Target detection method for power transmission tower detection in remote sensing image

Abstract

The invention relates to the technical field of power transmission tower detection, in particular to a target detection method for power transmission tower detection in a remote sensing image, and the method comprises the steps: improving a Faster RCNN network model, and obtaining a power transmission tower detection model; acquiring a remote sensing image data set of the power transmission tower, and performing image processing on the remote sensing image data set of the power transmission tower to obtain a sample data set for training a power transmission tower detection model; testing and training the power transmission tower detection model by using the sample data set to obtain a trained power transmission tower detection model; and inputting a remote sensing image of a to-be-detected power transmission tower into the trained power transmission tower detection model to obtain a power transmission tower image center coordinate. According to the method, the defect that the omission ratio of the power transmission tower is high due to the fact that feature information of different scales is ignored and only the feature layer of the minimum scale is detected is overcome, and the method can be used for automatically obtaining the geographic coordinates of the power transmission tower in the target area.

Description

Technical Field

[0001] The present invention relates to the technical field of transmission tower detection, and in particular to a target detection method for transmission tower detection in remote sensing images. Background Art

[0002] Target detection in remote sensing images is one of the current research hotspots in the field of image processing. Transmission towers are important national infrastructure and an important part of the power transmission system. The location information of transmission towers is important geospatial intelligence. However, since transmission towers are small and easily affected by the surrounding environment, there is a problem of low missed detection rate in transmission tower detection.

[0003] Traditional remote sensing image transmission tower detection methods include manual interpretation and traditional machine learning. Manual interpretation relies on the experience and skills of the interpreter and is less efficient. Traditional machine learning methods train machine learning models to identify transmission towers in remote sensing images, which has higher efficiency and accuracy than manual interpretation. However, they are limited by factors such as training samples and quality, and model complexity, and the detection effect of transmission towers in complex backgrounds and with weak targets is poor.

[0004] In order to overcome the defects of traditional remote sensing image transmission tower detection methods, more and more scholars have applied deep learning technology to object detection tasks in remote sensing images and achieved good results. At the same time, remote sensing satellite technology has developed rapidly, and the available high-resolution remote sensing image resources are becoming more and more abundant. This has made it mainstream to automatically extract the geographic information of various types of objects in remote sensing images based on high-resolution remote sensing images and target detection technology, and also provided new possibilities for extracting the geographic location information of transmission towers.

[0005] At present, the target detection algorithms for transmission towers in remote sensing images based on deep learning technology are mainly divided into two categories. One is a single-stage target detection method that converts detection into a regression problem, represented by YOLO (YOU ONLE LOOK ONCE). The original image is divided into grids and regression prediction is performed on the features in each grid. However, YOLO has low detection accuracy, especially for targets with weak samples, there is a high missed detection rate. The other is region-based target detection, such as RCNN (Region with CNN feature), Fast RCNN, and Faster RCNN. The region-based target detection model often ignores feature information of different scales and only detects the feature layer of the smallest scale, which often leads to the loss of small targets. In the detection process of transmission towers, there is also a high missed detection rate. Summary of the invention

[0006] In order to solve the technical problem that the existing transmission tower detection has a high missed detection rate, the purpose of the present invention is to provide a target detection method for transmission tower detection in remote sensing images. The technical solution adopted is as follows:

[0007] An embodiment of the present invention provides a target detection method for detecting transmission towers in remote sensing images, the method comprising the following steps:

[0008] A Faster RCNN network model is constructed, a spatial pyramid pooling module is added to the Faster RCNN network model, and the output result is connected to a multi-scale feature fusion network to obtain a transmission tower detection model;

[0009] Acquire a transmission tower remote sensing image dataset, perform image processing on each original remote sensing image in the transmission tower remote sensing image dataset, and obtain a sample dataset for training a transmission tower detection model;

[0010] Training and testing the transmission tower detection model based on the sample data set to obtain a trained transmission tower detection model;

[0011] The remote sensing image of the transmission tower to be detected is input into the trained transmission tower detection model to obtain the center coordinates of the transmission tower image.

[0012] Furthermore, the steps of constructing the transmission tower detection model include:

[0013] First, add the spatial pyramid pooling module to the deepest feature layer in the Faster RCNN network model;

[0014] Then, the output result of the spatial pyramid pooling module and two adjacent feature layers are added to the feature fusion network;

[0015] Then, the three output fused feature layers are tested respectively, and the fused feature layer with the optimal scale is selected to obtain the transmission tower detection model.

