A Method and System for Unmanned Aerial Vehicle Mine Target Detection Based on Visible Light Polarization Imaging

By combining visible light polarization imaging from UAVs with deep learning technology, the problem of effective detection of landmine targets in complex backgrounds has been solved, realizing automated detection and identification of landmine targets and enhancing the accuracy and robustness of detection.

CN118196661BActive Publication Date: 2026-11-17TONGJI UNIV +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410320591.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-03-20
Publication Date
2026-11-17
Estimated Expiration
2044-03-20

AI Technical Summary

Technical Problem

Existing technologies are insufficient for effectively detecting landmine targets in complex environments such as grasslands and beaches, and traditional methods are susceptible to interference, leading to false detections and false alarms.

Method used

A UAV-based landmine target detection method based on visible light polarization imaging is adopted, combined with deep learning technology. The visible light polarization imaging module mounted on the UAV acquires polarization intensity data, calculates the Stokes vector and polarization state parameters, and uses a deep neural network with feature pyramid network and attention enhancement module for automated detection.

Benefits of technology

It enables effective detection of landmine targets in complex backgrounds, enhances the detection and identification capabilities of small and multi-scale targets, and reduces interference from background clutter.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118196661B_ABST
    Figure CN118196661B_ABST
Patent Text Reader

Abstract

This invention discloses a method and system for detecting landmine targets on unmanned aerial vehicles (UAVs) based on visible light polarization imaging. The method utilizes polarization intensity data acquired by a visible light polarization imaging module mounted on the UAV. Steps include acquiring the polarization intensity data corresponding to several polarization angles, calculating the corresponding Stokes vectors, further calculating three parameters characterizing the polarization state of the light wave, and inputting the degree of polarization, polarization angle, and elliptic angle as three-channel inputs into a deep neural network model for automated detection of surface landmine targets. This invention, based on visible light polarization imaging and combined with deep learning technology, enables the detection of surface landmine targets in complex backgrounds such as grasslands, tidal flats, and beaches, and can serve as an important tool in current integrated UAV mine detection systems.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of remote sensing target detection and recognition technology, and more specifically, to a method and system for detecting landmine targets on unmanned aerial vehicles based on visible light polarization imaging. Background Technology

[0002] Due to their concealment, camouflage, and lethality, landmines not only cause significant casualties and vehicle damage during wartime, but also continue to destroy human lives and hinder economic and social development in post-war peacetime, with numerous minefields remaining around the world. Landmine targets are typically located in complex environments such as grasslands and sandy areas. Traditional ground-penetrating radar, microwave radiometers, and classic metal detectors and visible light imaging methods for landmine detection are easily affected by complex backgrounds, leading to false detections and alarms.

[0003] Polarization is the fourth important "information dimension" parameter describing the fundamental properties of electromagnetic waves, in addition to intensity, wavelength, and phase. The polarization characteristics of reflected or radiated light from an object are closely related to its material, geometry, texture, surface roughness, and physicochemical properties. Due to the scattering effect of light illuminating a rough surface, landmine targets with different surface roughness exhibit different degrees of polarization. When light interacts with a landmine surface, areas with low reflectivity (smooth surfaces) are dominated by single scattering, while areas with high reflectivity (rough surfaces) are dominated by multiple scattering. When reflected light is dominated by single scattering, the degree of polarization is larger; when multiple scattering occurs, the degree of polarization is lower due to the uncertainty of the polarization direction of the emitted light. Polarization imaging can serve as an effective supplement to visible light, infrared, and spectral imaging, and has excellent detection capabilities for targets in low signal-to-noise ratio complex background environments, strong scattering environments, and low-illuminance environments.

