Unmanned aerial vehicle detection method and device

Abstract

The invention provides an unmanned aerial vehicle detection method and device. The unmanned aerial vehicle detection method comprises the steps of obtaining a to-be-detected image; detecting a to-be-detected image based on the unmanned aerial vehicle detection model to obtain an unmanned aerial vehicle detection result; wherein the unmanned aerial vehicle detection model is obtained by training a neural network based on a Transform architecture by taking a mixed sample set as a training sample and taking a fusion feature as a training feature; the mixed sample set is obtained by shooting in a plurality of different unmanned aerial vehicle activity scenes through a multi-dimensional optical device; the fusion features are obtained by performing cross-modal calculation on a plurality of features extracted from the mixed sample set according to a Transform module. According to the method, the features in respective modes can be accurately captured, the fused features reserve key features in each mode, and the accuracy of the unmanned aerial vehicle detection result in a complex environment is improved through complementation and enhancement of the features.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and in particular to a method and device for detecting a drone. Background Art

[0002] The extremely small target drone detection algorithm can improve the detection capability of low-altitude, slow-speed, and small flying objects, helping to promptly detect and respond to potential air threats; for example, in the field of border security, detection drones can enhance monitoring of illegal drones crossing the border to prevent smuggling and illegal immigration; in the field of urban security, detection drones can help to promptly detect and stop illegal drone photography, delivery of contraband, etc.; in addition, in the security work of important activities or venues, detection drones can enhance the monitoring and defense capabilities of drones to ensure the safety of activities.

[0003] In the prior art, when using target detection technology to detect drone images, the drone outline information in a single scene is usually selected as a typical feature for drone target detection. Due to the diversity of drones, interference from complex backgrounds and rapid movement of drones, the characterization ability of the above detection features is poor, resulting in poor recognition and detection capabilities for drones in complex environments, and further leading to inaccurate drone detection results. Summary of the invention

[0004] The present invention provides a drone detection method and device, which are used to solve the defect that the prior art has poor capability of selecting drone feature representation, resulting in poor recognition and detection capabilities of drones in complex environments, and improve the accuracy of drone detection results.

[0005] The present invention provides a drone detection method, comprising: Acquire an image to be detected; Detect the image to be detected based on the drone detection model to obtain a drone detection result; Among them, the drone detection model is obtained by training a neural network based on the Transformer architecture using a mixed sample set as a training sample and fusion features as training features; the mixed sample set is obtained by photographing multiple different drone activity scenes using a multi-dimensional optical device; the fusion features are obtained by performing cross-modal calculations on multiple features extracted from the mixed sample set according to the Transformer module.

[0006] According to a drone detection method provided by the present invention, the neural network includes an encoding layer and a feedforward neural network corresponding to the Transformer architecture; The drone detection model is trained specifically through the following steps: Performing normalization processing on each sample image in the mixed sample set to obtain a normalized processed image; Marking the drone targets in each channel of the normalized processed image to obtain a plurality of channel marked images; Based on the encoding layer, the plurality of channel labeled images are feature encoded to obtain encoding features; and the encoding features are interacted and fused according to the feedforward neural network and the Transformer mechanism to obtain the fused features; The neural network is iteratively trained according to the fusion feature, and a loss function value corresponding to the neural network is calculated. When the loss function value is less than a loss threshold, it is determined that the neural network model training has converged, and the drone detection model is obtained.

[0007] According to a drone detection method provided by the present invention, the mixed sample set is obtained by the following steps: The mixed sample set is constructed by a multi-dimensional optical device based on RGB images collected by different filters, polarization images with multiple different polarization degrees, near-infrared images, long-wave infrared images and images of high dynamic channels.

[0008] According to a drone detection method provided by the present invention, the coding features are interacted and fused according to the feedforward neural network and the Transformer mechanism to obtain the fused features, which include: The coding features are flattened, and the flattened features are extracted according to a multi-layer and multi-head Transformer architecture to obtain new coding features, and the new coding features are merged with the corresponding flattened features to obtain the fused features.

[0009] According to a drone detection method provided by the present invention, the neural network uses the YOLOv9 target detection framework to extract features, the neural network includes a multi-feature fusion module and a detection head, and the detection head performs single-size candidate box training on the drone target by using bidirectional output.