[0016] Furthermore, the acquiring of the transmission tower remote sensing image dataset and performing image processing on each original remote sensing image in the transmission tower remote sensing image dataset include:

[0017] Acquire original remote sensing images of several target areas of the transmission tower at target resolutions to form a transmission tower remote sensing image dataset;

[0018] Performing image cropping on each of the original remote sensing images to obtain a first remote sensing image that meets the requirements of the transmission tower detection model;

[0019] Marking each of the first remote sensing images with a rectangular frame, recording the coordinates of the lower left corner of the original remote sensing image and the position information of the first remote sensing image, and obtaining each of the second remote sensing images;

[0020] The labeled brick file of each of the second remote sensing images is converted to obtain a sample data set that satisfies the training and testing of the transmission tower detection model.

[0021] Furthermore, after obtaining the sample data set, the method further includes:

[0022] The sample data set is divided into a training set and a test set, the training set is used to train the transmission tower detection model, and the test set is used to evaluate the performance of the trained transmission tower detection model.

[0023] Furthermore, the step of inputting the remote sensing image of the transmission tower to be detected into the trained transmission tower detection model to obtain the center coordinates of the transmission tower image includes:

[0024] Calculate the initial geographic coordinates of the transmission tower to be detected according to the coordinates of the lower left corner of the remote sensing image of the transmission tower to be detected, the position information of the first remote sensing image of the remote sensing image, the target resolution, and the spatial coordinates of the transmission tower image to be detected identified by the transmission tower detection model;

[0025] Calculate the geographic coordinate error according to the height of the transmission tower to be detected and the satellite shooting angle of the remote sensing image of the transmission tower to be detected;

[0026] According to the geographic coordinate error and the azimuth observation angle of the satellite, the initial geographic coordinates of the transmission tower to be detected are corrected to obtain the center coordinates of the transmission tower image.

[0027] Furthermore, the calculation formula of the initial geographic coordinates of the transmission tower to be detected is:

[0028] X i =X0+(a i -1)*ΔX+k*x;Y i =Y0+(b i -1)*ΔY+k*y; where [X i ,Y i ] represents the initial geographic coordinates of the transmission tower to be detected, [X0, Y0] represents the coordinates of the lower left corner of the remote sensing image of the transmission tower to be detected, [ΔX, ΔY] represents the length and width of the first remote sensing image of the remote sensing image of the transmission tower to be detected, k represents the target resolution, and (x, y) represents the spatial coordinates of the transmission tower image to be detected identified by the transmission tower detection model.

[0029] Furthermore, the calculating of the geographic coordinate error according to the height of the transmission tower to be detected and the satellite shooting angle of the remote sensing image of the transmission tower to be detected includes:

[0030] Where, e represents the geographic coordinate error, H represents the height of the transmission tower to be detected, A represents the satellite shooting angle of the remote sensing image of the transmission tower to be detected, and sin represents the sine function.

[0031] Furthermore, the initial geographic coordinates of the transmission tower to be detected are corrected according to the geographic coordinate error and the azimuth observation angle of the satellite to obtain the center coordinates of the transmission tower image, including:

[0032] x i =X i +e×sinAz;y i =Y i +e×cosAz; where [x i ,y i ] represents the center coordinates of the transmission tower image, [X i ,Y i ] represents the initial geographic coordinates of the transmission tower to be detected, e represents the geographic coordinate error, Az represents the azimuth observation angle of the satellite, cos represents the cosine function, and sin represents the sine function.

[0033] The present invention has the following beneficial effects:

[0034] The present invention provides a target detection method for power transmission tower detection in remote sensing images. The method adopts the Faster RCNN network structure as the basic framework for target detection, and integrates the spatial pyramid pooling structure and the feature fusion network to reduce the missed detection rate of power transmission towers in remote sensing images and automatically output the geographical data of the detected power transmission towers, which has high effectiveness and practicality. The spatial pyramid pooling and feature fusion network are used to fuse multi-scale feature information, enhance the model's detection ability for weak targets, and effectively reduce the model's false detection rate and missed detection rate in remote sensing image power transmission tower detection; directly input the remote sensing image of the power transmission tower to be detected into the trained power transmission tower detection model to realize target detection, and can automatically output the geographical coordinates of the detected power transmission tower, including error analysis and correction, simplify the subsequent processing links, and improve application efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0036] Figure 1 A flowchart of a target detection method for detecting transmission towers in remote sensing images according to an embodiment of the present invention;

