A spectral imaging system based on adaptive super-resolution

Abstract

The present invention proposes a spectral imaging system based on adaptive super-resolution, comprising a coaxial co-aperture objective lens, a collimator, a dichroic beam splitter, a coding template, a hyperspectral imaging component, a piezoelectric ceramic micro-scanning super-resolution device, and a multispectral detection component. The coaxial co-aperture objective lens converges target light onto the collimator, which shapes it into a parallel beam. The dichroic beam splitter splits the parallel beam into visible / near-infrared branch paths and mid-infrared branch paths. The coding template and hyperspectral imaging component are positioned on the visible / near-infrared branch paths. The piezoelectric ceramic micro-scanning super-resolution device and the multispectral detection component are positioned on the mid-infrared branch paths. This system utilizes a self-feedback mechanism to perform adaptive infrared spatial super-resolution imaging and visible spectrum super-resolution imaging, thereby ensuring both detection capability and data volume, thus ensuring real-time data processing.

Description

Technical Field

[0001] The present invention relates to the technical field of spectral imaging, and in particular to a spectral imaging system based on adaptive super-resolution. Background Art

[0002] Currently, infrared image super-resolution technology mainly focuses on how to use deep learning or multi-frame fusion technology to improve the ability to restore image details. In the "Infrared Image Super-Resolution Reconstruction Method Based on Deep Learning", Li Yan et al. used a multi-layer convolutional neural network and feature fusion strategy to perform end-to-end training on a large number of low-resolution infrared images to learn to restore image details and texture information. However, because this method relies on a large data set and offline training, its model has weak adaptive adjustment capabilities in real-time scenarios and cannot meet the high timeliness requirements in dynamic environments. Patent application number CN2024113011084, "Infrared Image Super-Resolution Reconstruction Method, Device, Equipment, Storage Medium and Process", adopts a strategy based on multi-frame image fusion and motion compensation to improve the spatial resolution of the image through precise image registration and offline reconstruction. However, this solution also focuses on offline processing of pre-collected images and lacks dynamic adjustment capabilities for real-time target changes. In the field of hyperspectral super-resolution, Xu et al. proposed an offline super-resolution reconstruction method based on generative adversarial networks in their paper "Hyperspectral Image Super-Resolution Using a Generative Adversarial Network" published in the IEEE Access 2019 journal. The method uses a deep network to restore hyperspectral image details. However, this method does not combine spectral encoding with real-time acquisition process, making it difficult to take into account both spatial and spectral resolution in dynamic scenes.

[0003] Most existing methods use machine learning to generate data with higher resolution and more spectral dimensions. Their authenticity is questionable, and there is a contradiction between the extremely large amount of data and the limited real-time processing capabilities.

[0004] In light of this, the present invention proposes a spectral imaging system based on adaptive super-resolution. In contrast, this system utilizes intelligent automatic feedback control to achieve self-feedback acquisition of high-spatial-resolution and high-spectral-dimensional data on targets of interest. This system employs a coarse detection mode to reduce data volume. Upon detection of a suspicious target, the system utilizes a self-feedback mechanism to perform adaptive infrared spatial super-resolution imaging and visible spectrum super-resolution imaging. This ensures both detection capability and reduced data volume, ensuring real-time data processing. Summary of the Invention

[0005] The purpose of the present invention is to propose a spectral imaging system based on adaptive super-resolution; using a self-feedback mechanism, adaptive infrared spatial super-resolution imaging and visible spectrum super-resolution imaging are performed, thereby ensuring detection capabilities while reducing the amount of data and ensuring real-time data processing.

[0006] To achieve the above object, the technical solution of the present invention is as follows:

[0007] A spectral imaging system based on adaptive super-resolution, comprising a coaxial common aperture objective lens, a collimating lens, a dichroic beam splitting element, a coding template, a hyperspectral imaging component, a piezoelectric ceramic micro-scanning super-resolution device, and a multispectral detection component;

[0008] The optical axes of the coaxial common-aperture objective lens and the collimator lens coincide with each other; the coaxial common-aperture objective lens converges the target light onto the collimator lens, which then shapes the target light into a parallel light beam; the dichroic beam splitting element splits the parallel light beam into a visible light / near-infrared light beam and a mid-infrared light beam according to a preset splitting ratio, thereby forming a visible light / near-infrared light branching optical path and a mid-infrared light branching optical path;