[0010] According to a drone detection method provided by the present invention, the step of normalizing each sample image in the mixed sample set to obtain a normalized processed image includes: De-noising, enhancement, normalization and format conversion operations are performed on each sample image in the mixed sample set to obtain the normalized processed image.

[0011] The present invention also provides a drone detection device, comprising: An image acquisition module, used for acquiring an image to be detected; An image detection module is used to detect the image to be detected based on a drone detection model to obtain a drone detection result; Among them, the drone detection model is obtained by training a neural network based on the Transformer architecture using a mixed sample set as a training sample and fusion features as training features; the mixed sample set is obtained by photographing multiple different drone activity scenes using a multi-dimensional optical device; the fusion features are obtained by performing cross-modal calculations on multiple features extracted from the mixed sample set according to the Transformer module.

[0012] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, any one of the above-described drone detection methods is implemented.

[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the drone detection method described in any one of the above is implemented.

[0014] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the drone detection method described above is implemented.

[0015] The drone detection method and device provided by the present invention detect the image to be detected by using a drone detection model obtained by training a neural network based on a Transformer architecture with a mixed sample set as a training sample and with fused features as training features, and obtain the drone detection result. The method and device can accurately capture the features under each modality, and the fused features retain the key features under each modality. The accuracy of the drone detection results in complex environments is improved through the complementarity and enhancement between features. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are 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.

[0017] Figure 1 This is one of the flow charts of the UAV detection method provided by the present invention.

[0018] Figure 2 It is a schematic diagram of the process of obtaining fusion features provided by the present invention.

[0019] Figure 3This is the second flow chart of the UAV detection method provided by the present invention.

[0020] Figure 4 It is a structural schematic diagram of the UAV detection device provided by the present invention.

[0021] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0022] 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 drawings of the present invention. Obviously, the described embodiments are 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 creative work are within the scope of protection of the present invention.

[0023] Combine the following Figure 1-Figure 4 The drone detection method and device of the present invention are described.

[0024] Figure 1 It is one of the flow charts of the drone detection method provided by the present invention, such as Figure 1 As shown, the drone detection method includes the following steps: Step 110: Acquire an image to be detected.

[0025] In this step, the image to be detected includes an image captured in real time or an image obtained from a drone image database.

[0026] For example, the images to be detected may be a large number of drone images collected in different scenarios, and the image content may include diverse environments such as cities, suburbs, mountains, and seas.

[0027] Step 120, detecting the image to be detected based on the drone detection model to obtain the drone detection result; wherein the drone detection model is obtained by training a neural network based on a Transformer architecture using a mixed sample set as a training sample and fusion features as training features; the mixed sample set is obtained by photographing in multiple different drone activity scenes using a multi-dimensional optical device; the fusion features are obtained by performing cross-modal calculations on multiple features extracted from the mixed sample set according to the Transformer module.

[0028] In this step, each sample image in the mixed sample set includes drones in various postures, distances, and lighting conditions to simulate complex scenes in the real world. Multi-scene drone images ensure the diversity and richness of the dataset.

[0029] In this embodiment, the mixed sample set is obtained by the following steps: constructing a mixed sample set by using a multi-dimensional optical device based on RGB images collected by different filters, multiple polarization images with different polarization degrees, near infrared images, long-wave infrared images, and images of high dynamic channels

[0030] In this embodiment, the multi-dimensional optical device can collect drone images in different scenes and convert them into images of different dimensions such as RGB images, polarization images with three different polarization degrees, long-wave infrared images, near-infrared images and high-dynamic images through data analysis.

[0031] In this embodiment, based on images of different modalities, a feature extraction network can be used to obtain image features under a single modality, feature interaction and fusion can be achieved based on a feature fusion module, and cross-modal computing can be effectively achieved based on a Transformer module.

[0032] In this embodiment, by designing a neural network including a network structure with partial weight sharing, the training complexity is reduced and the Transformer module is allowed to perform information interaction within and between modalities.

[0033] In this embodiment, by using multiple different modal feature extraction networks to extract features of different dimensions from a mixed sample set, the features under each modality can be accurately captured.

[0034] In this embodiment, the trained image features are sent to the detection head after feature splicing, and an efficient and deep target detection framework is input to output the final detection result.

[0035] In this embodiment, the neural network uses the YOLOv9 target detection framework to extract features. The neural network includes a multi-feature fusion module and a detection head. The detection head uses bidirectional output to train a single-size candidate box for the drone target.

