Observation pod and true and false target image characteristic identification method based on dual-spectrum detection

Abstract

The invention relates to an observation pod and true and false target image characteristic identification method based on dual-spectrum detection. The observation pod comprises a shell, a power module, an information processing module, an infrared lens, a visible light lens, a data transmission module, a servo control system, a gyroscope, an azimuth motor and a pitching motor, the shell comprises a shell body and a bottom plate, the shell body is of a circular cover structure, two lens holes are formed in the shell body, and the infrared lens and the visible light lens are installed in the two lens holes respectively; the bottom plate is a flat plate and is provided with a reinforcing structure, the information processing module, the servo control system, the gyroscope, the azimuth motor and the pitching motor are arranged in the shell, the servo control system is connected with the gyroscope, the azimuth motor and the pitching motor, the information processing module is connected with the infrared lens, the visible light lens, the servo control system and the data transmission module, and the power supply module supplies power. The equipment can perform all-weather and multi-dimensional observation on a target, and the mass and the size of the equipment are reduced by taking a modular structure and a flat plate with a reinforcing structure as a bottom plate.

Description

Technical Field

[0001] The invention relates to the technical field of pods of small-sized flying equipment, and in particular to an observation pod based on dual-spectrum detection and a method for identifying characteristics of true and false target images. Background Art

[0002] The general observation equipment mainly consists of: aircraft platform, measurement and control and information transmission system, mission payload system (observation pod), ground integrated support system, navigation flight control system, etc. In order to ensure the accuracy of the observed target, the selection of the mission payload system (observation pod) is particularly critical. Commonly used mission payload systems include visible light sensors, infrared detectors, radar lasers, etc. Most small flight observation equipment on the market uses a single mission payload system, either visible light or infrared. Although a single payload can also achieve the observation of the target, it is inevitably limited by the single payload, such as visible light cannot observe the target at night, infrared cannot observe the specific information of the target, and the overall recognition accuracy is low. If multiple detection methods are integrated, the volume and / or weight will increase, which will make it difficult to apply to small flight equipment. Summary of the invention

[0003] The first object of the present invention is to provide an observation pod based on dual-spectrum detection, which can be suitable for small-scale flying equipment.

[0004] The second object of the present invention is to provide a method for identifying true and false target image characteristics using an observation pod based on dual-spectrum detection.

[0005] In order to achieve the above first object, in a first aspect, the present invention provides an observation pod based on dual-spectrum detection, characterized in that it includes a housing, a power module, an information processing module, an infrared lens, a visible light lens, a data transmission module, a servo control system, a gyroscope, an azimuth motor and a pitch motor;

[0006] The outer shell is rotatably mounted on the pan / tilt frame, and comprises a shell body and a bottom plate. The shell body is a circular cover structure, and two lens holes are arranged on the shell body. The bottom plate is a flat plate, and a reinforcement structure is arranged on the flat plate.

[0007] The infrared lens and the visible light lens are both arranged in the housing and are respectively installed at a lens hole;

[0008] The servo control system, the gyro, the azimuth motor and the pitch motor are all arranged in the housing and mounted on the bottom plate, and the servo control system is connected with the gyro, the azimuth motor and the pitch motor;

[0009] The information processing module is arranged in the housing, mounted on the bottom plate, and connected with the infrared lens, the visible light lens, the servo control system and the data transmission module;

[0010] The power module is used to provide power.

[0011] Optionally, the reinforcement structure is a reinforcement rib or convex and concave portions arranged at intervals.

[0012] Optionally, the infrared lens and the visible light lens have parallel optical axes.

[0013] Optionally, the layout of the power module, information processing module, infrared lens, visible light lens, data transmission module, servo control system, gyroscope, azimuth motor and pitch motor satisfies the requirement of minimizing the distance between each part and the housing.

[0014] Optionally, a counterweight is also provided in the shell.