[0009] The coding template and hyperspectral imaging component are sequentially arranged on the visible light / near-infrared light branch optical path along the propagation direction of the visible light / near-infrared light beam; the coding template performs pixel-level modulation on the visible light / near-infrared light beam through a micromirror array to improve the spectral signal-to-noise ratio and resolution; the hyperspectral imaging component includes a dispersive element and a detector. After the coded visible light / near-infrared light beam is separated into each band spectrum by the dispersive element, a coded hyperspectral image is generated on the detector;

[0010] The piezoelectric ceramic micro-scanning super-resolution device and the multi-spectral detection component are sequentially arranged on the mid-infrared beam branching optical path along the propagation direction of the mid-infrared beam; the multi-spectral detection component includes a multi-path splitter and a multi-channel detector. The multi-path splitter divides the bandwidth of the beam passing through the piezoelectric ceramic micro-scanning super-resolution device into multiple narrow-band channels, which then enter the multi-channel detector to complete the synchronous acquisition of multi-spectral infrared images.

[0011] Preferably, the preset splitting ratio is 50:50.

[0012] Preferably, the dispersive element comprises a prism or a grating.

[0013] Preferably, the multi-channel optical splitting device includes a multi-channel bandpass filter or a multi-prism optical splitter.

[0014] Preferably, the multi-way optical splitter and the multi-channel detector are respectively a four-way optical splitter and a four-channel detector.

[0015] Preferably, it also includes an adjustable optical bracket, and the coaxial common aperture objective lens, collimating mirror, dichroic beam splitting element, coding template, hyperspectral imaging component, piezoelectric ceramic micro-scanning super-resolution device and multi-spectral detection component are arranged on the adjustable optical bracket for azimuth adjustment.

[0016] Preferably, it further includes an analysis and control module, which is connected to the hyperspectral imaging component, the piezoelectric ceramic micro-scanning super-resolution device and the multispectral detection component respectively; the working mode of the analysis and control module is as follows:

[0017] S1. Execute infrared detection mode and control the piezoelectric ceramic micro-scanning super-resolution device to be in a static state; for the same target scene, perform image acquisition through the mid-infrared beam branching optical path to obtain L frames of multispectral infrared images, where L represents the number of splitting paths of the multi-path splitting device;

[0018] S2. Acquire L frames of multispectral infrared images in infrared detection mode and perform target detection;

[0019] S3. If the confidence level of the target detection result is lower than the first confidence threshold, the hyperspectral and super-resolution infrared collaborative detection mode is started, and the image acquisition and detection of the current target scene are performed again; if the confidence level of the target detection result is between the first confidence threshold and the second confidence threshold, the infrared super-resolution mode is started, and the image acquisition and detection of the current target scene are performed again; if the confidence level of the target detection result is higher than the second confidence threshold, the target detection result is output; wherein the first confidence threshold is less than the second confidence threshold.

[0020] Preferably, the hyperspectral and ultra-resolution infrared collaborative detection mode is started, and image acquisition and detection of the current target scene are performed again; specifically as follows:

[0021] Initiating a hyperspectral and super-resolution infrared collaborative detection mode includes controlling a piezoelectric ceramic micro-scanning super-resolution device to be in an operating state; in the operating state, a vibrating component of the piezoelectric ceramic micro-scanning super-resolution device operates and causes a displacement of an output optical path, causing pixels on the multi-channel detector to shift at a sub-pixel level according to a preset trajectory between each frame;

[0022] For the current target scene: continuously collect L×N frames of multispectral infrared images with displacement information through the mid-infrared beam branch optical path, where N represents the number of multispectral infrared image frames continuously collected for each channel; and synchronously continuously collect M frames of encoded hyperspectral images through the visible light / near-infrared light branch optical path, where M represents the number of encoded hyperspectral image frames continuously collected;

[0023] Obtain L×N frames of multispectral infrared images with displacement information, and fuse the multispectral infrared images of the same channel to obtain a fused multispectral infrared image; obtain M frames of encoded hyperspectral images and decode them to obtain a super-resolution spectral image;

[0024] Target detection is performed based on the fusion of multispectral infrared images and super-resolution spectral images, and the target detection results are obtained and output.

[0025] Preferably, the infrared super-resolution mode is started, and image acquisition and detection of the current target scene are performed again; specifically as follows:

[0026] Starting the infrared super-resolution mode includes controlling the piezoelectric ceramic micro-scanning super-resolution device to be in an operating state; in the operating state, a vibrating component of the piezoelectric ceramic micro-scanning super-resolution device operates and causes a displacement of the output optical path, so that pixels on the multi-channel detector are shifted at a sub-pixel level according to a preset trajectory between each frame;

[0027] For the current target scene, L×N frames of multispectral infrared images with displacement information are continuously collected through the mid-infrared beam branching optical path, where N represents the number of multispectral infrared image frames continuously collected by each channel;

[0028] Obtain L×N frames of multispectral infrared images with displacement information, and fuse the multispectral infrared images of the same channel to obtain a fused multispectral infrared image;

[0029] Target detection is performed based on the fused multispectral infrared image, and the target detection results are obtained and output.

