Remote sensing target detection method and device based on evidence learning, equipment and medium

Abstract

The invention provides a remote sensing target detection method and device based on evidence learning, equipment and a medium. Relates to the technical field of remote sensing image processing. The method comprises the steps of preprocessing a remote sensing image data set, inputting the preprocessed remote sensing image data set into a basic target detection model, and performing model training to obtain a basic model; adding an evidence learning head to a detection head part of the basic model, calculating evidence values by using a function, and calculating the sum of the evidence values of different categories of each sample to obtain total intensity; a # imgabs0 # function is used for defining evidence learning loss, and evaluation of classification confidence is achieved by quantifying classification accuracy and evidence strength of a basic model for different categories; and according to the classification loss, the regression loss and the evidence learning loss of the basic model, retraining the basic target detection model to obtain a target detection model. According to the method and the device, quantification of detection uncertainty is realized by embedding the evidence learning module, so that the credibility of a detection result is higher.

Description

Technical Field

[0001] The present application relates to the technical field of remote sensing image processing, and in particular to a remote sensing target detection method, device, equipment and medium based on evidence learning. Background Art

[0002] With the rapid development of earth observation technology, remote sensing satellites have achieved significant improvements in spatial, temporal and spectral resolution, making a large number of high-resolution remote sensing images available as resources. As the core task of automatic interpretation of remote sensing images, remote sensing target detection has played an important role in many fields such as environmental monitoring, land use, urban planning, and disaster assessment. Its main purpose is to identify and locate specific targets (such as structured objects such as aircraft, vehicles, and ships) from images and return the target category and location to support automated remote sensing image interpretation.

[0003] Remote sensing images have the characteristics of bird's-eye view, large scale changes, and diverse target postures, which makes the target detection task face a series of unique technical challenges in remote sensing scenes. In recent years, deep learning technology has demonstrated a strong feature representation capability, which has brought significant improvements to the performance of remote sensing target detection, especially in multi-scale, rotation detection, weak feature response, small target detection and limited supervision detection. However, the complexity and uncertainty of remote sensing scenes still make traditional deep learning models have many bottlenecks when processing remote sensing data.

[0004] Among the existing remote sensing target detection methods, deep learning-based detection models have made significant progress in feature extraction capabilities and detection accuracy. However, these methods still have significant defects in quantifying the reliability of detection results. Remote sensing images have unique perspectives, complex ground features, and especially the diversity of scales and orientations of geographic objects, which makes it difficult for models to accurately identify targets in unfamiliar scenes and difficult to measure the reliability of detection results.

[0005] Defects and shortcomings of the existing technology:

[0006] 1. Inadequate uncertainty handling: Traditional remote sensing target detection models cannot directly quantify detection uncertainty, especially in out-of-distribution scenarios or uncommon target types, and the detection results are unreliable. This shortcoming brings great risks in practical applications. For example, in new terrain or changing environments, the model may misreport or miss targets, affecting the credibility of the detection results.

[0007] 2. Lack of robustness: Existing technologies usually rely on a large amount of labeled data for training. When the quality of the labels is uneven, the prediction results of the model may be biased, especially in scenes with weak features and small targets that are difficult to detect. These methods often perform poorly and cannot provide stable detection results. The lack of a mechanism to handle uncertainty makes these models less adaptable to abnormal or complex backgrounds. Summary of the invention

[0008] This application provides a remote sensing target detection method, device, equipment and medium based on evidence learning, aiming to improve the detection reliability and adaptability of remote sensing image target detection models in complex scenes. This application achieves quantification of detection uncertainty by embedding an evidence learning module, thereby making the detection results more credible.

[0009] In a first aspect, the present application provides a remote sensing target detection method based on evidence learning, comprising:

[0010] The remote sensing image dataset is preprocessed and input into the basic target detection model, and the classification loss is used With the regression loss L bbox Perform model training to obtain the basic model W;

[0011] Add an evidence learning head to the detection head of the basic model W and use the function F α Calculate the evidence value to convert the network output into evidence representation and restrict the input feature value range to The total strength is obtained by calculating the sum of the evidence values ​​of different categories of each sample, where is the minimum value of the restricted eigenvalue, y is the input eigenvalue, and n is the maximum value of the restricted eigenvalue;

[0012] use The function defines the evidence learning loss, which evaluates the classification confidence by quantifying the classification accuracy and evidence strength of the basic model for different categories.