[0004] Currently, lightweight unmanned aerial vehicles (UAVs) are finding increasingly widespread application in both military and civilian fields due to their significant advantages, including high maneuverability, low cost, minimal terrain limitations, and lack of loss of life. In particular, the rapid development of sensor and artificial intelligence technologies has continuously enhanced the intelligent perception, decision-making, and autonomous capabilities of UAV systems, making them important platforms and weapons for battlefield intelligence, surveillance, and reconnaissance. However, the technology for detecting multi-scale, small targets on the Earth's surface using visible light polarization imaging collected by UAVs is still lacking. Summary of the Invention

[0005] To address the aforementioned deficiencies in existing technologies, this invention provides a method and system for detecting landmine targets using unmanned aerial vehicles (UAVs). Based on visible light polarization imaging and combined with deep learning technology, it enables the detection of landmine targets on the surface against complex backgrounds such as grasslands, tidal flats, and beaches. This method can serve as an important tool in the current integrated UAV mine detection system.

[0006] To achieve the above objectives, on the one hand, the present invention provides a method for detecting landmine targets by unmanned aerial vehicles (UAVs) based on visible light polarization imaging, characterized in that, based on polarization intensity data acquired by a visible light polarization imaging module mounted on the UAV, the method includes the following steps:

[0007] Step S101: Obtain the polarization intensity data corresponding to several polarization angles;

[0008] Step S102: Calculate the corresponding Stokes vector and further calculate three parameters characterizing the polarization state of the light wave: degree of polarization, polarization angle, and elliptic angle;

[0009] Step S103: The polarization degree, polarization angle, and elliptic angle are used as three-channel inputs and fed into the deep neural network model for automated detection of landmine targets on the surface. The deep neural network model adopts a basic structure based on a feature pyramid network, and a sampling receptive field enhancement module is added to the top layer of the feature pyramid network. In the neck of the deep neural network model, the most important contextual features are separated using a spatial pyramid pool and the receptive field is increased. A path aggregation structure is used to fuse features from different feature layers. An attention enhancement module is introduced into the detection head of the deep neural network model.

[0010] Furthermore, the polarization intensity data of the polarization angle includes polarization intensity data of four polarization angles: 0°, 45°, 90°, and 135° for the scene and the target.

[0011] Furthermore, the Stokes vector equation is:

[0012]

[0013] In the above formula, I0, I 45 I 90 I 135 I represents the light intensity of the linearly polarized components at 0°, 45°, 90°, and 135°, respectively. R and I L S1 and S2 represent the light intensities of the right-hand circularly polarized and left-hand circularly polarized components of the light wave, respectively; S0 represents the total intensity of the light wave; S1 represents the intensity difference between linearly polarized light in the horizontal (0°) and vertical (90°) directions; S2 represents the intensity difference between linearly polarized light in the 45° and 135° directions; and S3 represents the intensity difference between right-hand and left-hand circularly polarized light. x (t), E y (t) is the electric field vector distributed along the x and y directions, and δ is the phase difference of the electric field vector in the x and y directions.

[0014] Further, in step S102, the formulas for obtaining the degree of polarization DoP, polarization angle AoP, and elliptic angle ε using the Stokes vector are as follows:

[0015]

[0016] The degree of polarization expresses the proportion of fully polarized light in the total light intensity, the polarization angle expresses the angle between the vibration direction of polarized light and the x-axis of the reference direction, and the elliptic angle represents the ratio of the minor axis to the major axis of the polarization ellipse.

[0017] On the other hand, the present invention provides a UAV mine target detection system based on visible light polarization imaging, characterized in that, for implementing the UAV mine target detection method as described above, it includes:

[0018] The data acquisition module acquires polarization intensity data at several polarization angles sent back by the UAV.

[0019] The data preprocessing module calculates the Stokes vector and parameters characterizing the polarization state of light waves: degree of polarization, polarization angle, and elliptic angle.

[0020] A deep neural network detector inputs the polarization degree, polarization angle, and elliptic angle into a deep neural network model to perform automated detection of landmine targets on the ground.