[0036] The drone detection method provided by the embodiment of the present invention detects the image to be detected by using a drone detection model obtained by training a neural network based on a Transformer architecture with a mixed sample set as a training sample and with fused features as training features, and obtains a drone detection result. The method can accurately capture the features under each modality, and the fused features retain the key features under each modality. The accuracy of drone detection results in complex environments is improved through complementarity and enhancement between features.

[0037] In some embodiments, the neural network includes an encoding layer and a feedforward neural network corresponding to the Transformer architecture; the drone detection model is specifically trained through the following steps: normalizing each sample image in the mixed sample set to obtain a normalized processed image; marking the drone targets in each channel of the normalized processed image to obtain multiple channel labeled images; encoding the multiple channel labeled images based on the encoding layer to obtain encoding features; and interacting and fusing the encoding features according to the feedforward neural network and Transformer mechanism to obtain fused features; iteratively training the neural network according to the fused features, and calculating the corresponding loss function value of the neural network. When the loss function value is less than the loss threshold, it is determined that the neural network model training converges to obtain the drone detection model.

[0038] In this embodiment, a data parsing and marking system applied to multidimensional optical data is constructed to extract useful information from the multidimensional optical data and perform effective data marking. Specifically, the data parsing and marking system reads image data from different optical devices and extracts basic information of the image (such as resolution, color space, etc.), and then marks the target where the drone is located in the multidimensional optical image and displays it in each channel image.

[0039] In this embodiment, the data parsing and marking system can also support the parsing and loading of multiple optical image formats, and provide a user-friendly user interface for users to perform image annotation.

[0040] For example, the data analysis and labeling system can provide annotation tools such as rectangular boxes, polygons, and key points so that users can accurately mark drone targets in multi-dimensional optical images. In the case where some channel labels are missing, the invention can automatically supplement the annotation information of other channels by annotating single-channel images.

[0041] In this embodiment, during the image annotation process performed by the data analysis and labeling system, all channel images except the long-wave infrared are in an aligned state. This is because there is a large deviation in the size of the long-wave infrared image and other images. The labeling system can assign labels to each channel image through preprocessing.

[0042] In this embodiment, a large number of drone images are collected based on a multi-dimensional optical device, and the collected images are standardized to improve the image quality and unify the data format.

[0043] Specifically, performing normalization processing on each sample image in the mixed sample set to obtain a normalized processed image includes: performing denoising, enhancement, normalization and format conversion operations on each sample image in the mixed sample set to obtain a normalized processed image.

[0044] In this embodiment, after obtaining the normalized processed images, these images are divided into a training set, a validation set, and a test set according to certain rules for subsequent network training and evaluation; and finally a multidimensional optical drone detection dataset is constructed as the basis for network training.

[0045] In this embodiment, the training set, validation set and test set can be divided according to a ratio, and the ratio coefficient is set according to user needs. For example, multiple normalized processed images are divided into a training set, a validation set and a test set according to a ratio of 7:2:1.

[0046] In this embodiment, the Transformer encoder is used to encode the features of images of different channels, and then the multi-head self-attention mechanism and feedforward neural network are used to achieve effective interaction and fusion between the features of different channels.

[0047] In this embodiment, the coding features are interacted and fused according to the feedforward neural network and the Transformer mechanism to obtain the fused features, including: flattening the coding features, extracting the flattened features according to a multi-layer and multi-head Transformer architecture to obtain new coding features, and merging the new coding features with the corresponding flattened features to obtain fused features.

[0048] Figure 2 is a schematic diagram of the process of obtaining fusion features provided by the present invention. Figure 2 In the illustrated embodiment, after the corresponding image features of images of different channels (input feature 1, input feature 2, and input feature 3) are concatenated, the features are flattened through a normalization layer, and the features are encoded in combination with a multi-head self-attention mechanism to obtain new encoded features, and the new encoded features and the corresponding flattened features are concatenated, and finally the concatenated features are separated to obtain new output features of different dimensions.

[0049] Specifically, Transformer-based multimodal feature fusion is achieved by subjecting the optical image features of each dimension output by the intermediate layer to sequence signal feature extraction processing.