[0013] Classification loss according to the base model Regression Loss and the evidence learning loss The basic target detection model is retrained to obtain a target detection model; the target detection model is used to realize target recognition in remote sensing images.

[0014] In one possible design, the remote sensing image dataset is preprocessed and then fed into the basic target detection model, and the classification loss is used. With the regression loss L bbox Perform model training to obtain the basic model W, including:

[0015] Preprocess the remote sensing images of the remote sensing image dataset. Separate the remote sensing images based on the overlapping pixel distance between the two segmented images. The resolution of the segmented images is subsize = 1024 * 1024. Use the scale factors rates = 0.5, 1.0, 1.5 to obtain multi-scale remote sensing images as the preprocessing dataset.

[0016] The labels of the remote sensing images are organized into the format of (x1, y1, x2, y2, x3, y3, x4, y4, n, d), where (x1, y1), (x2, y2), (x3, y3), and (x4, y4) represent the horizontal and vertical coordinates of the four points of the target frame, n represents the number of target categories, and d represents the difficulty of distinguishing the target;

[0017] The preprocessed dataset and the labels of the corresponding remote sensing images are input into the basic target detection model, and the classification loss is used based on the set learning rate. Regression Loss Train the model and obtain the trained basic model W.

[0018] In one possible design, an evidence learning head is added to the detection head of the base model W, using the function F α Calculate the evidence value to convert the network output into evidence representation and restrict the input feature value range to The total strength is calculated by summing up the evidence values ​​of different categories for each sample, including:

[0019] Add an evidence learning head to the detection head of the basic model W and use the function F α Calculate the evidence value, the category score output by the basic model for the i-th sample for the j-th category is y ij , apply an exponential function to the class score and convert it into evidence value α ij , the evidence value represents the confidence of each category, and the calculation process is expressed as:

[0020] F α =log(y ij +1)=α ij (1)

[0021] Limit the range of evidence values ​​to is the maximum value of the evidence value, and the calculation process is expressed as:

[0022]

[0023] In the formula, clamp represents a numerical function in the return range, which limits the increasing, decreasing or randomly changing values ​​to a range of values;

[0024] The total evidence strength of the i-th sample is calculated as S by the following formulai :

[0025]

[0026] In the formula, α ij represents the evidence of the i-th sample for the j-th category, and n represents the number of target categories.

[0027] In one possible design, the function:

[0028] The total evidence strength S of each sample in all categories is calculated based on the following formula i The evidence value α predicted by each class sample ij The error between:

[0029]

[0030] Determine the encoding of the sample's true label based on the following formula

[0031]

[0032] According to the encoding of the sample's true label Calculated Function, expressed as:

[0033]

[0034] In one possible design, the classification loss of the base model Regression Loss and the evidence learning loss Retrain the basic target detection model to obtain the target detection model, including:

[0035] Calculate the classification loss based on the output of the base model The loss function measures the classification accuracy of the model in each category, and the calculation process is:

[0036]

[0037] In the formula, Represents the probability that the i-th sample belongs to the j-th category;

[0038] Calculating regression loss Include Smoothing loss and bounding box regression loss The regression loss is used to measure the accuracy of the model in predicting the bounding box position and angle. It is expressed as:

[0039]

[0040] Where, t i,j is the target regression value of the i-th sample, is the predicted value of the i-th sample, K is the number of regression target dimensions, N is the total number of samples, and Smooth is the Smooth smoothing function;

[0041] Classification loss based on the base model Regression Loss and Evidence Learning The loss is obtained, and the total loss function is obtained;

[0042] The model is retrained based on the total loss function to obtain the model parameters W' based on evidence learning, and the model parameters W' are used to configure the basic target detection model to obtain the target detection model.

[0043] In one possible design, the classification loss based on the base model Regression Loss and the evidence learning loss The total loss function is obtained by the following formula;

[0044]

[0045] In the formula, represents the total loss, λ cls , reg and λ edl Respectively represent the corresponding classification loss Regression Loss and the evidence learning loss The weight of .

[0046] In a second aspect, the present application provides a remote sensing target detection device based on evidence learning, the device comprising:

[0047] The first training module is configured to input the remote sensing image dataset into the basic target detection model after preprocessing, and use the classification loss With the regression loss L bbox Perform model training to obtain the basic model W;

[0048] The total strength calculation module is configured to add an evidence learning head to the detection head part of the basic model W, using the function F α Calculate the evidence value to convert the network output into evidence representation and restrict the input feature value range to The total strength is obtained by calculating the sum of the evidence values ​​of different categories of each sample, where is the minimum value of the evidence value, y is the input feature value, and n is the maximum value of the restricted feature value;

[0049] The evidence learning loss determination module is configured to use The function defines the evidence learning loss, which evaluates the classification confidence by quantifying the classification accuracy and evidence strength of the basic model for different categories.