[0021] The deep neural network model adopts a feature pyramid network-based basic structure, and adds a sampling receptive field enhancement module to the top layer of the feature pyramid network; in the neck of the deep neural network model, a spatial pyramid pool is used to separate the most important contextual features and increase the receptive field, and a path aggregation structure is used as a connection module to fuse features from different feature layers; an attention enhancement module is introduced into the detection head of the deep neural network model.

[0022] Furthermore, the UAV includes an onboard visible light polarization imaging detection module, which includes a visible light polarization camera system and an onboard microcomputer; the onboard computer adjusts and automatically controls the imaging parameters of the polarization camera, and acquires and stores the corresponding polarization data.

[0023] Furthermore, the data acquisition module is located on the ground and acquires polarization intensity data of several polarization angles stored in the airborne computer.

[0024] Furthermore, the feature pyramid network has a 5-layer structure.

[0025] Furthermore, the sampling receptive field enhancement module first uses a 3×3 convolutional layer with a stride of 2 to reduce the feature map size while increasing the number of channels, so that subsequent convolutional kernels can acquire information over a larger spatial range. Then, it performs upsampling through simple nearest neighbor interpolation to restore the original size, and then uses a 1×1 convolutional layer to reduce the number of channels before stacking it with the feature map input to this module in terms of channel dimension. Finally, it refines and fuses the features through a 3×3 convolution.

[0026] Furthermore, the attention enhancement module dynamically learns weights based on the importance of different input features and calculates the weight information using the following formula:

[0027]

[0028] in, F is the attention-weighted feature map, where F is the input feature map. att (·;θ att ) is the attention weight calculation subnetwork, θ att These are the parameters of the sub-network. This indicates element-wise multiplication.

[0029] Compared with the prior art, the present invention has the following technical effects:

[0030] (1) This invention integrates existing visible light polarization industrial cameras and designs a visible light polarization imaging detection module that can be mounted on UAVs to realize visible light polarization imaging detection of ground targets by UAVs.

[0031] (2) This invention designs a landmine target detection system based on deep learning of visible light polarization features, proposes a basic network structure with enhanced receptive field to enhance the detection and recognition capabilities of targets of different scales and small targets, and designs an attention mechanism to reduce background clutter interference to the target.

[0032] (3) This invention can effectively detect landmine targets on the ground and can be used as an important means in the current UAV integrated mine detection system. Attached Figure Description

[0033] Figure 1 This is a flowchart of a UAV mine target detection method according to an embodiment of the present invention;

[0034] Figure 2 This is a structural block diagram of a deep neural network model in one embodiment of the present invention;

[0035] Figure 3 This is a structural block diagram of the receptive field enhancement module in one embodiment of the present invention;

[0036] Figure 4 This is a structural block diagram of the attention enhancement module in one embodiment of the present invention;

[0037] Figure 5 This is a structural diagram of a UAV mine target detection system according to an embodiment of the present invention.

[0038] Figure 6 This is a structural diagram of an airborne visible light polarization imaging detection module in one embodiment of the present invention. Detailed Implementation

[0039] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but these are not intended to limit the scope of the invention.

[0040] In the following detailed description, numerous specific details are set forth to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that well-known algorithms are not shown in detail to avoid obscuring the spirit of the invention. Algorithms not detailed in this specification are all well-known algorithms.

[0041] Furthermore, the execution order of actions, steps, etc. in the apparatus and methods shown in the claims, specification, and drawings can be implemented in any order, unless a specific order is explicitly specified, and as long as the output of the preceding processing is not used in the subsequent processing.

