Tone migration-based unmanned aerial vehicle target matching method and device, and medium

Abstract

The invention discloses an unmanned aerial vehicle target matching method and device based on hue migration, and a medium, belongs to the technical field of unmanned aerial vehicle target matching, and aims to solve the problem that the target matching accuracy is poor due to the influence of external factors such as the geographic position of an unmanned aerial vehicle, a lens shooting angle and ambient light. And the hue performance of the same target in the visual field of the unmanned aerial vehicle has significant difference, so that the target matching precision is reduced. The method comprises the following steps: acquiring a first target image of a to-be-matched target, and acquiring a template target image; performing hue migration on the first target image based on the template target image to obtain a second target image; and based on the second target image and the template target image, determining whether the target to be matched is matched with a template target corresponding to the template target image. According to the method, the accuracy of target matching of the unmanned aerial vehicle is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of drone target matching, and in particular to a method, device and medium for drone target matching based on hue migration. Background Art

[0002] Target matching is a technology widely used in many fields. Its core lies in associating and identifying the target to be matched with the corresponding object through a specific algorithm or method. The camera carried by the current drone has become a new perspective for target matching, especially in emergency rescue, security monitoring and other scenarios.

[0003] Traditional drone target matching can already meet the needs in simple scenarios. However, in actual applications, due to the influence of external factors such as the drone's geographical location, lens shooting angle, ambient light, etc., the color tone of the same target in the drone's field of view is significantly different. As a result, it may be impossible to identify and associate the same target in different drone fields of view, resulting in a decrease in the accuracy of drone target matching. Summary of the invention

[0004] The present invention provides a method, device and medium for unmanned aerial vehicle target matching based on hue migration, which is used to solve the existing technical problem that due to the influence of external factors such as the geographical location of the unmanned aerial vehicle, the shooting angle of the lens, the ambient light, etc., the hue expression of the same target in the field of view of the unmanned aerial vehicle has significant differences, which in turn leads to a decrease in the target matching accuracy.

[0005] In a first aspect, the present invention provides a method for drone target matching based on hue migration, comprising:

[0006] Acquire a first target image of a target to be matched, and acquire a template target image;

[0007] Performing tone migration on the first target image based on the template target image to obtain a second target image;

[0008] Based on the second target image and the template target image, it is determined whether the target to be matched matches the template target corresponding to the template target image.

[0009] In a feasible implementation manner, the step of acquiring a first target image of a target to be matched includes:

[0010] Get the current video frame of the drone screen;

[0011] Based on the object detector, detecting the object to be matched in the video frame;

[0012] Based on the target to be matched, the video frame is cropped to obtain a first target image corresponding to the target to be matched.

[0013] In a feasible implementation manner, performing tone migration on the first target image based on the template target image to obtain a second target image includes:

[0014] Performing color space conversion on the first target image and the template target image;

[0015] Calculate the first target image Img src The first mean src and the first standard deviation Std src ;

[0016] Calculate the template target image Img dst Template mean Mean dst and template standard deviation Std dst ;

[0017] Based on the formula Get the second target image Img trans ;

[0018] An inverse transformation of the color space transformation is performed on the second target image.

[0019] In a feasible implementation manner, determining whether the target to be matched matches a template target corresponding to the template target image based on the second target image and the template target image includes:

[0020] Determining a fusion feature of the target to be matched based on the first target image and the second target image;

[0021] Performing feature extraction on the template target image to obtain template features of the template target image;

[0022] Based on the similarity between the fusion feature and the template feature, it is determined whether the target to be matched matches the template target corresponding to the template target image.

[0023] In a feasible implementation manner, determining the fusion feature of the target to be matched based on the first target image and the second target image includes:

[0024] Performing feature extraction on the first target image and the second target image respectively to obtain a first feature corresponding to the first target image and a second feature corresponding to the second target image;

[0025] A fusion feature is obtained based on the first feature and the second feature.

[0026] In a feasible implementation manner, performing feature extraction on the first target image and the second target image respectively to obtain a first feature corresponding to the first target image and a second feature corresponding to the second target image includes:

[0027] Based on the ReID model, feature extraction is performed on the first target image and the second target image respectively, and the output of the feature layer in the ReID model is determined as the first feature and the second feature; wherein, the ReID model uses ResNet-50 as a feature encoder, and cross entropy loss and triplet loss are used for metric learning during training.

