Method, system, device and storage medium for remote sensing building detection

Abstract

The application discloses a method and system for remote sensing building detection; wherein the method comprises: constructing a training set, the training set being a remote sensing image with a known remote sensing building detection label; training a building detection model using the training set to obtain a trained building detection model, wherein an edge mask network is arranged in the trained building detection model; removing the edge mask network from the trained building detection model to obtain a corrected building detection model; obtaining a remote sensing image to be detected, inputting the remote sensing image to be detected into the corrected building detection model, and outputting a detection result of a building in the image. The edge mask label generated by the edge detection operator and the feature image generated by the edge mask network are used for training to enhance the position information, so that the object detection precision is improved without increasing the calculation cost.

Description

Technical Field

[0001] This invention relates to the field of remote sensing building image detection technology, and in particular to methods, systems, equipment and storage media for remote sensing building detection. Background Technology

[0002] The statements in this section merely refer to the background art related to this invention and do not necessarily constitute prior art.

[0003] Building detection from remote sensing imagery can replace traditional manual annotation, improving interpretation efficiency and playing a vital role in rural revitalization, urban planning, disaster detection, ecological environment management, and military applications. However, due to the complex backgrounds, varying resolutions, and diverse target representations in remote sensing images, achieving high-quality detection results is challenging. Currently, object detection algorithms based on deep convolutional neural networks are popular for building detection. Common single-stage object detection algorithms include the YOLO series, SSD, and RetinaNet, while common two-stage algorithms include R-CNN, Fast R-CNN, Faster R-CNN, Mask R-CNN, and Sparse R-CNN, all of which have achieved good detection accuracy in building detection.

[0004] Initially, the Faster R-CNN detection method used features from the last layer of the network to predict objects. However, its drawback was that it caused a decrease in accuracy for small objects in the detected image. Subsequently, a large number of studies emerged based on this, but due to the complex backgrounds of remote sensing building images and the diversity of target objects, the features generated by the backbone network and FPN for buildings still had a significant loss of location information. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method, system, device, and storage medium for remote sensing building detection. This invention utilizes edge mask labels generated by an edge detection operator and feature images generated by an edge mask network for training to enhance location information, thereby improving object detection accuracy without increasing computational costs.

[0006] In a first aspect, the present invention provides a method for remote sensing building detection;

[0007] Methods for remote sensing building detection include:

[0008] Construct a training set, which consists of remote sensing images with known remote sensing building detection labels;

[0009] The building detection model is trained using a training set to obtain a trained building detection model, wherein the trained building detection model includes an edge masking network;

[0010] The edge masking network is removed from the trained building detection model to obtain the corrected building detection model.

[0011] The system acquires the remote sensing image to be detected, inputs the image into the modified building detection model, and outputs the detection results of the buildings in the image.

[0012] Secondly, the present invention provides a system for remote sensing building detection;

[0013] Remote sensing building detection systems include:

[0014] The training set construction module is configured to: construct a training set, which is a remote sensing image with known remote sensing building detection labels;

[0015] The training module is configured to train the building detection model using a training set to obtain a trained building detection model, wherein the trained building detection model includes an edge masking network.

[0016] The removal module is configured to remove the edge masking network from the trained building detection model to obtain a corrected building detection model.

[0017] The detection module is configured to: acquire the remote sensing image to be detected, input the remote sensing image to be detected into the modified building detection model, and output the detection results of buildings in the image.

[0018] Thirdly, the present invention also provides an electronic device, comprising:

[0019] Memory, used for non-transitory storage of computer-readable instructions; and

[0020] Processor, for executing the computer-readable instructions,

[0021] When the computer-readable instructions are executed by the processor, they perform the method described in the first aspect above.

[0022] Fourthly, the present invention also provides a storage medium for non-transitory storage of computer-readable instructions, wherein, when the non-transitory computer-readable instructions are executed by a computer, the instructions for the method described in the first aspect are executed.