[0037] Figure 2 This is a schematic diagram of the existing Faster RCNN network model structure;

[0038] Figure 3 A schematic diagram of the structure of a spatial pyramid pooling module in one embodiment of the present invention;

[0039] Figure 4 A schematic diagram of a multi-scale feature fusion network in one embodiment of the present invention;

[0040] Figure 5 It is a schematic diagram of the structure of a transmission tower detection model in one embodiment of the present invention;

[0041] Figure 6 A schematic diagram of an image coordinate extraction mechanism in one embodiment of the present invention;

[0042] Figure 7 A schematic diagram of transmission tower coordinate error calculation in one embodiment of the present invention;

[0043] Figure 8 is a schematic diagram of a first remote sensing image in one embodiment of the present invention;

[0044] Fig. 9 Schematic diagram of calculation of transmission tower image center coordinates in one embodiment of the present invention. DETAILED DESCRIPTION

[0045] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the technical solutions proposed by the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0046] Unless defined otherwise, 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 belongs.

[0047] Regarding remote sensing image processing, typical object detection algorithms in object detection research include Faster RCNN, YOLO, and RetinaNet algorithms. Among them, the YOLO algorithm is a real-time target detection algorithm, suitable for application scenarios that require high speed, but the detection accuracy is low, and the detection speed requirement for objects such as transmission towers in remote sensing images is not high; the RetinaNet algorithm is often used to deal with the problem of class imbalance in images, and has good accuracy for detecting weak targets, but is very sensitive to complex environments and has poor anti-interference ability in the face of complex remote sensing image backgrounds; the Faster RCNN algorithm is a region-based target detection algorithm that maintains good detection performance for complex remote sensing images, but the computational complexity of the model is high, and the detection effect of weak targets is not ideal.

[0048] The above-mentioned prior art ignores feature information of different scales, resulting in reduced utilization of feature information, and for the weak features of transmission tower targets, when the scale of the target detection layer is small, it may cause small targets to be lost during the transmission tower detection process; and the prior art does not fully consider the influence of satellite observation angles and transmission tower characteristics on coordinate accuracy, which may result in insufficient accuracy of the acquired transmission tower coordinates, affecting subsequent applications.

[0049] In order to overcome the above-mentioned defects and deficiencies in the prior art, an embodiment of the present invention provides a target detection method for detecting transmission towers in remote sensing images, such as Figure 1 As shown, the following steps are included:

[0050] S1, build the Faster RCNN network model, add the spatial pyramid pooling module to the Faster RCNN network model, and connect the output result to the multi-scale feature fusion network to obtain the transmission tower detection model.

[0051] Here, the transmission tower detection model is a model that detects the location coordinates of the transmission tower through remote sensing images. The transmission tower detection model obtained by improving the Faster RCNN network model can improve the detection performance of weak targets. The schematic diagram of the Faster RCNN network model structure is shown in Figure 2 shown.

[0052] It should be noted that the Faster R-CNN network model can fully utilize its powerful target detection capabilities, multi-task learning, end-to-end training, efficient candidate region generation, and adaptability to complex environments to provide a high-precision and high-efficiency transmission tower detection solution. In addition, the flexibility and scalability of Faster R-CNN enables it to be optimized according to different application scenarios to meet different needs.

[0053] The above step S1 can be implemented by the following steps:

[0054] The first step is to add a spatial pyramid pooling module to the deepest feature layer in the Faster RCNN network model.

[0055] In this embodiment, in the spatial pyramid pooling structure, a cross-stage connection is used to replace the 1×1 pooling layer in the pooling layer, and the optimal pooling layer parameters are obtained through experiments [3, 5, 7]. The schematic diagram of the spatial pyramid pooling module structure is shown in Figure 3 shown.