[0028] In a feasible implementation manner, obtaining a fusion feature based on the first feature and the second feature includes:

[0029] Based on Formula F fuse =αF org +(1-α)F trans , get the fusion feature, where F fuse is the fusion feature, F org For the first feature, F trans is the second feature, α is the weighted weight, and its value range is [0, 1].

[0030] In a feasible implementation manner, determining whether the target to be matched matches the template target corresponding to the template target image based on the similarity between the fusion feature and the template feature includes:

[0031] Calculating the cosine similarity between the fusion feature and the template feature;

[0032] Based on the cosine similarity, it is determined whether the target to be matched matches the template target corresponding to the template target image.

[0033] In a second aspect, the present invention provides a device for drone target matching based on hue migration, the device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor so that the at least one processor can execute a method for drone target matching based on hue migration as described in any of the above embodiments.

[0034] In a third aspect, the present invention provides a non-volatile computer storage medium, which is a non-volatile computer-readable storage medium, and the non-volatile computer-readable storage medium stores at least one program, each of which includes instructions, and when the instructions are executed by a terminal, the terminal executes a method for drone target matching based on hue migration as described in any of the above embodiments.

[0035] The method, device and medium for drone target matching based on hue migration provided by the present invention have the following beneficial technical effects compared with the prior art:

[0036] (1) After obtaining a first target image and a template target image of a target to be matched, the present invention performs tone migration on the first target image based on the template target image to obtain a second target image after tone migration. The second target image has the same tone as the template target image, thereby reducing the image difference caused by inconsistent tone and improving the accuracy of drone target matching.

[0037] (2) Based on the first target image and the second target image, the fusion feature of the target to be matched is determined, that is, the fusion feature is determined based on the two images before and after the color tone migration. The fusion feature combines the features in the original image acquired by the drone and the features of the second target image with the same color tone as the template target image, which not only reduces the image difference caused by inconsistent color tones, but also retains the features of the original image. By performing target matching through the fusion feature, the accuracy of target matching can be improved.

[0038] (3) When receiving the first target image, the present invention transfers the color tone of the first target image of the target to be matched to the template target image, then performs feature extraction based on the ReID model, and performs target matching based on the features extracted from the second target image, thereby achieving the purpose of unifying the target color tone and improving the matching accuracy under different environmental conditions, thereby ensuring the robustness and practicality of target matching. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0040] Figure 1 A flowchart of a method for UAV target matching based on hue migration provided by the present invention;

[0041] Figure 2 A schematic diagram of the architecture of the hardware part of a system for drone target matching based on hue migration provided by the present invention;

[0042] Figure 3 A flowchart of another method for drone target matching based on hue migration provided by the present invention;

[0043] Figure 4 A schematic structural diagram of a device for drone target matching based on hue migration provided by the present invention. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0045] The current drone-mounted lens has become a new perspective for target matching, especially in emergency rescue, security monitoring and other scenarios. Traditional drone target matching can already meet the needs in simple scenarios, but in actual applications, due to the influence of external factors such as shooting angle and ambient lighting, the color tone of the same target in the drone's field of view is significantly different, so there is an urgent need for a target matching method that can adapt to color tone changes.

[0046] The present invention aims to solve the problem of decreased target matching accuracy in drone target matching due to the different geographical locations, lens shooting angles, and ambient light of drones. Through the technical solution of the present invention, when a drone receives a target image, it can transfer the color tone of the image to be matched to the template target image according to the current picture color tone, and then perform ReID image matching, so as to achieve the purpose of unifying the target color tone under different environmental conditions and improving the matching accuracy, thereby ensuring the robustness and practicality of the matching model.

[0047] The technical solution proposed by the present invention is described in detail below with reference to the accompanying drawings.

[0048] Figure 1 The present invention provides a flow chart of a method for matching drone targets based on hue migration. Figure 1 As shown, the method includes the following execution steps:

[0049] S101, obtaining a first target image of a target to be matched, and obtaining a template target image.