[0023] Fifthly, the present invention also provides a computer program product, including a computer program that, when run on one or more processors, is used to implement the method described in the first aspect above.

[0024] Compared with the prior art, the beneficial effects of the present invention are:

[0025] This invention utilizes edge mask labels generated by an edge detection operator and feature images generated by an edge mask network for training to enhance location information, thereby improving object detection accuracy without increasing computational cost. Furthermore, it can enhance the learning of location information and improve the accuracy of building detection without increasing model complexity. Attached Figure Description

[0026] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0027] Figure 1 This is a flowchart of the remote sensing building detection method of the present invention;

[0028] Figures 2(a) and 2(b) are schematic diagrams of the internal structure of the remote sensing building detection model of the present invention;

[0029] Figure 3 This is a schematic diagram of the internal structure of the edge masking network of the present invention;

[0030] Figures 4(a) to 4(i) This is the edge mask image generated after filtering the edge detection operator and the real bounding box according to the present invention.

[0031] Figures 5(a) to 5(f) The image shows the test results of this invention. Detailed Implementation

[0032] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0033] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. Furthermore, it should be understood that the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0034] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0035] All data acquisition in this embodiment is carried out in accordance with laws and regulations and with user consent, and the data is used legally.

[0036] Example 1

[0037] This embodiment provides a method for remote sensing building detection;

[0038] like Figure 1 As shown, the method for remote sensing building detection includes:

[0039] S101: Construct a training set, which consists of remote sensing images with known remote sensing building detection labels;

[0040] S102: The building detection model is trained using a training set to obtain a trained building detection model, wherein the trained building detection model includes an edge masking network;

[0041] S103: Remove the edge masking network from the trained building detection model to obtain the corrected building detection model;

[0042] S104: Acquire the remote sensing image to be detected, input the remote sensing image to be detected into the corrected building detection model, and output the detection results of buildings in the image.

[0043] Furthermore, the specific process of constructing the training set includes:

[0044] Obtain remote sensing images with known remote sensing building detection labels, preprocess the remote sensing images, and then process the preprocessed remote sensing images using an edge detection operator to obtain edge mask labels.

[0045] Select labeled optical remote sensing images, with no limit on the length and width of the images. Each batch of images will be preprocessed, the images within the batch will be scaled, and then filled into images of uniform size. Then, 80% of the images will be used to form the training set, and 20% of the images will be used to form the validation set and test set.

[0046] Furthermore, the preprocessing of the remote sensing image includes:

[0047] For each batch of data, the data is horizontally flipped with a 50% probability. Then, the longer side is scaled to an integer multiple of 32 between 800 and 1333 pixels, while the shorter side is scaled to maintain the fixed aspect ratio of the image. If the length or width is less than an integer multiple of 32, it is padded with 0.

[0048] Further, as shown in Figures 2(a) and 2(b), in step S102: the building detection model is trained using a training set to obtain a trained building detection model; wherein, the network structure of the trained building detection model includes:

[0049] ResNet-50 network and Feature Pyramid Networks (FPN);

[0050] The ResNet-50 network includes: a first convolutional block conv_1, a second convolutional block conv_2, a third convolutional block conv_3, a fourth convolutional block conv_4, and a fifth convolutional block conv_5 connected sequentially from bottom to top; the output of the first convolutional block conv_1 outputs feature map c1, the output of the second convolutional block conv_2 outputs feature map c2, the output of the third convolutional block conv_3 outputs feature map c3, the output of the fourth convolutional block conv_4 outputs feature map c4, and the output of the fifth convolutional block conv_5 outputs feature map c5.