[0042] Example 1

[0043] See Figure 1 This embodiment provides a method for detecting landmine targets using a drone, based on polarization intensity data acquired by a visible light polarization imaging module mounted on the drone, including the following steps:

[0044] Step S101: Obtain the polarization intensity data corresponding to several polarization angles;

[0045] Step S102: Calculate the corresponding Stokes vector and further calculate three parameters characterizing the polarization state of the light wave: degree of polarization, polarization angle, and elliptic angle;

[0046] Step S103: Input the degree of polarization, polarization angle, and elliptic angle as three-channel inputs into the deep neural network model for automated detection of landmine targets on the ground: See [link to relevant documentation] Figure 2 The deep neural network model introduces an attention mechanism into the YOLOv5 object detection framework to enhance the network's ability to detect small targets such as landmines. Its basic structure is based on a feature pyramid network, with a sampling receptive field enhancement module added to the top layer of the feature pyramid network (see [link to relevant documentation]). Figure 2 The blue dashed box and Figure 3 In the neck region of the deep neural network model, spatial pyramid pooling is used to separate the most important contextual features and increase the receptive field (see...). Figure 2 (Middle blue dashed box), using path aggregation structure (see...) Figure 2 (Green line in the middle) to fuse features from different feature layers; an attention enhancement module is introduced into the detection head of the deep neural network model (see... Figure 2 The red dashed box in the middle and Figure 4 ).

[0047] As a preferred example, the polarization intensity data of the polarization angle includes polarization intensity data for four polarization angles: 0°, 45°, 90°, and 135° for the scene and the target. The corresponding Stokes vector equation is:

[0048]

[0049] In the above formula, I0, I 45 I 90 I 135 I represents the light intensity of the linearly polarized components at 0°, 45°, 90°, and 135°, respectively. R and I L S1 and S2 represent the light intensities of the right-hand circularly polarized and left-hand circularly polarized components of the light wave, respectively; S0 represents the total intensity of the light wave; S1 represents the intensity difference between linearly polarized light in the horizontal (0°) and vertical (90°) directions; S2 represents the intensity difference between linearly polarized light in the 45° and 135° directions; and S3 represents the intensity difference between right-hand and left-hand circularly polarized light. x (t), E y (t) is the electric field vector distributed along the x and y directions, and δ is the phase difference of the electric field vector in the x and y directions.

[0050] The formulas for obtaining the degree of polarization DoP, polarization angle AoP, and elliptic angle ε using the Stokes vector are as follows:

[0051]

[0052] The degree of polarization expresses the proportion of fully polarized light in the total light intensity, the polarization angle expresses the angle between the vibration direction of polarized light and the x-axis of the reference direction, and the elliptic angle represents the ratio of the minor axis to the major axis of the polarization ellipse.

[0053] See Figure 2 The receptive field enhancement module in the P5 layer of the feature pyramid network can expand the receptive field. In the detection head of the model, an attention enhancement module is used to focus feature learning on the target and reduce the interference of background noise.

[0054] If the aforementioned UAV mine target detection method is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Therefore, the essence of this technical solution, or the part that contributes to the prior art, or a portion of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0055] Example 2

[0056] See Figure 5 This embodiment provides a UAV mine target detection system based on visible light polarization imaging to implement the UAV mine target detection method described in Embodiment 1, including:

[0057] The data acquisition module 200 acquires polarization intensity data of several polarization angles sent back by the UAV 100.

[0058] The data preprocessing module 300 calculates the Stokes vector and parameters characterizing the polarization state of light waves: degree of polarization, polarization angle, and elliptic angle.

[0059] The deep neural network detector 400 inputs the polarization degree, polarization angle and elliptic angle into the deep neural network model to perform automated detection of landmine targets on the ground.

[0060] See Figure 2 Similar to Embodiment 1, the deep neural network model 400 adopts a basic structure based on a feature pyramid network, and adds a sampling receptive field enhancement module to the top layer of the feature pyramid network; in the neck of the deep neural network model, a spatial pyramid pool is used to separate the most important contextual features and increase the receptive field, and a path aggregation structure is used as a connection module to fuse features from different feature layers; an attention enhancement module is introduced into the detection head of the deep neural network model.