[0050] The present invention determines whether there is a template target image that matches the target to be matched in at least one template target image. For example, the target to be matched may be a vehicle in a drone image, and the template target image may be an image corresponding to the vehicle to be searched, such as an image of a white car in good light.

[0051] As a possible implementation manner, obtaining a first target image of the target to be matched in S101 includes:

[0052] S101A, obtaining the current video frame of the drone image.

[0053] S101B, based on the target detector, detecting the target to be matched in the video frame.

[0054] The present invention requires continuous tracking of the target to be matched in the drone image (eg, all vehicles in the image).

[0055] For example, in the scenario where the target to be matched is a vehicle, in order to avoid repeatedly matching the same target, YoLov8 and DeepSORT may be used to detect and track the target to be matched.

[0056] After acquiring the current video frame, the video frame is fed into the YoLov8 target detector for multi-target tracking. This target detector only detects vehicles and is pre-trained using the VisDrone dataset (vehicle dataset from drone perspective). In order to better adapt to the usage scenario, the target detector can be fine-tuned by using the training data corresponding to the scenario.

[0057] According to the target detection results, DeepSORT is used as the tracker. Since the feature extraction network of the tracker was originally designed and trained for pedestrian appearance features, the ReID model described below can be reused to enhance the tracker's ability to extract appearance features and achieve stable tracking of vehicles in the picture.

[0058] S101C: based on the target to be matched, crop the video frame to obtain a first target image corresponding to the target to be matched.

[0059] S102: Perform tone migration on the first target image based on the template target image to obtain a second target image.

[0060] When a drone matches a target through captured images, weather conditions, shooting angles and ambient light often change dynamically over time, resulting in inconsistent tones between the first target image and the template target image, and reducing the similarity between the target to be matched in the first target image and the template target. Tone migration (or color migration) is an image processing technology that aims to transfer the color features of an image (source image) to another image (target image) so that the two images have similar color features. After tone migration, the second target image in the present invention has the same tones as the template target image, and still retains the main features of the first target image.

[0061] As a possible implementation, S102 performs tone migration on the first target image based on the template target image to obtain the second target image, including the following steps:

[0062] S102A: Perform color space conversion on the first target image and the template target image.

[0063] Since there is a strong correlation between RGB channels in hue migration, it is necessary to perform color space conversion on the template target image and the first target image, and convert the template target image and the first target image from RGB space to Lab space. The conversion formula is as follows.

[0064]

[0065] S102B, calculate the first target image Img src The first mean src and the first standard deviation Std src ; Calculate the template target image Img dst Template mean Mean dst and template standard deviation Std dst ; Based on the formula Get the second target image Img trans .

[0066] S102C, performing an inverse transformation of the color space transformation in S102A on the second target image.

[0067] After obtaining the second target image, the present invention converts it into an image in RGB space to complete the hue migration. At this time, the hues of the second target image and the template target image remain consistent, which reduces the phenomenon that the color of the target to be matched in the image changes due to lighting changes.

[0068] S103: Based on the second target image and the template target image, determine whether the target to be matched matches the template target corresponding to the template target image.

[0069] In the present invention, the second target image has the same tone as the template target image, thereby reducing the difference between the target to be matched and the template target caused by inconsistent image tones. In this way, target matching can be performed by matching the second target image after tone migration with the template target image, thereby improving the accuracy of target matching.

[0070] For example, the template target image is a vehicle image taken on a sunny day, but the drone acquires the first target image on a cloudy day. Compared to sunny days, the color of the vehicle body in the first target image may change, which will cause the accuracy of target matching to decrease. After the color tone migration, the second target image has the same color tone as the template target image, reducing the difference caused by inconsistent color tones and improving the accuracy of target matching.

[0071] As a possible implementation manner, S103 determines whether the target to be matched matches the template target corresponding to the template target image based on the second target image and the template target image, including:

[0072] S103A: Determine fusion features of the target to be matched based on the first target image and the second target image.

[0073] In the present invention, the first target image corresponds to the original image acquired by the drone, that is, the image before the hue migration, and the second target image is the image after the hue migration; based on the first target image and the second target image, the fusion features of the target to be matched are determined, that is, the fusion features determined based on the two images before and after the hue migration.