[0030] Preferably, the first confidence threshold is set to 30%; the second confidence threshold is set to 80%.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] The present invention achieves collaborative optimization in various links such as acquisition, encoding, reconstruction, feedback control and data fusion: (1) High imaging accuracy and rich detail restoration: Utilizing high-speed micro-scanning motion components, sub-pixel-level multi-frame acquisition is achieved, and then through multi-frame fusion reconstruction technology, the resolution limitation of traditional infrared detectors is greatly broken through, and super-resolution infrared images with richer details are generated, maintaining the authenticity of physical super-resolution; (2) Fine spectral information acquisition: The incident light is modulated by a coding template, and the effective encoding of hyperspectral data is completed through a micromirror array, which effectively suppresses the noise introduced by the encoding process and improves the signal-to-noise ratio and resolution of the reconstructed spectrum; (3) Real-time adaptive control: Real-time monitoring of image reconstruction effects and dynamic adjustment of micro-scanning parameters enable the system to maintain the best working state in different light fields and detection environments, meeting the needs of dynamic and changing applications; (4) Compact, lightweight and low power consumption: Utilizing a wide-spectrum common aperture optical solution, sharing the same acquisition channel reduces the number of optical components, which not only simplifies the system structure, but also reduces the overall volume and power consumption, making it very suitable for use on drones, satellites and other portable platforms. (5) Comprehensive multispectral data processing capabilities: The system integrates infrared and visible / near-infrared band data, and provides sufficient data support for target detection, classification and accurate identification through back-end data fusion and multi-modal feature extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a system structure diagram of the present invention.

[0034] In the picture:

[0035] 1- Coaxial common aperture objective lens; 2- Collimating lens; 3- Dichroic beam splitter; 4- Coding template; 5- Hyperspectral imaging component; 6- Piezoelectric ceramic micro-scanning super-resolution device; 7- Multispectral detection component. DETAILED DESCRIPTION

[0036] The following is combined with Figure 1 , the technical solution of the present invention is described in detail.

[0037] The present invention proposes a spectral imaging system based on adaptive super-resolution, comprising a coaxial common aperture objective lens 1, a collimator 2, a dichroic beam splitter 3, a coding template 4, a hyperspectral imaging component 5, a piezoelectric ceramic micro-scanning super-resolution device 6, and a multispectral detection component 7;

[0038] The optical axes of the coaxial co-aperture objective lens 1 and the collimator lens 2 coincide with each other; the coaxial co-aperture objective lens 1 converges the target light onto the collimator lens 2, which then shapes the target light into a parallel light beam; the dichroic beam splitting element 3 splits the parallel light beam into a visible light / near-infrared light beam and a mid-infrared light beam according to a preset splitting ratio, thereby forming a visible light / near-infrared light branching optical path and a mid-infrared light branching optical path;

[0039] The coding template 4 and the hyperspectral imaging component 5 are sequentially arranged on the visible light / near-infrared light branch optical path along the propagation direction of the visible light / near-infrared light beam; the coding template 4 performs pixel-level modulation on the visible light / near-infrared light beam through a micromirror array to improve the spectral signal-to-noise ratio and resolution; the hyperspectral imaging component 5 includes a dispersive element and a detector. After the encoded visible light / near-infrared light beam is separated into various spectrum bands by the dispersive element, a coded hyperspectral image is generated on the detector;

[0040] The piezoelectric ceramic micro-scanning super-resolution device 6 and the multi-spectral detection component 7 are arranged in sequence on the mid-infrared beam branching optical path along the propagation direction of the mid-infrared beam; the multi-spectral detection component 7 includes a multi-way splitter device and a multi-channel detector. The multi-way splitter device divides the bandwidth of the light beam passing through the piezoelectric ceramic micro-scanning super-resolution device 6 into multiple narrow-band channels, and then enters the multi-channel detector to complete the synchronous acquisition of the multi-spectral infrared image.

[0041] In this embodiment, the preset splitting ratio is 50:50.

[0042] In this embodiment, the dispersion element includes a prism or a grating.

[0043] In this embodiment, the multi-channel optical splitting device includes a multi-channel bandpass filter or a multi-prism optical splitter.