[0056] It should be noted that the transmission tower module contains a spatial pyramid pooling module that can process features of different scales. Because the traditional convolutional neural network generally requires the input image to be of the same size, the spatial pyramid pooling module can effectively process targets of different sizes and scales in the input image by performing pooling at different scales. This is of great significance for transmission tower detection, especially for transmission tower targets at different distances, perspectives or angles. In addition, the spatial pyramid pooling module extracts more representative features through pooling operations, thereby improving the robustness of the model to noise and interference, which has great advantages for transmission tower detection in the presence of cluttered backgrounds, and at different perspectives and resolutions.

[0057] In the second step, the output result of the spatial pyramid pooling module is added to the feature fusion network together with the two adjacent feature layers.

[0058] In this embodiment, the spatial pyramid pooling structure processes the minimum scale feature layer, and sends the output result and the feature layers of two adjacent scales into the feature fusion network, which helps to improve the multi-scale adaptability, positioning accuracy, robustness and expression ability of the transmission tower detection model. This fusion method enhances the model's comprehensive processing capabilities for information of different scales and levels, enabling it to more accurately identify and locate transmission tower targets in complex backgrounds, different perspectives and occlusion conditions. In addition, feature fusion also improves training efficiency and enhances the model's anti-interference ability, which is of great significance for transmission tower detection in practical applications. Among them, the schematic diagram of the multi-scale feature fusion network is shown in the figure. Figure 4 shown.

[0059] In the third step, the three fused feature layers of the output are tested separately, and the fused feature layer with the optimal scale is selected to obtain the transmission tower detection model.

[0060] In this embodiment, by testing the three output layers of the feature fusion network, it is found that the middle layer performs best in the transmission tower detection work, and then the feature layer with the most appropriate scale is selected for subsequent detection tasks. Figure 5 shown.

[0061] It should be noted that the fusion of feature layers of different scales helps the model capture more detailed information at different levels. Usually, the low-scale layer captures a large range of global information, while the high-scale layer captures local details and high-frequency information. By selecting the fusion feature layer of the optimal scale, the model's detection accuracy of transmission towers can be effectively improved, reducing missed detections and false detections. In practical applications, by selecting the fusion feature layer of the optimal scale, the model only needs to process the most representative and effective scale information, thereby reducing unnecessary consumption of computing resources and improving the computational efficiency of the model. In addition, transmission towers usually have complex result features and different perspectives. After selecting the fusion feature layer of the optimal scale, the model can more accurately locate the transmission tower and better distinguish the difference between the transmission tower and the background, thereby improving the overall recognition effect.

[0062] So far, this embodiment has obtained a model that can accurately detect the location information of the transmission tower, that is, the transmission tower detection model.

[0063] S2, obtaining a transmission tower remote sensing image dataset, performing image processing on each original remote sensing image in the transmission tower remote sensing image dataset, and obtaining a sample dataset for training a transmission tower detection model.

[0064] Here, image processing may include cropping, numbering and labeling. The purpose of image processing is to enable the image processing of the transmission tower remote sensing image dataset to be used for training the transmission tower detection model. The sample dataset refers to an image dataset that can be used for training the transmission tower detection model after image processing. Generally, the sample dataset can be divided into two parts, including a training set and a test set. In this embodiment, the training set and the test set are divided according to a ratio of 4:1, wherein the training set contains 837 sample images and the test set contains 209 sample images.

[0065] The above step S2 can be implemented by the following steps:

[0066] The first step is to obtain the original remote sensing images of several target areas of the transmission tower with target resolution to form a transmission tower remote sensing image dataset.

[0067] In this embodiment, 1046 original remote sensing images with a target resolution of 0.5m are obtained for the target area, and all the obtained original remote sensing images constitute a transmission tower remote sensing image dataset. There are many data sources for obtaining original remote sensing images, such as satellite remote sensing (such as Landsat, Sentinel, WorldView, etc.), aerial remote sensing (such as aircraft photography), drone remote sensing (such as drone images), etc. They have different resolutions, coverage and application scenarios, which are not specifically limited here; the target resolution of the remote sensing image needs to be greater than or equal to 0.5.

[0068] In the second step, each original remote sensing image is cropped to obtain the first remote sensing image that meets the requirements of the transmission tower detection model.

[0069] In this embodiment, the length and width of the cropped image should be multiples of 32 to meet the model requirements, and the cropped original remote sensing image that meets the model requirements is used as the first remote sensing image. Figure 8 shown.

[0070] The third step is to mark each first remote sensing image with a rectangular frame, record the lower left corner coordinates of the original remote sensing image and the position information of the first remote sensing image, and obtain each second remote sensing image.