[0074] This fusion feature combines the features of the original image acquired by the drone and the features of the second target image with the same tone as the template target image, which not only reduces the image difference caused by inconsistent tone, but also retains the features of the original image. By performing target matching through this fusion feature, the accuracy of target matching can be improved.

[0075] As a possible implementation manner, S103A determines the fusion features of the target to be matched based on the first target image and the second target image, including:

[0076] SA1. Perform feature extraction on the first target image and the second target image respectively to obtain a first feature corresponding to the first target image and a second feature corresponding to the second target image.

[0077] As a possible implementation method, based on the ReID model, features are extracted for the first target image and the second target image respectively, and the output of the feature layer in the ReID model is determined as the first feature and the second feature; wherein, the ReID model uses ResNet-50 as the feature encoder, and cross entropy loss and triplet loss are used for metric learning during training.

[0078] SA2. Based on the first feature and the second feature, a fusion feature is obtained.

[0079] As a possible implementation method, based on formula F fuse =αF org +(1-α)F trans , get the fusion feature, where F fuse is the fusion feature, F org is the first feature, F trans is the second feature, α is the weighted value, and its value range is [0, 1].

[0080] S103B, extracting features of the template target image to obtain template features of the template target image.

[0081] The present invention can adopt the same feature extraction method as the first feature and the second feature to extract features from the template target image to obtain the template features of the template target image.

[0082] S103C: Based on the similarity between the fusion feature and the template feature, determine whether the target to be matched matches the template target corresponding to the template target image.

[0083] In the present invention, the fusion feature represents the feature of the target to be matched, the template feature represents the feature of the template target, and the similarity between the fusion feature and the target feature also represents the similarity between the target to be matched and the template target. Therefore, whether the target to be matched matches the template target can be determined based on the similarity between the fusion feature and the template feature.

[0084] In some embodiments, the target to be matched may be matched with multiple template targets, and the template target with the highest similarity may be determined as the template target that matches the target to be matched, thereby completing target matching.

[0085] In some embodiments, a similarity threshold may be set based on the above example. When the highest similarity exceeds the similarity threshold, the template target with the highest similarity is determined as the template target that matches the target to be matched.

[0086] As a possible implementation method, the cosine similarity between vectors can characterize the similarity between two vectors. S103C determines whether the target to be matched matches the template target corresponding to the template target image based on the similarity between the fusion feature and the template feature, including: calculating the cosine similarity between the fusion feature and the template feature; based on the cosine similarity, determining whether the target to be matched matches the template target corresponding to the template target image.

[0087] Exemplarily, a method for drone target matching based on hue migration provided by the present invention can be implemented by a drone target matching system based on hue migration. The system is mainly divided into two parts: a hardware part and a software part.

[0088] Figure 2 A schematic diagram of the architecture of the hardware part of a system for drone target matching based on hue migration provided by the present invention, such as Figure 2As shown, the hardware part includes: drones and drone AI kits (high-performance airborne edge computing platform). The drone AI kit is a drone expansion product, equipped with the NVIDIA Orin NX high-performance image computing unit, which is used with drones. It is small, light and easy to disassemble. The drone AI kit is connected to the drone through the mounting port, including power and network connection; the drone AI kit includes an external interface board, a main control board and a wireless networking module; the external interface board is used for power supply and network management, and is connected to the main control board through the network, and is connected to the wireless networking board through USB and HDMI; the main control board includes the NVIDIA Orin NX high-performance image computing unit, which is connected to the 5G module, hard disk, and monitoring microcontroller; the wireless networking board includes a wireless networking module; the drone AI kit also includes a 5G antenna and a self-organizing network antenna.

[0089] The software part is the target tracking, tone migration and ReID algorithms deployed in the drone AI kit (a high-performance airborne edge computing platform), which is responsible for video stream decoding, model inference, etc.

[0090] First, it is necessary to continuously track the targets to be matched in the picture (for example, all the vehicles in the picture) to avoid repeated matching of the same target. Therefore, YoLov8+DeepSORT is used to detect and track the target instances.