[0061] As a preferred technical solution, further: see Figure 6 The UAV 100 includes an onboard visible light polarization imaging detection module, which includes a visible light polarization camera system 101 and an onboard microcomputer 102. The onboard computer 102 adjusts and automatically controls the imaging parameters of the polarization camera, and acquires and stores the corresponding polarization data.

[0062] See Figure 5 Preferably, the data acquisition module is located on the ground and acquires polarization intensity data of several polarization angles stored in the airborne computer.

[0063] As a preferred technical solution, further: the feature pyramid network has a 5-layer structure to meet the needs of target semantic feature extraction. The sampling receptive field enhancement module first uses a 3×3 convolutional layer with a stride of 2 to reduce the feature map size while increasing the number of channels, enabling subsequent convolutional kernels to acquire information over a larger spatial range; then, it performs upsampling through simple nearest neighbor interpolation to restore the original size, and then uses a 1×1 convolutional layer to reduce the number of channels before stacking it with the feature map input to this module in terms of channel dimension; finally, it refines and fuses the features through a 3×3 convolution. Because landmine targets are generally small in images viewed from a UAV perspective, the sampling receptive field enhancement module with size variation can enhance the detection capability of this invention for small targets and multiple scales.

[0064] Preferably, the attention enhancement module dynamically learns weights based on the importance of different input features and calculates the weight information using the following formula:

[0065]

[0066] in, F is the attention-weighted feature map, where F is the input feature map. att (·;θ att ) is the attention weight calculation subnetwork, θ att These are the parameters of the sub-network. This indicates element-wise multiplication.

[0067] The aforementioned UAV mine target detection system can be implemented by combining firmware and software. If the UAV mine target detection function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0068] In summary, this invention provides a method and system for detecting landmine targets on unmanned aerial vehicles (UAVs) based on visible light polarization imaging. The UAV landmine target detection method utilizes polarization intensity data acquired by a visible light polarization imaging module mounted on the UAV. The method includes acquiring the polarization intensity data corresponding to several polarization angles, calculating the corresponding Stokes vectors, further calculating three parameters characterizing the polarization state of the light wave, and inputting the degree of polarization, polarization angle, and elliptic angle as three-channel inputs into a deep neural network model for automated detection of surface landmine targets. This invention, based on visible light polarization imaging and combined with deep learning technology, enables the detection of surface landmine targets in complex backgrounds such as grasslands, tidal flats, and beaches, and can serve as an important tool in current integrated UAV mine detection systems.

[0069] Those skilled in the art should understand that variations can be implemented by combining existing technology with the above embodiments, which will not be elaborated here. Such variations do not affect the essence of the present invention, and will not be elaborated here either.

[0070] The preferred embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and the devices and structures not described in detail should be understood as being implemented in a conventional manner in the art. Any person skilled in the art can make many possible variations and modifications to the technical solutions of the present invention using the methods and techniques disclosed above, or modify them into equivalent embodiments with equivalent changes, without departing from the scope of the present invention. This does not affect the essential content of the present invention. Therefore, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the present invention's technical solutions still fall within the protection scope of the present invention.

[0071] Those skilled in the art will recognize that the units, i.e., algorithm steps, of the various examples described in connection with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