[0051] The Feature Pyramid Network (FPN) comprises, from top to bottom, an upsampling unit y1, an adder J1, an upsampling unit y2, an adder J2, an upsampling unit y3, and an adder J3. The input of the upsampling unit y1 is connected to the output of the fifth convolutional block conv_5 through a 1*1 convolutional layer M1. The input of the adder J1 is connected to the output of the fourth convolutional block conv_4 through a 1*1 convolutional layer M2. The input of the adder J2 is connected to the output of the third convolutional block conv_3 through a 1*1 convolutional layer M3. The input of the adder J3 is connected to the output of the second convolutional block conv_2 through a 1*1 convolutional layer M4.

[0052] The output of the 1*1 convolutional layer M1 is connected to the input of the 3*3 convolutional layer L1; the output of adder J1 is connected to the input of the 3*3 convolutional layer L2; the output of adder J2 is connected to the input of the 3*3 convolutional layer L3; and the output of adder J3 is connected to the input of the 3*3 convolutional layer L4.

[0053] Furthermore, the output of the 3*3 convolutional layer L1 is connected to the input of the Region Proposal Network (RPN) and the input of the region feature aggregation module RoIAlign, respectively.

[0054] The output of the 3*3 convolutional layer L2 is connected to the input of the Region Proposal Network (RPN) and the input of the region feature aggregation module RoIAlign, respectively.

[0055] The output of the 3*3 convolutional layer L3 is connected to the input of the Region Proposal Network (RPN) and the input of the region feature aggregation module RoIAlign, respectively.

[0056] The output of the 3*3 convolutional layer L4 is connected to the input of the Region Proposal Network (RPN) and the input of the region feature aggregation module RoIAlign, respectively.

[0057] The output of the 3*3 convolutional layer L4 is connected to the input of the edge masking network, and the output of the edge masking network is used to output the feature image.

[0058] The output of convolutional layer L1 outputs feature map P5, the output of convolutional layer L2 outputs feature map P4, the output of convolutional layer L3 outputs feature map P3, and the output of convolutional layer L4 outputs feature map P2.

[0059] The output of the Region Generation Network (RPN) is connected to the input of the Region Feature Aggregation Module (ROIAlign); the output of the ROIAlign module is connected to the classifier and the regressor, respectively.

[0060] Furthermore, the first convolutional block conv_1 includes a convolutional layer; the second convolutional block conv_2 includes a downsampling layer and three convolutional sub-blocks connected in sequence; the third convolutional block conv_3 includes a downsampling layer and four convolutional sub-blocks connected in sequence; the fourth convolutional block conv_4 includes a downsampling layer and six convolutional sub-blocks connected in sequence; and the fifth convolutional block conv_5 includes a downsampling layer and three convolutional sub-blocks connected in sequence. Each convolutional sub-block includes a 1*1 convolutional layer, a 3*3 convolutional layer, and a 1*1 convolutional layer connected in sequence.

[0061] The input of the first convolutional block conv1 is used to input the remote sensing image. The remote sensing image is processed by an edge detection operator to obtain edge mask labels.

[0062] Furthermore, the working principle of the trained building detection model is as follows:

[0063] The ResNet-50 network is used for feature extraction, the Feature Pyramid Network (FPN) is used for cross-scale feature fusion, and classifiers and regressors are used to accurately predict buildings.

[0064] Furthermore, such as Figure 3 As shown, the edge masking network includes, in sequence: a first upsampling layer, a 3×3 convolutional layer, a ReLU activation function layer, a second upsampling layer, a 3×3 convolutional layer, a ReLU activation function layer, and a convolutional layer.

[0065] Furthermore, such as Figure 3 As shown, the edge masking network works as follows:

[0066] Edge masking networks are used to upsample and constrain the feature maps generated by feature pyramid networks (FPNs).

[0067] The edge masking network enables the entire ResNet-50 network and the Feature Pyramid Network (FPN) to learn obvious edge information during the training phase through pseudo-labels; the pseudo-labels are generated using real bounding boxes and the Laplace edge algorithm.