[0015] In order to achieve the above second objective, in a second aspect, the present invention further provides a method for identifying true and false target image characteristics, using an observation pod as in any implementation of the first aspect to perform true and false target image characteristics identification, comprising the following steps:

[0016] Step 100: Obtain visible and infrared original images of the real target, fix the image size, and input them into the deep convolutional neural network model;

[0017] Step 101: extracting visible light and infrared image feature information respectively through a feature extraction network, and classifying the feature information according to categories and batches;

[0018] Step 102: Using a feature fusion network to fuse shallow image location information and deep image semantics, the algorithm can dynamically adjust the depth and range of feature extraction according to the size, shape and complexity of the target;

[0019] Step 103: Analyze the category and environment of the detection target based on the rich image feature information, and adaptively select and call different loss function models according to the different targets and environments;

[0020] Step 104: Calculate the difference between the target to be measured and the real target in the library using the loss function, and then analyze and judge the authenticity of the target by combining the image feature parameters of the target to be measured and the feature curve model of the real target.

[0021] Optionally, the network model and the feature extraction model are continuously trained, the loss function is gradient descended, and the model is iteratively updated to reduce the category loss and the bounding box loss.

[0022] The above technical solution of the present invention has the following advantages:

[0023] The observation pod based on dual-spectrum detection provided by the present invention includes a shell, a power module, an information processing module, an infrared lens, a visible light lens, a data transmission module, a servo control system, a gyroscope, an azimuth motor and a pitch motor. The shell is rotatably mounted on a pan-tilt frame, and the shell includes a shell and a bottom plate. The shell is a circular cover structure, and two lens holes are provided on the shell. The bottom plate is a flat plate, and a reinforcement structure is provided on the flat plate. The infrared lens and the visible light lens are both arranged in the shell, and are respectively installed at one lens hole. The servo control system, the gyroscope, the azimuth motor and the pitch motor are all arranged in the shell and installed on the bottom plate. The servo control system is connected with the gyroscope, the azimuth motor and the pitch motor. The information processing module is arranged in the shell, installed on the bottom plate, and connected with the infrared lens, the visible light lens, the servo control system and the data transmission module. The power module is used to provide power. The device can observe the target in all weather and multi-dimensionally. By using a modular structure and a flat plate with a reinforcement structure as a bottom plate, the quality and volume of the device are greatly reduced without affecting the flexibility and reliability of the device.

[0024] The method for identifying true and false target image characteristics provided by the present invention uses an algorithm to compare and analyze the targets observed by the observation crane, comprehensively analyze and judge the authenticity of the targets. For the image data collected by the observation crane, the characteristics of the targets are analyzed from the visible and infrared angles respectively, and the authenticity of the targets is distinguished by the difference in the image characteristic curves of the true and false targets. The target feature algorithm combines adaptive and dynamic feature fusion technology. It can provide more fine-grained features for small targets and allocate more global features for large targets by dynamically selecting the network depth and width. This method greatly improves the accuracy and efficiency of judging true and false targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The drawings of the present invention are provided for illustrative purposes only, and the proportions and quantities of the components in the drawings may not necessarily be consistent with the actual products.

[0026] Figure 1 is a structural schematic diagram of an observation pod based on dual-spectrum detection in an embodiment of the present invention;

[0027] Figure 2 It is a block diagram of the structural connection relationship of an observation pod based on dual-spectrum detection in an embodiment of the present invention;

[0028] Figure 3 It is a flow chart of a method for identifying true and false target image characteristics in an embodiment of the present invention.

[0029] In the figure:

[0030] 1: PTZ stand;

[0031] 2: Observation pod;

[0032] 21: Shell;

[0033] 211: Shell;

[0034] 212: bottom plate;

[0035] 22: Information processing module;

[0036] 23: Infrared lens;

[0037] 24: Visible light lens;

[0038] 25: data transmission module;

[0039] 26: Servo control system;

[0040] 27: Gyro;

[0041] 28: Azimuth motor;

[0042] 29: Pitch motor. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments 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.

[0044] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0045] like Figure 1 and Figure 2 As shown, the observation pod 2 based on dual-spectrum detection provided by an embodiment of the present invention includes a shell 21, a power module, an information processing module 22, an infrared lens 23, a visible light lens 24, a data transmission module 25, a servo control system 26, a gyroscope 27, an azimuth motor 28 and a pitch motor 29.