Claims

1. A method for detecting landmine targets using a UAV based on visible light polarization imaging, characterized in that, Based on the polarization intensity data acquired by the visible light polarization imaging module mounted on the UAV, the following steps are included: Step S101: Obtain the polarization intensity data corresponding to several polarization angles; Step S102: Calculate the corresponding Stokes vector and further calculate three parameters characterizing the polarization state of the light wave: degree of polarization, polarization angle, and elliptic angle; the equation of the Stokes vector is: , In the above formula, , , , These represent the light intensities of the linearly polarized components at 0°, 45°, 90°, and 135°, respectively. and These represent the light intensities of the right-hand circularly polarized and left-hand circularly polarized components of the light wave, respectively. Represents the total intensity of light waves. This represents the intensity difference of linearly polarized light in the horizontal (0°) and vertical (90°) directions. This represents the intensity difference of linearly polarized light in the 45° and 135° directions. This represents the intensity difference between right-handed and left-handed circularly polarized light; , It is distributed along The electric field vector in the direction, For the electric field vector in Phase difference in direction; polarization degree DoP, polarization angle AoP, and elliptic angle are obtained using the Stokes vector. The formula is: , , The degree of polarization expresses the proportion of fully polarized light in the total light intensity, while the polarization angle expresses the direction of vibration of the polarized light relative to the reference direction. The angle between the axes, the elliptic angle represents the ratio of the minor axis to the major axis of the polarization ellipse; Step S103: Polarization degree, polarization angle, and elliptic angle are used as three-channel inputs and fed into a deep neural network model for automated detection of landmine targets on the surface. The deep neural network model adopts a feature pyramid network-based basic structure, and a sampling receptive field enhancement module is added to the top layer of the feature pyramid network. At the neck of the deep neural network model, a spatial pyramid pooling method is used to separate the most important contextual features and increase the receptive field. A path aggregation structure is used to fuse features from different feature layers. An attention enhancement module is introduced into the detection head of the deep neural network model. The sampling receptive field enhancement module first passes through a 3×3 convolutional layer with a stride of 2, which reduces the feature map size while increasing the number of channels, so that subsequent convolutional kernels can obtain information from a larger spatial range. Then, it performs upsampling through simple nearest neighbor interpolation to restore the original size, and then uses a 1×1 convolutional layer to reduce the number of channels before stacking it with the feature map input to this module in terms of channel dimension. Finally, a 3×3 convolution is used to refine and fuse the features.

2. The method for detecting landmine targets by unmanned aerial vehicles based on visible light polarization imaging according to claim 1, characterized in that, The polarization intensity data of the polarization angle includes polarization intensity data of the scene and the target at four polarization angles: 0°, 45°, 90°, and 135°.

3. A UAV mine target detection system based on visible light polarization imaging, characterized in that, The method for detecting landmine targets by unmanned aerial vehicles as described in claim 1 includes: The data acquisition module acquires polarization intensity data at several polarization angles sent back by the UAV. The data preprocessing module calculates the Stokes vector and parameters characterizing the polarization state of light waves: degree of polarization, polarization angle, and elliptic angle. A deep neural network detector inputs the polarization degree, polarization angle, and elliptic angle into a deep neural network model to perform automated detection of landmine targets on the ground. The deep neural network model adopts a feature pyramid network-based basic structure, and adds a sampling receptive field enhancement module to the top layer of the feature pyramid network; in the neck of the deep neural network model, a spatial pyramid pool is used to separate the most important contextual features and increase the receptive field, and a path aggregation structure is used as a connection module to fuse features from different feature layers; an attention enhancement module is introduced into the detection head of the deep neural network model.

4. The UAV mine target detection system based on visible light polarization imaging according to claim 3, characterized in that, The UAV includes an onboard visible light polarization imaging detection module, which includes a visible light polarization camera system and an onboard microcomputer. The onboard microcomputer adjusts and automatically controls the imaging parameters of the polarization camera, and acquires and stores the corresponding polarization data.

5. The UAV mine target detection system based on visible light polarization imaging according to claim 4, characterized in that, The data acquisition module is located on the ground and acquires polarization intensity data of several polarization angles stored in the airborne microcomputer.

6. A UAV mine target detection system based on visible light polarization imaging according to claim 3 or 4, characterized in that, The feature pyramid network has a 5-layer structure.

7. A UAV mine target detection system based on visible light polarization imaging according to claim 3 or 4, characterized in that, The attention enhancement module dynamically learns weights based on the importance of different input features and calculates the weight information using the following formula: in, It is the attention-weighted feature map. These are the input feature maps. It is the attention weight calculation subnetwork. These are the parameters of the sub-network. This indicates element-wise multiplication.