[0068] Further, in step S102: the building detection model is trained using a training set to obtain a trained building detection model; wherein the training process includes:

[0069] S102-1: Input the preprocessed data into the ResNet-50 network and the Feature Pyramid Network (FPN) for feature extraction, and generate feature maps P2, P3, P4 and P5;

[0070] S102-2: Input the generated feature maps P2, P3, P4 and P5 into the RPN network to generate proposal boxes;

[0071] S102-3: Input feature maps P2, P3, P4 and P5 and proposal boxes into the ROIAlign module, classifier and regressor for training;

[0072] S102-4: Generate an edge mask by filtering the image using edge detection operators and the true bounding box;

[0073] S102-5: Generate feature map A2 from P2 using an edge network;

[0074] S102-6: Training is performed by calculating the mean squared error loss function of feature map A2 and edge mask;

[0075] Calculate the first loss function value between the edge mask label and the processed feature image;

[0076] Calculate the second loss function value of the classifier and the third loss function value of the regressor;

[0077] The total loss function value is obtained by summing the first, second, and third loss function values.

[0078] When the total loss function value no longer decreases, training is stopped, and the trained building detection model is obtained.

[0079] During training, the Laplace edge detection operator is used to initialize the 3×3 convolution kernel. Then, the initialized 3×3 convolution kernel is used to convolve the original image to generate a feature map. Finally, the true bounding boxes are used to filter the backgrounds that do not contain buildings, and only the edge information of the buildings is retained as the edge mask.

[0080] Further, in S103: the edge mask network is removed from the trained building detection model to obtain a modified building detection model; wherein, the difference between the network model of the modified building detection model and the trained building detection model is that the modified building detection model does not have an edge mask network, while the trained building detection model does have an edge mask network.

[0081] Further, step S104: acquiring the remote sensing image to be detected, inputting the remote sensing image to be detected into the corrected building detection model, and outputting the detection results of buildings in the image, specifically includes:

[0082] The remote sensing image to be detected is input into the modified building detection model, and the model's classifier outputs the classification labels of the buildings in the remote sensing image.

[0083] Further, step S104: acquiring the remote sensing image to be detected, inputting the remote sensing image to be detected into the corrected building detection model, and outputting the detection results of buildings in the image, specifically includes:

[0084] The preprocessed data is input into the ResNet-50 network and the Feature Pyramid Network (FPN) for feature extraction, generating feature maps P2, P3, P4, and P5.

[0085] Feature maps P2, P3, P4, and P5 are fed into the classifier and regressor to predict the class and bounding box, and then non-maximum suppression is applied to generate the detection results.

[0086] The detection results are mapped to the corresponding positions in the original image and the predicted category information is labeled.

[0087] This invention utilizes edge mask labels generated by an edge detection operator and feature images generated by an edge mask network for training, thereby enhancing the location information of the building detection model and improving object detection accuracy without increasing computational cost.

[0088] An edge masking network is constructed and trained using its embedding. The edge masking labels generated by the edge detection operator and the feature images generated by the edge masking network are used for training to enhance the location information of the building detection model. This improves the detection capability of the building detection model without increasing the computational cost.

[0089] Simulation test conditions: Server GPU: GeForce RTX2080 Ti, 12G video memory.

[0090] The software platform for the simulation experiment of this invention is: Ubuntu 18.04 system, Python 3.6, PyTorch-GPU 1.19.0;

[0091] The input images used in the simulation experiments of this invention are the publicly available SSDD dataset, DIOR dataset, and 5M-Building dataset.

[0092] The SSDD dataset was released by Wuhan University in 2017. Its images are extracted from RadarSat-2, TerraSAR-X, and Sentinel-1, and it contains 1160 SAR ship images, each with a size of 500×500×3. We used 1000 ship images as the training set and 160 images as the validation set.

[0093] The DIOR dataset was released by Northwestern Polytechnical University in 2019. Its images are extracted from Google Earth, with a size of 800×800×3 pixels, and include 20 classes and a total of 190,288 targets. We used 5,862 remote sensing images as the training set and 5,862 images as the validation set for testing and evaluation.