[0091] Secondly, hue transfer is an image processing technology that transfers the color features (hue, contrast) of a reference image to a target image, making the two more similar in color. For example, an autumn-style image can appear in summer tones after transfer. Therefore, the target image to be matched is cropped for the target tracking detection frame, and the template target image (the vehicle to be searched, such as a white car in good light) is used as the specified hue. Color Transfer is used to transfer the hue to reduce the impact of light on the target appearance color.

[0092] Finally, ReID is used to extract features from the image without tone migration and the image after migration, and the corresponding feature vectors of the two are integrated, and finally the similarity is calculated with the feature vector of the template target image.

[0093] When a drone searches for a specified target, the weather conditions, shooting angle and ambient light of the target often change dynamically over time, resulting in significant differences in the appearance of the template target and the target to be matched in the image, thereby reducing the similarity between the two and limiting the use scenarios of drone target matching. The present invention continuously tracks the target to be matched in the image through target tracking, and uses hue migration to unify the hue to the template target image, and combines the features of the images before and after the migration to perform ReID image matching, thereby enhancing the use scope of drone target matching.

[0094] Taking the target to be matched as a vehicle as an example, the method provided by the present invention is described as follows: Figure 3 Shown is a flow chart of another method for UAV target matching based on hue migration provided by the present invention, which includes the following steps.

[0095] Get the current video frame and send it to the YoLov8 target detector for multi-target tracking. It only detects vehicles as targets and uses the VisDrone dataset (drone-perspective vehicle dataset) for pre-training. In order to better adapt to the usage scenario, it is fine-tuned in its own dataset.

[0096] According to the target detection results, DeepSORT is used as a tracker. Since the feature extraction network was originally designed and trained for pedestrian appearance features, this example reuses the ReID feature extraction model described below to enhance the tracker's ability to extract appearance features and achieve stable tracking of vehicles in the picture. At the same time, the feature vector corresponding to the template target is calculated and saved. Here, it is only calculated once until the algorithm receives a new template target.

[0097] According to the recorded IDs that have been attempted to match, select the target frame that has not been matched, crop the image, and save it. In hue migration, because there is a strong correlation between RGB channels, the template target image and the cropped image (first target image) need to be converted to color space, and RGB is converted to lab space. The conversion formula is as follows:

[0098]

[0099] Then, the template target image (Img dst ) and cropped image (Img src )The mean of each channel dst 、Mean src ) and standard deviation (Std dst 、Std src ), and perform zero-mean normalization on the cropped image and then set Mean dst 、Std dst Transform to the cropped image, the formula is as follows:

[0100]

[0101] Save Std here dst With Mean dst , so as to facilitate subsequent color migration. Finally, it is converted into an image in RGB space to complete the color migration. At this time, the tones of the two images are consistent, which reduces the color change of the vehicle body caused by the change of lighting.

[0102] The ReID model is used to extract features of the images before and after the tone transfer, respectively: F org 、F trans , which uses ResNet-50 as the feature encoder, and uses cross entropy loss and triple loss for metric learning during training; the classification layer is discarded during inference, and the output of the feature layer is used as the image feature. org 、F trans Find the weighted average as follows:

[0103] F fuse =αF org +(1-α)F trans

[0104] Among them, α is the weighted weight, with a value range of [0,1], to achieve the fusion of feature vectors; then the distance matrix is ​​calculated with the feature vector of the template target using cosine similarity. If the similarity of the target with the highest similarity is greater than the set threshold, the target is matched, otherwise all target IDs of this attempt to match are recorded, and then the video frame is obtained again.

[0105] Corresponding to the above embodiment, the present invention also provides a device for drone target matching based on hue migration. Figure 4 A structural schematic diagram of a device for drone target matching based on hue migration provided in an embodiment of the present invention, the device may include: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor so that the at least one processor can execute a method for drone target matching based on hue migration as described in the above embodiment.

[0106] In a possible implementation of the present invention, the aforementioned at least one processor is capable of executing the following steps: obtaining a first target image of a target to be matched and obtaining a template target image; performing tone migration on the first target image based on the template target image to obtain a second target image; and determining whether the target to be matched matches the template target corresponding to the template target image based on the second target image and the template target image.