[0046] Among them, the outer shell 21 is rotatably installed on the pan-tilt stand 1, and the outer shell 21 includes a shell body 211 and a bottom plate 212. The shell body 211 is a circular cover structure, and two lens holes are arranged on the shell body 211. The bottom plate 212 is a flat plate, and a reinforcement structure is arranged on the flat plate.

[0047] The infrared lens 23 and the visible light lens 24 are both disposed in the housing 21 and are installed at a lens hole respectively.

[0048] The servo control system 26 , the gyroscope 27 , the azimuth motor 28 and the pitch motor 29 are all arranged in the housing 21 and mounted on the bottom plate 212 . The servo control system 26 is connected to the gyroscope 27 , the azimuth motor 28 and the pitch motor 29 .

[0049] The information processing module 22 is disposed in the housing 21 , mounted on the bottom plate 212 , and connected to the infrared lens 23 , the visible light lens 24 , the servo control system 26 , and the data transmission module 25 .

[0050] The data module 25 is used to transmit the image and video stream data of the infrared lens 23 and the visible light lens 24 to the ground detection platform. The information processing module 22 is the core module. Each command received is transmitted to each load by the information processing module 22, and the feedback information of each load is uploaded to the host computer. The power module is used to provide power to each power load of the observation pod 2.

[0051] When in use, the observation pod 2 is mounted on a small flying device (such as a drone). The observation pod 2 uses visible light and infrared payloads to observe a target at the same time. It can have the advantages of both payloads. The visible light has high resolution, captures rich target information, and has strong anti-interference ability. The infrared has a thermal imaging function and can work all-weather, adapt to nighttime or low visibility conditions, and has strong resistance to electromagnetic interference. The two complement each other, so that the device can observe the target all-weather and in multiple dimensions. Through the modular structure, the selection of each module can start with the selection of lightweight components as much as possible to achieve the overall lightweight of the device. In addition, the redundant part of the lower sphere is removed and designed into a plane with a reinforced structure. Without affecting the flexibility and reliability of the equipment, the weight and volume of the equipment are greatly reduced.

[0052] In the example of this embodiment, the reinforcement structure may be a reinforcement rib or convex and concave portions arranged at intervals.

[0053] In this embodiment, the infrared lens 23 and the visible light lens 24 both have parallel optical axes.

[0054] In this embodiment, preferably, the layout of the power module, information processing module 22, infrared lens 23, visible light lens 24, data transmission module 25, servo control system 26, gyroscope 27, azimuth motor 28 and pitch motor 29 satisfies the minimum distance between each part and the housing. That is, the assembly of the device adopts the method of seamlessly fitting the functional module and the housing, which can reduce the size of the device and increase the heat dissipation function of each functional module.

[0055] In one example, a counterweight is further provided in the housing 21 as required.

[0056] See also Figure 3 As shown, this embodiment also provides a method for identifying true and false target image characteristics, using any of the observation pods in the above embodiments to identify true and false target image characteristics, including the following steps:

[0057] Step 100: Obtain visible and infrared original images of the real target, fix the image size, and input them into the deep convolutional neural network model;

[0058] Step 101: extract visible light and infrared image feature information respectively through a feature extraction network, and classify the feature information according to categories and batches.

[0059] Step 102: Use the feature fusion network to fuse the shallow position information and deep semantics of the image, so that the algorithm can dynamically adjust the depth and range of feature extraction according to the size, shape and complexity of the target.

[0060] Step 103: Analyze the category and environment of the detection target based on the rich image feature information, and adaptively select and call different loss function models according to the different targets and environments.

[0061] Step 104: Calculate the difference between the target to be measured and the real target in the library using the loss function, and then analyze and judge the authenticity of the target by combining the image feature parameters of the target to be measured and the feature curve model of the real target.