[0094] The 5M-Building dataset was released by Jinan University in 2019. Its images are from the Gaofen-2 satellite and include nearly 30,000 panchromatic, multispectral, and fused images. We used 6,999 images as the training set and 3,557 images as the validation set.

[0095] Simulation Experiment 1 uses the RetinaNet method from the prior art under the above simulation conditions and obtains detection results. Simulation Experiment 2 uses the FCOS method from the prior art under the above simulation conditions and obtains detection results. Simulation Experiment 3 uses the YOLOF method from the prior art under the above simulation conditions and obtains detection results. Simulation Experiment 4 uses the Sparse R-CNN method from the prior art under the above simulation conditions and obtains detection results. Simulation Experiment 5 uses the Mask R-CNN method from the prior art under the above simulation conditions and obtains detection results. Simulation Experiment 6 uses the Deformable-DETR method from the prior art under the above simulation conditions and obtains detection results. Simulation Experiment 7 uses the Faster R-CNN method from the prior art under the above simulation conditions and obtains detection results. Simulation Experiment 7 introduces an Edge Masking Network (EMN) into the existing Faster R-CNN method and conducts an experiment under the above simulation conditions to obtain detection results.

[0096] To verify the effectiveness of this invention for remote sensing image detection, the detection results were evaluated using the COCO evaluation index. The calculation results are plotted in Table 1, and refer to Tables 1, 2, and 3.

[0097] Analysis of simulation test results:

[0098] According to Table 1, compared with the baseline model Faster R-CNN, the introduction of the edge masking network improved AP50, AP70 and AP by 1.6%, 4.5% and 10%, respectively.

[0099] Table 1

[0100]

[0101] Combining Tables 2 and 3, replacing the backbone network ResNet50 with ResNeXt50, which has stronger feature extraction capabilities, significantly improves the accuracy of the detection model.

[0102] Table 2

[0103]

[0104] Table 3

[0105]

[0106]

[0107] Table 1 is a comparison table of simulation experiment results on the SSDD dataset of the present invention; Table 2 is a comparison table of simulation experiment results on the SSDD dataset of the present invention; Table 3 is a comparison table of simulation experiment results on the SSDD dataset of the present invention.

[0108] Figures 4(a)-4(i) The detection results of the baseline model and the edge mask network method are shown. Figures 4(a) and 4(d) are the input images, Figures 4(b) and 4(c) are the detection results of Faster R-CNN, and images c and f are the detection results of the edge mask network. From the detection results in Figures 4(b) and 4(c), it can be seen that the Faster R-CNN network has more false positives, while ResNext50+EMN, although having some false negatives or missed detections, has better overall detection results than Faster R-CNN. Figures 5(a) to 5(f) The image shows the test results of this invention.

[0109] Example 2

[0110] This embodiment provides a system for remote sensing building detection;

[0111] Remote sensing building detection systems include:

[0112] The training set construction module is configured to: construct a training set, which is a remote sensing image with known remote sensing building detection labels;

[0113] The training module is configured to train the building detection model using a training set to obtain a trained building detection model, wherein the trained building detection model includes an edge masking network.

[0114] The removal module is configured to remove the edge masking network from the trained building detection model to obtain a corrected building detection model.

[0115] The detection module is configured to: acquire the remote sensing image to be detected, input the remote sensing image to be detected into the modified building detection model, and output the detection results of buildings in the image.

[0116] It should be noted that the training set construction module, training module, removal module, and detection module described above correspond to steps S101 to S104 in Embodiment 1. The examples and application scenarios implemented by these modules and their corresponding steps are the same, but they are not limited to the content disclosed in Embodiment 1. It should also be noted that these modules, as part of the system, can be executed in a computer system, such as a set of computer-executable instructions.

[0117] The descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0118] The proposed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative, and the division of modules described above is only a logical functional division. In actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed.