[0050] For example, the image features of each modality are flattened and used as sequence information to be sent into a multi-layer multi-head Transformer architecture for feature extraction. The obtained results are split and sent together with the original features to the subsequent feature extraction network for further calculation, ultimately achieving the fusion of modal features of multiple different dimensions and obtaining the corresponding fused features.

[0051] In this embodiment, the above-mentioned neural network corresponding loss function can be a cross entropy loss function, a positioning loss function, or a combination of the cross entropy loss function and the positioning loss function.

[0052] Figure 3 This is the second flow chart of the drone detection method provided by the present invention. Figure 3 In the illustrated embodiment, data analysis is performed on each sample image in the mixed sample set by multiple optical image acquisition devices to obtain five groups of images or data of different modalities, including RGB images, near-infrared images, polarization images, long-wave infrared images, and multi-dimensional optical drone detection data sets; a pre-trained backbone network (including feature extraction network 1, feature extraction network 2, feature extraction network 3, feature extraction network 4, and feature extraction network 5) is then used to extract features from the grouped multi-dimensional images of the five modalities, and the extracted features are then sent to a feature fusion network based on a Transformer architecture for cross-channel feature splicing, and finally the detection head is used to perform target detection on the fused features.

[0053] In this embodiment, the trained drone detection model can be deployed on an algorithm server, and drone target detection can be performed on the received multi-dimensional optical images using a target detection method using multimodal feature fusion.

[0054] In this embodiment, the neural network is trained based on a GPU (Graphics Processing Unit) to obtain a trained drone detection model (for example, a trained weight file), and the model or file is deployed on an algorithm server, and the received multi-dimensional optical image is processed according to the above process to achieve drone target detection.

[0055] In this embodiment, the dataset inference results show that the input image is 4096×3000×9, and the feature fusion module is used with the RGB image for separate detection. After feature fusion, the target detection accuracy reaches 81.39%, while the target detection accuracy based on RGB image is only 63.23%, which fully proves the effectiveness of the UAV target detection method after multi-dimensional optical image feature fusion.

[0056] The drone detection method of this embodiment can be freely deployed on an algorithm server without being restricted by the hardware system, and can maximize the mining of data features.

[0057] The drone detection method provided by the embodiment of the present invention performs normalization processing on each sample image in the mixed sample set, marks the drone target of each channel in the normalized processed image, obtains multiple channel marked images, and then performs feature encoding on the multiple channel marked images based on the coding layer to obtain coding features; and interacts and fuses the coding features according to the feedforward neural network and the Transformer mechanism, and finally iteratively trains the neural network according to the fused features to obtain a drone detection model. The drone detection model integrates multi-feature information, has strong anti-interference ability, and has low computational cost, thereby realizing efficient and accurate detection of drone targets.

[0058] The drone detection device provided by the present invention is described below. The drone detection device described below and the drone detection method described above can be referenced to each other.

[0059] Figure 4 : is a schematic diagram of the structure of the drone detection device provided by the present invention, such as Figure 4 As shown, the drone detection device includes: an image acquisition module 410 and an image detection module 420.

[0060] An image acquisition module 410 is used to acquire an image to be detected; An image detection module 420 is used to detect the image to be detected based on the drone detection model to obtain a drone detection result; Among them, the drone detection model is obtained by training a neural network based on the Transformer architecture using a mixed sample set as training samples and fusion features as training features; the mixed sample set is obtained by photographing multiple different drone activity scenes using multi-dimensional optical equipment; the fusion features are obtained by performing cross-modal calculations on multiple features extracted from the mixed sample set according to the Transformer module.

[0061] The drone detection device provided by the embodiment of the present invention detects the image to be detected by using a drone detection model obtained by training a neural network based on a Transformer architecture with a mixed sample set as a training sample and with fused features as training features, and obtains a drone detection result. The device can accurately capture the features under each modality, and the fused features retain the key features under each modality. The accuracy of drone detection results in complex environments is improved through the complementarity and enhancement between features.

[0062] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the drone detection method, which includes: obtaining an image to be detected; detecting the image to be detected based on a drone detection model to obtain a drone detection result; wherein the drone detection model is obtained by training a neural network based on a Transformer architecture using a mixed sample set as a training sample and a fusion feature as a training feature; the mixed sample set is obtained by shooting in multiple different drone activity scenes using a multi-dimensional optical device; the fusion feature is obtained by performing cross-modal calculations on multiple features extracted from the mixed sample set according to the Transformer module.