[0062] This method for identifying true and false target image characteristics can use algorithms to compare and analyze targets observed by observation cranes, comprehensively analyze and judge the authenticity of targets. For the image data collected by observation cranes, the characteristics of targets are analyzed from visible and infrared angles respectively, and the authenticity of targets is distinguished by the difference in the image characteristic curves of true and false targets. The target feature algorithm combines adaptive and dynamic feature fusion technology. It can provide more fine-grained features for small targets and assign more global features to large targets by dynamically selecting the depth and width of the network. This method greatly improves the accuracy and efficiency of judging true and false targets.

[0063] It should be noted that the various algorithms mentioned in the above method are existing algorithms and will not be described in detail here.

[0064] In order to further improve the reliability of the algorithm model and the accuracy of target authenticity judgment, preferably, the network model and feature extraction model are continuously trained, the loss function is gradient descended, the weight parameters of the convolution, full connection and other layers in the model are iteratively updated, the category loss and bounding box loss are continuously reduced, and the reliability of the algorithm model and the accuracy of target authenticity judgment are improved.

[0065] 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 not every embodiment contains only one independent technical solution. In the absence of a conflict of solutions, the various technical features mentioned in each embodiment can be combined in any manner to form other implementation methods that can be understood by those skilled in the art.

[0066] In addition, without departing from the scope of the present invention, the technical solutions described in the aforementioned embodiments may be modified, or some of the technical features may be replaced by equivalents, without causing the essence of the corresponding technical solutions to depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An observation pod based on dual-spectrum detection, characterized in that: It includes a housing, a power module, an information processing module, an infrared lens, a visible light lens, a data transmission module, a servo control system, a gyroscope, an azimuth motor and a pitch motor; The housing is rotatably mounted on the pan / tilt stand, and comprises a shell and a bottom plate. The shell is a circular cover structure, and two lens holes are arranged on the shell. The bottom plate is a flat plate, and a reinforcement structure is arranged on the flat plate. The infrared lens and the visible light lens are both arranged in the housing and are respectively installed at one of the lens holes; The servo control system, gyroscope, azimuth motor and pitch motor are all arranged in the housing and mounted on the bottom plate, and the servo control system is connected with the gyroscope, azimuth motor and pitch motor; The information processing module is arranged in the housing, mounted on the bottom plate, and connected with the infrared lens, the visible light lens, the servo control system and the data transmission module; The power module is used to provide power.

2. The observation pod according to claim 1, characterized in that: The reinforcement structure is a reinforcement rib or convex and concave parts arranged at intervals.

3. The observation pod according to claim 1, characterized in that: The infrared lens and the visible light lens both have parallel optical axes.

4. The observation pod according to claim 1, characterized in that: The layout of the power module, information processing module, infrared lens, visible light lens, data transmission module, servo control system, gyroscope, azimuth motor and pitch motor is such that the distance between each part and the housing is minimized.

5. The observation pod according to claim 1, characterized in that: A counterweight is also arranged in the shell.

6. A method for identifying true and false target image characteristics, characterized in that: Using the observation pod as described in any one of claims 1 to 5 to perform true and false target image feature recognition, the following steps are included: Step 100: Obtain visible and infrared original images of the real target, fix the image size, and input them into the deep convolutional neural network model; Step 101: extracting visible light and infrared image feature information respectively through a feature extraction network, and classifying the feature information according to categories and batches; Step 102: Using a feature fusion network to fuse shallow image location information and deep image semantics, the algorithm can dynamically adjust the depth and range of feature extraction according to the size, shape and complexity of the target; Step 103: Analyze the category and environment of the detection target based on the rich image feature information, and adaptively select and call different loss function models according to the different targets and environments; Step 104: Calculate the difference between the target to be measured and the real target in the library using the loss function, and then analyze and judge the authenticity of the target by combining the image feature parameters of the target to be measured and the feature curve model of the real target.

7. The method for identifying true and false target image characteristics according to claim 6, characterized in that: The network model and feature extraction model are continuously trained, the gradient descent is performed on the loss function, and the model is iteratively updated to reduce the category loss and bounding box loss.

Images