[0119] Example 3

[0120] This embodiment also provides an electronic device, including: one or more processors, one or more memories, and one or more computer programs; wherein, the processor is connected to the memory, and the one or more computer programs are stored in the memory. When the electronic device is running, the processor executes the one or more computer programs stored in the memory to cause the electronic device to perform the method described in Embodiment 1.

[0121] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0122] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.

[0123] In the implementation process, each step of the above method can be completed by the integrated logic circuits in the processor hardware or by instructions in software.

[0124] The method in Embodiment 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.

[0125] Those skilled in the art will recognize that the units and algorithm steps described in connection with the various examples of this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention.

[0126] Example 4

[0127] This embodiment also provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the method described in Embodiment 1.

[0128] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for remote sensing building detection, characterized by: include: Construct a training set, which consists of remote sensing images with known remote sensing building detection labels; The building detection model is trained using a training set to obtain a trained building detection model, wherein the trained building detection model includes an edge masking network; The edge masking network is removed from the trained building detection model to obtain the corrected building detection model. The remote sensing image to be detected is acquired, and the image is input into the modified building detection model. The detection results of buildings in the image are then output. The trained building detection model has the following network structure: ResNet-50 network and Feature Pyramid Network (FPN); The outputs of the 3*3 convolutional layers L1, L2, L3, and L4 are connected to the inputs of the Region Generation Network (RPN) and the Region Feature Aggregation Module (RoI Align), respectively; the output of the 3*3 convolutional layer L4 is connected to the input of the edge masking network, and the output of the edge masking network is used to output the feature image. The output of convolutional layer L1 outputs feature map P5, the output of convolutional layer L2 outputs feature map P4, the output of convolutional layer L3 outputs feature map P3, and the output of convolutional layer L4 outputs feature map P2. The output of the Region Generation Network (RPN) is connected to the input of the Region Feature Aggregation Module (ROI Align); the output of the ROI Align is connected to the classifier and the regressor, respectively.

2. The remote sensing building detection method as described in claim 1, characterized in that, The ResNet-50 network includes: a first convolutional block conv_1, a second convolutional block conv_2, a third convolutional block conv_3, a fourth convolutional block conv_4, and a fifth convolutional block conv_5 connected sequentially from bottom to top; the output of the first convolutional block conv_1 outputs feature map c1, the output of the second convolutional block conv_2 outputs feature map c2, the output of the third convolutional block conv_3 outputs feature map c3, the output of the fourth convolutional block conv_4 outputs feature map c4, and the output of the fifth convolutional block conv_5 outputs feature map c5. The Feature Pyramid Network (FPN) comprises, from top to bottom, an upsampling unit y1, an adder J1, an upsampling unit y2, an adder J2, an upsampling unit y3, and an adder J3. The input of the upsampling unit y1 is connected to the output of the fifth convolutional block conv_5 through a 1*1 convolutional layer M1. The input of the adder J1 is connected to the output of the fourth convolutional block conv_4 through a 1*1 convolutional layer M2. The input of the adder J2 is connected to the output of the third convolutional block conv_3 through a 1*1 convolutional layer M3. The input of the adder J3 is connected to the output of the second convolutional block conv_2 through a 1*1 convolutional layer M4. The output of the 1*1 convolutional layer M1 is connected to the input of the 3*3 convolutional layer L1; the output of adder J1 is connected to the input of the 3*3 convolutional layer L2; the output of adder J2 is connected to the input of the 3*3 convolutional layer L3; and the output of adder J3 is connected to the input of the 3*3 convolutional layer L4.

3. The remote sensing building detection method as described in claim 1, characterized in that, The working principle of the trained building detection model is as follows: The ResNet-50 network is used for feature extraction, the Feature Pyramid Network (FPN) is used for cross-scale feature fusion, and classifiers and regressors are used to accurately predict buildings.