[0063] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0064] On the other hand, the present invention also provides a computer program product, which includes a computer program, and the computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the drone detection method provided by the above methods, and the method includes: acquiring an image to be detected; detecting the image to be detected based on a drone detection model to obtain a drone detection result; wherein the drone detection model is obtained by training a neural network based on a Transformer architecture using a mixed sample set as a training sample and fusion features as training features; the mixed sample set is obtained by photographing multiple different drone activity scenes using a multi-dimensional optical device; the fusion features are obtained by performing cross-modal calculations on multiple features extracted from the mixed sample set according to the Transformer module.

[0065] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the drone detection method provided by the above-mentioned methods, the method comprising: acquiring an image to be detected; detecting the image to be detected based on a drone detection model to obtain a drone detection result; wherein the drone detection model is obtained by training a neural network based on a Transformer architecture using a mixed sample set as a training sample and fusion features as training features; the mixed sample set is obtained by photographing a plurality of different drone activity scenes using a multi-dimensional optical device; the fusion features are obtained by performing cross-modal calculations on a plurality of features extracted from the mixed sample set according to a Transformer module.

[0066] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0067] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0068] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention 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 make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A drone detection method, characterized in that: include: Acquire an image to be detected; Detect the image to be detected based on the drone detection model to obtain a drone detection result; Among them, the drone detection model is obtained by training a neural network based on the Transformer architecture using a mixed sample set as a training sample and fusion features as training features; the mixed sample set is obtained by photographing multiple different drone activity scenes using a multi-dimensional optical device; the fusion features are obtained by performing cross-modal calculations on multiple features extracted from the mixed sample set according to the Transformer module.

2. The drone detection method according to claim 1, characterized in that: The neural network includes an encoding layer and a feedforward neural network corresponding to the Transformer architecture; The drone detection model is trained specifically through the following steps: Performing normalization processing on each sample image in the mixed sample set to obtain a normalized processed image; Marking the drone targets in each channel of the normalized processed image to obtain a plurality of channel marked images; Performing feature encoding on the plurality of channel labeled images based on the encoding layer to obtain encoding features; and interacting and fusing the encoding features according to the feedforward neural network and the Transformer mechanism to obtain the fused features; The neural network is iteratively trained according to the fusion feature, and a loss function value corresponding to the neural network is calculated. When the loss function value is less than a loss threshold, it is determined that the neural network model training has converged, and the drone detection model is obtained.

3. The drone detection method according to any one of claims 1-2, characterized in that: The mixed sample set is obtained by the following steps: The mixed sample set is constructed by a multi-dimensional optical device based on RGB images collected by different filters, polarization images with multiple different polarization degrees, near-infrared images, long-wave infrared images and images of high dynamic channels.

4. The drone detection method according to claim 2, characterized in that: The interaction and fusion of the encoding features according to the feedforward neural network and the Transformer mechanism to obtain the fusion features includes: The coding features are flattened, and the flattened features are extracted according to a multi-layer and multi-head Transformer architecture to obtain new coding features, and the new coding features are merged with the corresponding flattened features to obtain the fused features.

5. The drone detection method according to claim 2, characterized in that: The neural network uses the YOLOv9 target detection framework to extract features. The neural network includes a multi-feature fusion module and a detection head. The detection head uses bidirectional output to perform single-size candidate box training on the drone target.

6. The drone detection method according to claim 2, characterized in that: The step of normalizing each sample image in the mixed sample set to obtain a normalized processed image comprises: De-noising, enhancement, normalization and format conversion operations are performed on each sample image in the mixed sample set to obtain the normalized processed image.

7. A drone detection device, characterized in that: include: An image acquisition module, used for acquiring an image to be detected; An image detection module is used to detect the image to be detected based on a drone detection model to obtain a drone detection result; Among them, the drone detection model is obtained by training a neural network based on the Transformer architecture using a mixed sample set as a training sample and fusion features as training features; the mixed sample set is obtained by photographing multiple different drone activity scenes using a multi-dimensional optical device; the fusion features are obtained by performing cross-modal calculations on multiple features extracted from the mixed sample set according to the Transformer module.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the drone detection method according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the drone detection method according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the drone detection method according to any one of claims 1 to 6 is implemented.

Images