[0044] In this embodiment, the multi-way optical splitter and the multi-channel detector are respectively a four-way optical splitter and a four-channel detector.

[0045] In this embodiment, an adjustable optical bracket is also included, and the coaxial common-aperture objective lens 1, collimating lens 2, dichroic spectrometer 3, coding template 4, hyperspectral imaging component 5, piezoelectric ceramic micro-scanning super-resolution device 6 and multispectral detection component 7 are arranged on the adjustable optical bracket for azimuth adjustment.

[0046] In this embodiment, an analysis and control module is further included, which is connected to the hyperspectral imaging component 5, the piezoelectric ceramic micro-scanning super-resolution device 6, and the multispectral detection component 7 respectively; the working mode of the analysis and control module is as follows:

[0047] S1. Execute infrared detection mode (coarse detection mode) and control the piezoelectric ceramic micro-scanning super-resolution device 6 to be in a static state; for the same target scene, perform image acquisition through the mid-infrared beam branching optical path to obtain L frames of multispectral infrared images, where L represents the number of splitting paths of the multi-path splitting device;

[0048] S2. Acquire L frames of multispectral infrared images in infrared detection mode and perform target detection;

[0049] S3. If the confidence level of the target detection result is lower than the first confidence threshold, the hyperspectral and super-resolution infrared collaborative detection mode is started, and the image acquisition and detection of the current target scene are performed again; if the confidence level of the target detection result is between the first confidence threshold and the second confidence threshold, the infrared super-resolution mode is started, and the image acquisition and detection of the current target scene are performed again; if the confidence level of the target detection result is higher than the second confidence threshold, the target detection result is output; wherein the first confidence threshold is less than the second confidence threshold.

[0050] In addition, the analysis and control module can also be connected to the encoding template 4 to dynamically adjust the encoding strategy according to the image reconstruction effect (image detection result); wherein, the encoding template 4 is a prior art, and the details are referred to patent application CN107687896A.

[0051] In this embodiment, the hyperspectral and ultra-resolution infrared collaborative detection mode is started, and image acquisition and detection of the current target scene are performed again; the details are as follows:

[0052] Initiating a hyperspectral and super-resolution infrared collaborative detection mode includes controlling the piezoelectric ceramic micro-scanning super-resolution device 6 to be in an operating state; in the operating state, a vibrating component of the piezoelectric ceramic micro-scanning super-resolution device 6 operates and causes a displacement of the output optical path, causing the pixels on the multi-channel detector to shift at a sub-pixel level according to a preset trajectory between each frame interval;

[0053] For the current target scene: continuously collect L×N frames of multispectral infrared images with displacement information through the mid-infrared beam branch optical path, where N represents the number of multispectral infrared image frames continuously collected for each channel; and synchronously continuously collect M frames of encoded hyperspectral images through the visible light / near-infrared light branch optical path, where M represents the number of encoded hyperspectral image frames continuously collected;

[0054] Obtain L×N frames of multispectral infrared images with displacement information, and fuse the multispectral infrared images of the same channel to obtain a fused multispectral infrared image; obtain M frames of encoded hyperspectral images and decode them to obtain a super-resolution spectral image;

[0055] Target detection is performed based on the fusion of multispectral infrared images and super-resolution spectral images, and the target detection results are obtained and output.

[0056] In this embodiment, the infrared super-resolution mode is started, and image acquisition and detection of the current target scene are performed again; specifically, as follows:

[0057] Starting the infrared super-resolution mode includes controlling the piezoelectric ceramic micro-scanning super-resolution device 6 to be in an operating state; in the operating state, the vibrating component of the piezoelectric ceramic micro-scanning super-resolution device 6 operates and causes the output optical path to shift, so that the pixels on the multi-channel detector are shifted at the sub-pixel level according to a preset trajectory between each frame;

[0058] For the current target scene, L×N frames of multispectral infrared images with displacement information are continuously collected through the mid-infrared beam branching optical path, where N represents the number of multispectral infrared image frames continuously collected by each channel;

[0059] Obtain L×N frames of multispectral infrared images with displacement information, and fuse the multispectral infrared images of the same channel to obtain a fused multispectral infrared image;

[0060] Target detection is performed based on the fused multispectral infrared image, and the target detection results are obtained and output.

[0061] In this embodiment, the first confidence threshold is set to 30%; the second confidence threshold is set to 80%.