[0050] The second training module is configured to be based on the classification loss of the base model Regression Loss and the evidence learning loss The basic target detection model is retrained to obtain a target detection model; the target detection model is used to realize target recognition in remote sensing images.

[0051] In a third aspect, an embodiment of the present application provides an electronic device comprising: at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the remote sensing target detection method based on evidence learning as described in the first aspect and various possible designs of the first aspect.

[0052] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer execution instructions are stored. When a processor executes the computer execution instructions, the remote sensing target detection method based on evidence learning as described in the first aspect and various possible designs of the first aspect is implemented.

[0053] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the remote sensing target detection method based on evidence learning as described in the first aspect and various possible designs of the first aspect.

[0054] The remote sensing target detection method, device, equipment and medium based on evidence learning provided in this application can effectively quantify detection uncertainty, significantly improve the reliability and adaptability of the model in different scenarios, and have stronger robustness. The main advantages are summarized as follows:

[0055] 1. Quantify uncertainty and improve detection reliability: This application introduces an evidence learning framework to quantify the confidence of the detection results, so that the detection model not only outputs the target category and location, but also provides a reliability estimate for each detection result. Compared with traditional methods, this application can dynamically adjust the credibility of the output results in unfamiliar scenes or out-of-distribution data. This significantly reduces the probability of false detection and missed detection of the detection system in practical applications, especially in new terrain or environmental changes, and enhances the reliability and stability of the results.

[0056] 2. Enhanced robustness and adaptation to complex backgrounds and diverse scenarios: This application effectively improves the adaptability of the model through the evidence learning mechanism, so that it still performs well under complex backgrounds or noise interference. Since remote sensing images usually have multi-scale and multi-directional characteristics, traditional models have difficulty handling such complex backgrounds, and this application can significantly improve the detection accuracy and robustness of the model. Especially when the target position changes greatly or is irregularly distributed, the adaptability of this application enables it to maintain stable detection performance in a variety of scenarios and has stronger generalization capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0058] Figure 1 A flowchart of a remote sensing target detection method based on evidence learning provided in an embodiment of the present application;

[0059] Figure 2 A flowchart of the basic model training provided in the embodiment of the present application;

[0060] Figure 3 A total strength calculation diagram provided for an embodiment of the present application;

[0061] Figure 4 A flow chart for evaluating classification confidence provided in an embodiment of the present application;

[0062] Figure 5 A flowchart of the basic model training provided in the embodiment of the present application;

[0063] Figure 6 A flow chart of target detection model training provided in an embodiment of the present application;

[0064] Figure 7 A schematic diagram of the structure of a remote sensing target detection device based on evidence learning provided in an embodiment of the present application.

[0065] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0066] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0067] In the technical solution of this application, the collection, storage, use, processing, transmission, provision and disclosure of information such as financial data or user data involved shall comply with the provisions of relevant laws and regulations and shall not violate public order and good morals.

[0068] It should be noted that in the embodiments of the present application, certain software, components, models and other existing solutions in the industry may be mentioned, and they should be regarded as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of the present application, but it does not mean that the applicant has or will necessarily use the solution.

[0069] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0070] The embodiment of the present application provides a remote sensing target detection method based on evidence learning. Figure 1 A flowchart of a remote sensing target detection method based on evidence learning is provided in an embodiment of the present application. Figure 1 As shown, the remote sensing target detection method based on evidence learning includes the following steps S10 to S40, which are described in detail as follows.

[0071] S10, the remote sensing image dataset is preprocessed and input into the basic target detection model, and the classification loss is used With the regression loss L bbox Perform model training to obtain the basic model W.

[0072] In some embodiments, Figure 2 As shown, the detailed steps of step S10 are as follows:

[0073] S101, preprocessing the samples of the training data set, setting the overlapping pixel distance gap between the two segmented images to 200, segmenting the training sample images, the resolution of the segmented images is subsize = 1024 * 1024, and using the scale factors rates = 0.5, 1.0, 1.5 to obtain multi-scale training images as the data set.