[0071] In this embodiment, in the labeling process, the image where the transmission tower is located is first found according to prior knowledge, and then the remote sensing image of the transmission tower is labeled with a rectangular frame using Lableimg to generate a labeling file in xml format, and the labeling file should have the same name as the original image file; then, by parsing the xml file, the labeling template category information and the rectangular frame position information can be obtained. The lower left corner coordinates of the original remote sensing image and the position information of the first remote sensing image are recorded here to determine the basis for the subsequent conversion of the image space coordinates to the geographic coordinates, and the first remote sensing image that has been labeled is used as the second remote sensing image. Among them, the process of labeling by Lableimg is a prior art, which is not within the scope of protection of the present invention and will not be elaborated here.

[0072] The fourth step is to convert the labeled brick files of each second remote sensing image to obtain a sample data set that meets the training and testing requirements of the transmission tower detection model.

[0073] In this embodiment, the annotation file in XML format is converted into a sample data set that satisfies the training and testing of the Faster RCNN network through a Python script file. The process of implementing annotation file conversion with a Python script file is a prior art and is not within the scope of protection of the present invention, and will not be elaborated on in detail here.

[0074] So far, this embodiment has obtained a sample data set that satisfies the training and testing of the transmission tower detection model.

[0075] S3, training and testing the transmission tower detection model based on the sample data set to obtain a trained transmission tower detection model.

[0076] In this embodiment, the sample data set can be divided into a training set and a test set. The training set is used to train the transmission tower detection model, and the test set is used to evaluate the performance of the trained transmission tower detection model. Specifically, the transmission tower detection model is trained by the training set to obtain a training result, and then the training result is tested by the test set to obtain a trained network model that meets the performance requirements, that is, a trained transmission tower detection model. Among them, the training and testing process of the transmission tower detection model is a prior art, which is not within the protection scope of the present invention and will not be elaborated here.

[0077] It should be noted that the model training parameter settings in the transmission tower detection model training process include: the input image resolution size is 640*640, the iteration batch size is 2, the iteration batch is 240 times, the initial learning rate is 0.0001, the candidate box size parameter is [4, 6, 8], the candidate box aspect ratio parameter is [0.5, 1, 1.5], corresponding to 9 candidate boxes for each feature point, the impulse unit momentum is set to 0.9, there is no weight decay, the SDG (Stochastic Grident Descent) stochastic gradient descent method is used to optimize the model parameters, and the cos learning rate descent method is used to adjust the learning rate; and the transmission tower detection model is trained by training and testing. It can be tested once every certain iteration batch and the weights can be saved to evaluate the model performance at this time. The parameters are set to 20 batches here. Since the evaluation takes a lot of time, frequent evaluation is not recommended; finally, the best performance that meets the requirements after the test is used as the model used for transmission tower detection. In practical applications, the cropped image is input into the model to automatically output the detected transmission tower image coordinates.

[0078] At this point, this embodiment has obtained a transmission tower detection model that has completed training.

[0079] S4, inputting the remote sensing image of the transmission tower to be detected into the trained transmission tower detection model to obtain the center coordinates of the transmission tower image.

[0080] In this embodiment, the remote sensing image of the transmission tower to be detected is an image that meets the performance requirements of the transmission tower detection model. The process of inputting the remote sensing image into the trained transmission tower detection model to obtain the center coordinates of the transmission tower image may include the following steps:

[0081] The first step is to calculate the initial geographic coordinates of the transmission tower to be detected based on the coordinates of the lower left corner of the remote sensing image of the transmission tower to be detected, the position information of the first remote sensing image of the remote sensing image, the target resolution, and the spatial coordinates of the transmission tower image to be detected identified by the transmission tower detection model.

[0082] In this embodiment, the initial geographic coordinates refer to the geographic coordinates without considering the final transmission tower geographic coordinate errors, that is, the center point of the target detection model detection image should be the midpoint of the transmission tower after tilting, and it only builds a mapping relationship based on the original remote sensing image parameters recorded in step S2 to calculate the geographic coordinates of the transmission tower. Figure 6 shown.

