced image data into the convolutional layer of the target detection model to obtain a feature map; Inputting the feature map into the RPN network of the target detection model to obtain candidate regions of multiple scales; Inputting the feature map and the multiple candidate regions of the proportions into the ROI pooling layer of the target detection model to obtain multiple candidate regions of the same proportions; The multiple candidate regions of the same proportion are input into the classifier of the target detection model to obtain the target detection result.

Images