[0074] S102, organize the training sample labels into the format of (x1, y1, x2, y2, x3, y3, x4, y4, n, d), where (x1, y1), (x2, y2), (x3, y3), and (x4, y4) represent the horizontal and vertical coordinates of the four points of the target frame, n represents the number of categories of the target, and d represents the difficulty of distinguishing the target;

[0075] S103, input the sorted data set into the basic model network, set the learning rate to l = 0.0002, and use the classification loss of the basic model Regression Loss The model is trained to obtain a trained basic model W. The sorted data set refers to a data set obtained by combining the data set with the corresponding training sample labels.

[0076] S20, add the evidence learning head to the detection head of the basic model W, and use the function F α Calculate the evidence value to convert the network output into evidence representation and restrict the input feature value range to

[0077] The total strength is obtained by calculating the sum of the evidence values ​​of different categories of each sample, where is the minimum value of the restricted eigenvalue, y is the input eigenvalue, and n is the maximum value of the restricted eigenvalue.

[0078] In some embodiments, Figure 3 As shown, the detailed steps of step S20 are as follows:

[0079] S201, add an evidence learning head to the detection head of the basic model W, and use the function F α Calculate the evidence value, the category score output by the basic model for the i-th sample for the j-th category is y ij , for y ij Apply the exponential function to convert it into evidence value α ij , these evidence values ​​represent the confidence of each category, then the process can be characterized as formula (1): ij

[0080] F α =log(y ij +1)=α ij (1)

[0081] S202. Limit the range of evidence values ​​to

[0082] In this embodiment, in order to improve the stability of the value, the evidence value is restricted. For example, the range of the evidence value is limited to and are the minimum and maximum values ​​of the evidence value, that is, clipping it to a reasonable range, ensuring the stability of the value during calculation and avoiding the disappearance or explosion of the gradient caused by too large or too small values. The process can be represented by formula (2):

[0083]

[0084] In the formula, clamp represents a numerical function that returns a range, which limits the increasing, decreasing, or randomly changing values ​​to a range of values.

[0085] S203, sum up the evidence values ​​of all categories to obtain the evidence strength S of each sample i , for sample i, its evidence for n categories is α i1 , α i2 , α i3 , ..., α in , the total evidence strength of sample i is S i , which quantifies the overall confidence of the model for the sample, the process can be represented by formula (3):

[0086]

[0087] S30, use The function defines the evidence learning loss, which evaluates the classification confidence by quantifying the classification accuracy and evidence strength of the basic model for different categories.

[0088] In some embodiments, Figure 4 As shown, the detailed steps of step S30 are as follows:

[0089] S301, by calculating the total evidence strength S of each sample i on all categories i The evidence value α predicted by each class sample ij The error between them represents the difference in the strength of evidence between different categories in the sample. The process can be expressed as formula (4):

[0090]

[0091] S302: Encoding of the real label of the sample If the i-th sample belongs to the j-th category, then No The process can be expressed as formula (5):

[0092]

[0093] S303, encoding according to the real label of the sample Get evidence to learn the loss function Ensuring that the model achieves a balance between overall confidence and confidence between categories helps the model to be stable when dealing with uncertainty and confidence, and improves the ability to distinguish different categories. The evidence learning loss function can be expressed as formula (6):

[0094]

[0095] S40, classification loss according to the basic model Regression Loss and the evidence learning loss Retrain the basic object detection model to obtain the object detection model.

[0096] In some embodiments, Figure 5 As shown, the detailed steps of step S40 are as follows:

[0097] S401. Calculate classification loss based on the output of the basic model The classification loss function measures the classification accuracy of the model in each category, and the process can be represented as formula (7):

[0098]

[0099] In the formula, It represents the probability that the i-th sample belongs to the j-th category.

[0100] S402. Calculate regression loss This includes smoothing loss and bounding box regression loss The regression loss measures the accuracy of the model in predicting the bounding box position and angle. It can be expressed as formula (8):

[0101]

[0102] In the formula, t i,j is the target regression value of the i-th sample, is the predicted value of the i-th sample, K is the number of regression target dimensions, N is the total number of samples, and Smooth is the Smooth smoothing function.

[0103] S403, classification loss combined with the basic model Regression Loss and the evidence learning loss The corresponding weights λ cls , reg and λ edl Multiply them together to get the total loss function And retrain the model to obtain the model parameters W' based on evidence learning, then the process can be represented as formula (9):

[0104]

[0105] In the formula, represents the total loss, λ cls , reg and λ edl Respectively represent the corresponding classification loss Regression Loss and the evidence learning loss The weight of .