4. The remote sensing building detection method as described in claim 1, characterized in that, The edge masking network includes, in sequence: a first upsampling layer, a 3×3 convolutional layer, a ReLU activation function layer, a second upsampling layer, a 3×3 convolutional layer, a ReLU activation function layer, and a convolutional layer; The edge masking network works as follows: Edge masking networks are used to upsample and constrain the feature maps generated by feature pyramid networks (FPNs). The edge masking network enables the entire ResNet-50 network and the Feature Pyramid Network (FPN) to learn edge information during the training phase through pseudo-labels; the pseudo-labels are generated using real bounding boxes and the Laplace edge algorithm.

5. The remote sensing building detection method as described in claim 1, characterized in that, The building detection model is trained using a training set to obtain the trained building detection model; the training process includes: The preprocessed data is input into the ResNet-50 network and the Feature Pyramid Network (FPN) for feature extraction, generating feature maps P2, P3, P4, and P5. The generated feature maps P2, P3, P4 and P5 are fed into the RPN network to generate proposal boxes; The feature maps P2, P3, P4, and P5, along with the proposal boxes, are input into the ROI Align module, classifier, and regressor for training. The image is processed by edge detection operators and filtering of the true bounding boxes to generate an edge mask; P2 is used to generate feature map A2 through an edge network; Training is performed by calculating the mean squared error loss function of feature map A2 and edge mask; Calculate the first loss function value between the edge mask label and the processed feature image; Calculate the second loss function value of the classifier and the third loss function value of the regressor; The total loss function value is obtained by summing the first, second, and third loss function values. When the total loss function value no longer decreases, training is stopped, and the trained building detection model is obtained.

6. The remote sensing building detection method as described in claim 1, characterized in that, The process involves acquiring a remote sensing image to be detected, inputting the image into a modified building detection model, and outputting the detection results of buildings in the image, specifically including: The preprocessed data is input into the ResNet-50 network and the Feature Pyramid Network (FPN) for feature extraction, generating feature maps P2, P3, P4, and P5. Feature maps P2, P3, P4, and P5 are fed into the classifier and regressor to predict the class and bounding box, and then non-maximum suppression is applied to generate the detection results. The detection results are mapped to the corresponding positions in the original image and the predicted category information is labeled.

7. A remote sensing building detection system, characterized in that, include: The training set construction module is configured to: construct a training set, which is a remote sensing image with known remote sensing building detection labels; The training module is configured to train the building detection model using a training set to obtain a trained building detection model, wherein the trained building detection model includes an edge masking network. The removal module is configured to remove the edge masking network from the trained building detection model to obtain a corrected building detection model. The detection module is configured to: acquire the remote sensing image to be detected, input the remote sensing image to be detected into the modified building detection model, and output the detection results of buildings in the image; The trained building detection model has the following network structure: ResNet-50 network and Feature Pyramid Network (FPN); The outputs of the 3*3 convolutional layers L1, L2, L3, and L4 are connected to the inputs of the Region Generation Network (RPN) and the Region Feature Aggregation Module (RoI Align), respectively; the output of the 3*3 convolutional layer L4 is connected to the input of the edge masking network, and the output of the edge masking network is used to output the feature image. The output of convolutional layer L1 outputs feature map P5, the output of convolutional layer L2 outputs feature map P4, the output of convolutional layer L3 outputs feature map P3, and the output of convolutional layer L4 outputs feature map P2. The output of the Region Generation Network (RPN) is connected to the input of the Region Feature Aggregation Module (ROI Align); the output of the ROI Align is connected to the classifier and the regressor, respectively.

8. An electronic device, characterized in that it comprises: Memory is used to store computer-readable instructions in a non-transitory manner. as well as Processor, for executing the computer-readable instructions, When the computer-readable instructions are executed by the processor, they perform the method described in any one of claims 1-6.

9. A storage medium characterized by being non-transitory. The non-transitory stored computer-readable instructions are executed by a computer, wherein the instructions of the method according to any one of claims 1-6 are executed.

Images