[0107] In a specific implementation, the present invention can also be provided as a non-volatile computer storage medium, wherein the storage medium is a non-volatile computer-readable storage medium, and the non-volatile computer-readable storage medium stores at least one program, each program including instructions, which, when executed by a terminal, enable the terminal to execute a method for drone target matching based on hue migration as described in the above embodiment.

[0108] In a possible implementation of the present invention, the aforementioned terminal executes, obtaining a first target image of the target to be matched and obtaining a template target image; performing tone migration on the first target image based on the template target image to obtain a second target image; and determining whether the target to be matched matches the template target corresponding to the template target image based on the second target image and the template target image.

[0109] The various embodiments of the present invention are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0110] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0111] The above description is only an embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for drone target matching based on hue migration, characterized in that: The method comprises: Acquire a first target image of a target to be matched, and acquire a template target image; Performing tone migration on the first target image based on the template target image to obtain a second target image; Based on the second target image and the template target image, it is determined whether the target to be matched matches the template target corresponding to the template target image.

2. The method for drone target matching based on hue migration according to claim 1, characterized in that: The step of acquiring a first target image of a target to be matched comprises: Get the current video frame of the drone screen; Based on the object detector, detecting the object to be matched in the video frame; Based on the target to be matched, the video frame is cropped to obtain a first target image corresponding to the target to be matched.

3. The method for drone target matching based on hue migration according to claim 1, characterized in that: Performing tone migration on the first target image based on the template target image to obtain a second target image includes: Performing color space conversion on the first target image and the template target image; Calculate the first target image Img src The first mean src and the first standard deviation Std src ; Calculate the template target image Img dst Template mean Mean dst and template standard deviation Std dst ; Based on the formula Get the second target image Img trans ; An inverse transformation of the color space transformation is performed on the second target image.

4. The method for drone target matching based on hue migration according to claim 1, characterized in that: Determining whether the target to be matched matches a template target corresponding to the template target image based on the second target image and the template target image includes: Determining a fusion feature of the target to be matched based on the first target image and the second target image; Performing feature extraction on the template target image to obtain template features of the template target image; Based on the similarity between the fusion feature and the template feature, it is determined whether the target to be matched matches the template target corresponding to the template target image.

5. The method for drone target matching based on hue migration according to claim 4, characterized in that: Determining the fusion feature of the to-be-matched target based on the first target image and the second target image includes: Performing feature extraction on the first target image and the second target image respectively to obtain a first feature corresponding to the first target image and a second feature corresponding to the second target image; A fusion feature is obtained based on the first feature and the second feature.

6. The method for drone target matching based on hue migration according to claim 5, characterized in that: Extracting features from the first target image and the second target image respectively to obtain a first feature corresponding to the first target image and a second feature corresponding to the second target image includes: Based on the ReID model, feature extraction is performed on the first target image and the second target image respectively, and the output of the feature layer in the ReID model is determined as the first feature and the second feature; wherein, the ReID model uses ResNet-50 as a feature encoder, and cross entropy loss and triplet loss are used for metric learning during training.

7. The method for drone target matching based on hue migration according to claim 5, characterized in that: Based on the first feature and the second feature, a fusion feature is obtained, including: Based on Formula F fuse =αF org +(1-α)F trans , get the fusion feature, where F fuse is the fusion feature, F org For the first feature, F trans is the second feature, α is the weighted weight, and its value range is [0, 1].

8. The method for drone target matching based on hue migration according to claim 4, characterized in that: Determining whether the target to be matched matches a template target corresponding to the template target image based on the similarity between the fusion feature and the template feature includes: Calculating the cosine similarity between the fusion feature and the template feature; Based on the cosine similarity, it is determined whether the target to be matched matches the template target corresponding to the template target image.

9. A device for drone target matching based on hue migration, characterized in that: The device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, so that the at least one processor can execute the method for drone target matching based on hue migration according to any one of claims 1-8.

10. A non-volatile computer storage medium, characterized in that: The storage medium is a non-volatile computer-readable storage medium, which stores at least one program, each of which includes instructions, and when the instructions are executed by the terminal, the terminal executes a method for drone target matching based on hue migration according to any one of claims 1-8.

Images