[0106] Exemplarily, an architecture diagram of a basic target detection model is provided, such as Figure 6 As shown in the figure, the basic target detection model includes a backbone network, a feature pyramid and a target detection head. The target detection head is connected to the classification head, regression head and evidence learning head, which correspond to classification loss, regression loss and evidence loss respectively. The basic target detection model first inputs the training set into the backbone network in batches, and then generates feature maps of different sizes through convolution calculation, pooling and other operations; the feature maps are sequentially passed through the feature pyramid and the target detection head, and the prediction vector is output. The model is trained using the total loss function to obtain the target detection model.

[0107] The present application also provides a remote sensing target detection device based on evidence learning, such as Figure 7 As shown, the remote sensing target detection device based on evidence learning includes:

[0108] The first training module 701 is configured to input the remote sensing image data set into the basic target detection model after preprocessing, and use the classification loss With the regression loss L bbox Perform model training to obtain the basic model W;

[0109] The total strength calculation module 702 is configured to add an evidence learning head to the detection head part of the basic model W, using the function F α Calculate the evidence value to convert the network output into evidence representation and restrict the input feature value range to The total strength is obtained by calculating the sum of the evidence values ​​of different categories of each sample, where is the minimum value of the evidence value, y is the input feature value, and n is the number of target categories;

[0110] The evidence learning loss determination module 703 is configured to use The function defines the evidence learning loss, which evaluates the classification confidence by quantifying the classification accuracy and evidence strength of the basic model for different categories.

[0111] The second training module 704 is configured to classify the loss of the base model according to Regression Loss and the evidence learning loss The basic target detection model is retrained to obtain a target detection model; the target detection model is used to realize target recognition in remote sensing images.

[0112] In some embodiments, the first training module is further configured to:

[0113] Preprocess the remote sensing images of the remote sensing image dataset. Separate the remote sensing images based on the overlapping pixel distance between the two segmented images. The resolution of the segmented images is subsize = 1024 * 1024. Use the scale factors rates = 0.5, 1.0, 1.5 to obtain multi-scale remote sensing images as the preprocessing dataset.

[0114] The labels of the remote sensing images are organized into the format of (x1, y1, x2, y2, x3, y3, x4, y4, n, d), where (x1, y1), (x2, y2), (x3, y3), and (x4, y4) represent the horizontal and vertical coordinates of the four points of the target frame, n represents the number of target categories, and d represents the difficulty of distinguishing the target;

[0115] The preprocessed dataset and the labels of the corresponding remote sensing images are input into the basic target detection model, and the classification loss is used based on the set learning rate. Regression Loss Train the model and obtain the trained basic model W.

[0116] In some embodiments, the total intensity calculation module is further configured to:

[0117] Add an evidence learning head to the detection head of the basic model W and use the function F α Calculate the evidence value, the category score output by the basic model for the i-th sample for the j-th category is y ij , apply an exponential function to the class score and convert it into evidence value α ij , the evidence value represents the confidence of each category, and the calculation process is expressed as:

[0118] F α =log(y ij +1)=α ij (1)

[0119] Limit the range of evidence values ​​to is the maximum value of the evidence value, and the calculation process is expressed as:

[0120]

[0121] In the formula, clamp represents a numerical function in the return range, which limits the increasing, decreasing or randomly changing values ​​to a range of values;

[0122] The total evidence strength of the i-th sample is calculated as S by the following formula i :

[0123]

[0124] In the formula, α ij represents the evidence of the i-th sample for the j-th category, and n represents the number of target categories.

[0125] In some embodiments, the evidential learning loss determination module is further configured to determine function:

[0126] The total evidence strength S of each sample in all categories is calculated based on the following formula i The evidence value α predicted by each class sample ij The error between:

[0127]

[0128] Determine the encoding of the sample's true label based on the following formula

[0129]

[0130] According to the encoding of the sample's true label Calculated Function, expressed as:

[0131]

[0132] In some embodiments, the second training module is further configured to:

[0133] Calculate the classification loss based on the output of the base model The loss function measures the classification accuracy of the model in each category, and the calculation process is:

[0134]

[0135] In the formula, Represents the probability that the i-th sample belongs to the j-th category;

[0136] Calculating regression loss Include Smoothing loss and bounding box regression loss The regression loss is used to measure the accuracy of the model in predicting the bounding box position and angle. It is expressed as:

[0137]

[0138] Where, t i,j is the target regression value of the i-th sample, is the predicted value of the i-th sample, K is the number of regression target dimensions, N is the total number of samples, and Smooth is the Smooth smoothing function;

[0139] Classification loss based on the base model Regression Loss and Evidence Learning The loss is obtained, and the total loss function is obtained;

[0140] The model is retrained based on the total loss function to obtain the model parameters W' based on evidence learning, and the model parameters W' are used to configure the basic target detection model to obtain the target detection model.