[0083] As an example, the calculation formula for the initial geographic coordinates of the transmission tower to be detected may be:

[0084] X i =X0+(a i -1)*ΔX+k*x;Y i =Y0+(b i -1)*ΔY+k*y; where [X i ,Y i ] represents the initial geographic coordinates of the transmission tower to be detected, [X0, Y0] represents the coordinates of the lower left corner of the remote sensing image of the transmission tower to be detected, [ΔX, ΔY] represents the length and width of the first remote sensing image of the remote sensing image of the transmission tower to be detected, k represents the target resolution, and (x, y) represents the spatial coordinates of the transmission tower image to be detected identified by the transmission tower detection model.

[0085] In the second step, the geographic coordinate error is calculated according to the height of the transmission tower to be detected and the satellite shooting angle of the remote sensing image of the transmission tower to be detected.

[0086] In this embodiment, the transmission tower in the remote sensing image is affected by the satellite shooting angle, resulting in a geographic coordinate error. Therefore, the geographic coordinate error is related to the height of the transmission tower and the satellite shooting angle of the remote sensing image. That is, the geographic coordinate error can be determined by the height of the transmission tower to be detected and the satellite shooting angle. The schematic diagram of the transmission tower coordinate error calculation is shown in FIG. Figure 7 Among them, the height of the medium voltage transmission tower can be 10 to 25 meters, the height of the high voltage transmission tower can be 30 to 60 meters, and the height of the ultra-high voltage transmission tower can be 100 to 150 meters. The satellite shooting angle of the remote sensing image can usually be 30° to 60°.

[0087] As an example, the calculation formula for geographic coordinate error can be:

[0088] Where, e represents the geographic coordinate error, H represents the height of the transmission tower to be detected, that is, the height of the ground transmission tower, A represents the satellite shooting angle of the remote sensing image of the transmission tower to be detected, that is, the pitch observation angle of the remote sensing image satellite relative to the ground, and sin represents the sine function.

[0089] The third step is to correct the initial geographic coordinates of the transmission tower to be detected according to the geographic coordinate error and the azimuth observation angle of the satellite to obtain the coordinates of the center of the transmission tower image.

[0090] After obtaining the geographic coordinate error and initial geographic coordinates, the initial geographic coordinates of the transmission tower to be detected are corrected in combination with the satellite's azimuth observation angle to obtain the transmission tower image center coordinates. The transmission tower image center coordinate calculation diagram is shown in the figure: Fig. 9 shown.

[0091] As an example, the calculation formula for the center coordinates of the transmission tower image can be:

[0092] x i =X i +e×sinAz;y i =Y i +e×cosAz; where [x i ,y i ] represents the center coordinates of the transmission tower image, [X i ,Y i ] represents the initial geographic coordinates of the transmission tower to be detected, e represents the geographic coordinate error, Az represents the azimuth observation angle of the satellite, cos represents the cosine function, and sin represents the sine function.

[0093] It should be noted that the transmission tower image center coordinates in the model training process are consistent with the implementation process of step S4, both of which predict and correct errors through remote sensing image satellite imaging features and transmission tower feature analysis to improve application efficiency.

[0094] In summary, a target detection method for transmission tower detection in remote sensing images uses the Faster RCNN network structure as the basic framework for target detection, and integrates the spatial pyramid pooling structure and feature fusion network to reduce the missed detection rate of transmission towers in remote sensing images and automatically output the geographic data of the detected transmission towers, which has high effectiveness and practicality. Using spatial pyramid pooling and feature fusion network, multi-scale feature information is integrated to enhance the model's detection ability for weak targets, effectively reducing the model's false detection rate and missed detection rate in remote sensing image transmission tower detection; directly inputting the remote sensing image of the transmission tower to be detected into the trained transmission tower detection model to realize target detection, it can automatically output the geographic coordinates of the detected transmission tower, including error analysis and correction, simplifying subsequent processing links, and improving application efficiency.

[0095] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A target detection method for transmission tower detection in remote sensing images, characterized in that: The following steps are involved: A Faster RCNN network model is constructed, a spatial pyramid pooling module is added to the Faster RCNN network model, and the output result is connected to a multi-scale feature fusion network to obtain a transmission tower detection model; Acquire a transmission tower remote sensing image dataset, perform image processing on each original remote sensing image in the transmission tower remote sensing image dataset, and obtain a sample dataset for training a transmission tower detection model; Training and testing the transmission tower detection model based on the sample data set to obtain a trained transmission tower detection model; The remote sensing image of the transmission tower to be detected is input into the trained transmission tower detection model to obtain the center coordinates of the transmission tower image.