[0141] In some embodiments, the second training module is further configured to be based on the classification loss of the base model Regression Loss and the evidence learning loss The total loss function is obtained by the following formula;

[0142]

[0143] In the formula, represents the total loss, λ cls , reg and λ edl Respectively represent the corresponding classification loss Regression Loss and the evidence learning loss The weight of .

[0144] An embodiment of the present application provides an electronic device, which may include: a processor and a memory, wherein the processor and the memory may communicate with each other; illustratively, the processor and the memory communicate with each other via a communication bus.

[0145] The processor executes the computer execution instructions stored in the memory, so that the processor executes the scheme in the above embodiment. The processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit ASIC, a field programmable gate array FPGA or other programmable logic devices, discrete gates or transistor logic devices, and discrete hardware components.

[0146] The communication bus can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The system bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus. The transceiver is used to realize the communication between the database access device and other computers (such as clients, read-write libraries, and read-only libraries). The memory may include random access memory (RAM) and may also include non-volatile memory.

[0147] The electronic device provided in the embodiment of the present application may be the terminal device of the above embodiment.

[0148] An embodiment of the present application also provides a computer-readable storage medium, which stores computer instructions. When the computer instructions are executed on a computer, the computer executes the technical solution of the remote sensing target detection method based on evidence learning in the above embodiment.

[0149] An embodiment of the present application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When at least one processor executes the computer program, it can implement the technical solution of the remote sensing target detection method based on evidence learning in the above embodiment.

[0150] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of modules is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0151] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to implement the solution of this embodiment.

[0152] In addition, each functional module in each embodiment of the present application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The above-mentioned module-composed unit can be implemented in the form of hardware or in the form of hardware plus software functional units.

[0153] The above-mentioned integrated module implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to perform some steps of the methods of various embodiments of the present application.

[0154] It should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the invention may be directly implemented as being executed by a hardware processor, or may be implemented by a combination of hardware and software modules in the processor.

[0155] The memory may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk memory, and may also be a USB flash drive, a mobile hard disk, a read-only memory, a magnetic disk or an optical disk, etc.

[0156] The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application is not limited to only one bus or one type of bus.

[0157] The above storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The storage medium can be any available medium that can be accessed by a general or special purpose computer.

[0158] An exemplary storage medium is coupled to a processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the storage medium can also exist as discrete components in an electronic control unit or a main control device.

[0159] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0160] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A remote sensing target detection method based on evidence learning, characterized in that: The method comprises: The remote sensing image dataset is preprocessed and input into the basic target detection model, and the classification loss is used With the regression loss L bbox Perform model training to obtain the basic model W; Add an evidence learning head to the detection head of the basic model W and use the function F α Calculate the evidence value to convert the network output into evidence representation and restrict the input feature value range to The total strength is obtained by calculating the sum of the evidence values ​​of different categories of each sample, where is the minimum value of the restricted eigenvalue, y is the input eigenvalue, and n is the maximum value of the restricted eigenvalue; use The function defines the evidence learning loss, which evaluates the classification confidence by quantifying the classification accuracy and evidence strength of the basic model for different categories. Classification loss according to the base model Regression Loss and the evidence learning loss The basic target detection model is retrained to obtain a target detection model; the target detection model is used to realize target recognition in remote sensing images.

2. The remote sensing target detection method based on evidence learning according to claim 1 is characterized in that: The remote sensing image dataset is preprocessed and input into the basic target detection model, and the classification loss is used With the regression loss L bbox Perform model training to obtain the basic model W, including: Preprocess the remote sensing images of the remote sensing image dataset. Separate the remote sensing images based on the overlapping pixel distance between the two segmented images. The resolution of the segmented images is subsize = 1024 * 1024. Use the scale factors rates = 0.5, 1.0, 1.5 to obtain multi-scale remote sensing images as the preprocessing dataset. The labels of the remote sensing images are organized into the format of (x1, y1, x2, y2, x3, y3, x4, y4, n, d), where (x1, y1), (x2, y2), (x3, y3), and (x4, y4) represent the horizontal and vertical coordinates of the four points of the target frame, n represents the number of target categories, and d represents the difficulty of distinguishing the target; The preprocessed dataset and the labels of the corresponding remote sensing images are input into the basic target detection model, and the classification loss is used based on the set learning rate. Regression Loss Train the model and obtain the trained basic model W.

3. The remote sensing target detection method based on evidence learning according to claim 1 is characterized in that: Add an evidence learning head to the detection head of the basic model W and use the function F α Calculate the evidence value to convert the network output into evidence representation and restrict the input feature value range to The total strength is calculated by summing up the evidence values ​​of different categories for each sample, including: Add an evidence learning head to the detection head of the basic model W and use the function F α Calculate the evidence value, the category score output by the basic model for the i-th sample for the j-th category is y ij , apply an exponential function to the class score and convert it into evidence value α ij , the evidence value represents the confidence of each category, and the calculation process is expressed as: F α =log(y ij +1)=a ij (1) Limit the range of evidence values ​​to is the maximum value of the evidence value, and the calculation process is expressed as: In the formula, clamp represents a numerical function within the return range; The total evidence strength of the i-th sample is calculated as S by the following formula i : In the formula, α ij represents the evidence of the i-th sample for the j-th category, and n represents the number of target categories.

4. The remote sensing target detection method based on evidence learning according to claim 3 is characterized in that: Determine by the following method function: The total evidence strength S of each sample in all categories is calculated based on the following formula i The evidence value α predicted by each class sample ij The error between: Determine the encoding of the sample's true label based on the following formula According to the encoding of the sample's true label Calculated Function, expressed as: 。 5. The remote sensing target detection method based on evidence learning according to claim 4 is characterized in that: Classification loss according to the base model Regression Loss and the evidence learning loss Retrain the basic target detection model to obtain the target detection model, including: Calculate the classification loss based on the output of the base model The loss function measures the classification accuracy of the model in each category, and the calculation process is: In the formula, Represents the probability that the i-th sample belongs to the j-th category; Calculating regression loss Include Smoothing loss and bounding box regression loss The regression loss is used to measure the accuracy of the model in predicting the bounding box position and angle. It is expressed as: Where, t i,j is the target regression value of the i-th sample, is the predicted value of the i-th sample, K is the number of regression target dimensions, N is the total number of samples, and Smmoth is the Smmoth smoothing function; Classification loss based on the base model Regression Loss and Evidence-Based Learning The loss is obtained, and the total loss function is obtained; The model is retrained based on the total loss function to obtain the model parameters W' based on evidence learning, and the model parameters W' are used to configure the basic target detection model to obtain the target detection model.

6. The remote sensing target detection method based on evidence learning according to claim 5, characterized in that: Classification loss based on the base model Regression Loss and the evidence learning loss The total loss function is obtained by the following formula; In the formula, represents the total loss, λ cls , reg and λ edl Respectively represent the corresponding classification loss Regression Loss and the evidence learning loss The weight of .

7. A remote sensing target detection device based on evidence learning, characterized in that: The device comprises: The first training module is configured to input the remote sensing image dataset into the basic target detection model after preprocessing, and use the classification loss With the regression loss L bbox Perform model training to obtain the basic model W; The total strength calculation module is configured to add an evidence learning head to the detection head part of the basic model W, using the function F α Calculate the evidence value to convert the network output into evidence representation and restrict the input feature value range to The total strength is obtained by calculating the sum of the evidence values ​​of different categories of each sample, where is the minimum value of the restricted eigenvalue, y is the input eigenvalue, and n is the maximum value of the restricted eigenvalue; The evidence learning loss determination module is configured to use The function defines the evidence learning loss, which evaluates the classification confidence by quantifying the classification accuracy and evidence strength of the basic model for different categories. The second training module is configured to be based on the classification loss of the base model Regression Loss and the evidence learning loss The basic target detection model is retrained to obtain a target detection model; the target detection model is used to realize target recognition in remote sensing images.

8. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the remote sensing target detection method based on evidence learning as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the remote sensing target detection method based on evidence learning as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the remote sensing target detection method based on evidence learning as described in any one of claims 1 to 6 is implemented.

Images