2. The target detection method for transmission tower detection in remote sensing images according to claim 1, characterized in that: The steps of constructing the transmission tower detection model include: First, add the spatial pyramid pooling module to the deepest feature layer in the Faster RCNN network model; Then, the output result of the spatial pyramid pooling module and two adjacent feature layers are added to the feature fusion network; Then, the three output fused feature layers are tested respectively, and the fused feature layer with the optimal scale is selected to obtain the transmission tower detection model.

3. The target detection method for transmission tower detection in remote sensing images according to claim 1, characterized in that: The step of acquiring a transmission tower remote sensing image dataset and performing image processing on each original remote sensing image in the transmission tower remote sensing image dataset includes: Acquire original remote sensing images of several target areas of the transmission tower at target resolutions to form a transmission tower remote sensing image dataset; Performing image cropping on each of the original remote sensing images to obtain a first remote sensing image that meets the requirements of the transmission tower detection model; Marking each of the first remote sensing images with a rectangular frame, recording the coordinates of the lower left corner of the original remote sensing image and the position information of the first remote sensing image, and obtaining each of the second remote sensing images; The labeled brick file of each of the second remote sensing images is converted to obtain a sample data set that satisfies the training and testing of the transmission tower detection model.

4. The target detection method for transmission tower detection in remote sensing images according to claim 3, characterized in that: After obtaining the sample data set, the method further includes: The sample data set is divided into a training set and a test set, the training set is used to train the transmission tower detection model, and the test set is used to evaluate the performance of the trained transmission tower detection model.

5. The target detection method for transmission tower detection in remote sensing images according to claim 3, characterized in that: The step of inputting the remote sensing image of the transmission tower to be detected into the trained transmission tower detection model to obtain the center coordinates of the transmission tower image includes: Calculate the initial geographic coordinates of the transmission tower to be detected according to the coordinates of the lower left corner of the remote sensing image of the transmission tower to be detected, the position information of the first remote sensing image of the remote sensing image, the target resolution, and the spatial coordinates of the transmission tower image to be detected identified by the transmission tower detection model; Calculate the geographic coordinate error according to the height of the transmission tower to be detected and the satellite shooting angle of the remote sensing image of the transmission tower to be detected; According to the geographic coordinate error and the azimuth observation angle of the satellite, the initial geographic coordinates of the transmission tower to be detected are corrected to obtain the center coordinates of the transmission tower image.

6. The target detection method for transmission tower detection in remote sensing images according to claim 5, characterized in that: The calculation formula of the initial geographic coordinates of the transmission tower to be detected is: X i =X0+(a i -1)*ΔX+k*x;Y i =Y0+(b i -1)*ΔY+k*y; where [X i ,Y i ] represents the initial geographic coordinates of the transmission tower to be detected, [X0, Y0] represents the coordinates of the lower left corner of the remote sensing image of the transmission tower to be detected, [ΔX, ΔY] represents the length and width of the first remote sensing image of the remote sensing image of the transmission tower to be detected, k represents the target resolution, and (x, y) represents the spatial coordinates of the transmission tower image to be detected identified by the transmission tower detection model.

7. The target detection method for transmission tower detection in remote sensing images according to claim 6, characterized in that: The step of calculating the geographic coordinate error according to the height of the transmission tower to be detected and the satellite shooting angle of the remote sensing image of the transmission tower to be detected includes: Where, e represents the geographic coordinate error, H represents the height of the transmission tower to be detected, A represents the satellite shooting angle of the remote sensing image of the transmission tower to be detected, and sin represents the sine function.

8. The target detection method for transmission tower detection in remote sensing images according to claim 7, characterized in that: The method of correcting the initial geographic coordinates of the transmission tower to be detected according to the geographic coordinate error and the azimuth observation angle of the satellite to obtain the center coordinates of the transmission tower image includes: x i =X i +e×sinAz;y i =Y i +e×cosAz; where [x i ,y i ] represents the center coordinates of the transmission tower image, [X i ,Y i ] represents the initial geographic coordinates of the transmission tower to be detected, e represents the geographic coordinate error, Az represents the azimuth observation angle of the satellite, cos represents the cosine function, and sin represents the sine function.

Images