[0062] In summary, this system achieves self-feedback acquisition of high-spatial-resolution, high-spectral-dimensional data on targets of interest through intelligent automatic feedback control. Furthermore, this system can be widely used in a variety of scenarios, including military reconnaissance and security, remote sensing and environmental monitoring, airborne platforms and unmanned systems, as well as industrial inspection and safety monitoring. Through a "coarse-first, fine-second" adaptive feedback operating mode, it rapidly scans and locates the target area over a large area, then automatically switches to infrared spatial super-resolution and visible spectrum super-resolution imaging, achieving real-time acquisition of high-spatial-resolution and high-spectral-dimensional data. This significantly reduces the amount of initial data while ensuring the precise detection and analysis of key targets.

[0063] The above are preferred embodiments of the present invention. Any changes made according to the technical solution of the present invention, as long as the resulting functions and effects do not exceed the scope of the technical solution of the present invention, shall fall within the scope of protection of the present invention.

Claims

1. A spectral imaging system based on adaptive super-resolution, characterized in that: It includes a coaxial common aperture objective lens (1), a collimating lens (2), a dichroic beam splitting element (3), a coding template (4), a hyperspectral imaging component (5), a piezoelectric ceramic micro-scanning super-resolution device (6), and a multi-spectral detection component (7); The optical axes of the coaxial common-aperture objective lens (1) and the collimator lens (2) coincide with each other; the coaxial common-aperture objective lens (1) converges the target light to the collimator lens (2), and the collimator lens (2) shapes the target light into a parallel light beam; the dichroic beam splitting element (3) splits the parallel light beam into a visible light / near-infrared light beam and a mid-infrared light beam according to a preset splitting ratio, thereby forming a visible light / near-infrared light branching optical path and a mid-infrared light beam branching optical path; The coding template (4) and the hyperspectral imaging component (5) are sequentially arranged on the visible light / near infrared light branch optical path along the propagation direction of the visible light / near infrared light beam; the coding template (4) performs pixel-level modulation on the visible light / near infrared light beam through a micromirror array to improve the spectral signal-to-noise ratio and resolution; the hyperspectral imaging component (5) includes a dispersion element and a detector, and after the coded visible light / near infrared light beam is separated into spectrums of each band by the dispersion element, a coded hyperspectral image is generated on the detector; The piezoelectric ceramic micro-scanning super-resolution device (6) and the multi-spectral detection component (7) are sequentially arranged on a mid-infrared light beam branching optical path along the propagation direction of the mid-infrared light beam; the multi-spectral detection component (7) comprises a multi-path optical splitter and a multi-channel detector, and the multi-path optical splitter divides the bandwidth of the light beam passing through the piezoelectric ceramic micro-scanning super-resolution device (6) into a plurality of narrow-band channels, which then enter the multi-channel detector to complete the synchronous acquisition of the multi-spectral infrared image; The system further comprises an analysis and control module, wherein the analysis and control module is connected to the hyperspectral imaging component (5), the piezoelectric ceramic micro-scanning super-resolution device (6) and the multispectral detection component (7) respectively; the working mode of the analysis and control module is as follows: S1, executing the infrared detection mode, controlling the piezoelectric ceramic micro-scanning super-resolution device (6) to be in a static state; for the same target scene, performing image acquisition through the mid-infrared light beam branching optical path, obtaining L frames of multi-spectral infrared images, where L represents the number of splitting paths of the multi-path splitting device; S2. Acquire L frames of multispectral infrared images in infrared detection mode and perform target detection; S3. If the confidence level of the target detection result is lower than the first confidence threshold, the hyperspectral and super-resolution infrared collaborative detection mode is activated, and image acquisition and detection of the current target scene are performed again; If the confidence level of the target detection result is between the first confidence threshold and the second confidence threshold, the infrared super-resolution mode is activated, and image acquisition and detection of the current target scene are performed again; If the confidence level of the target detection result is higher than the second confidence threshold, the target detection result is output; wherein the first confidence threshold is lower than the second confidence threshold.

2. The spectral imaging system based on adaptive super-resolution according to claim 1, characterized in that: The preset splitting ratio is 50:

50.

3. The spectral imaging system based on adaptive super-resolution according to claim 1, characterized in that: The dispersive element includes a prism or a grating.

4. The spectral imaging system based on adaptive super-resolution according to claim 1, characterized in that: The multi-path optical splitting device includes a multi-path bandpass filter or a multi-prism optical splitter.

5. The spectral imaging system based on adaptive super-resolution according to claim 1, characterized in that: The multi-way light splitting device and the multi-channel detector are respectively a four-way light splitting device and a four-channel detector.

6. The spectral imaging system based on adaptive super-resolution according to claim 1, characterized in that: The invention also includes an adjustable optical bracket, wherein the coaxial common aperture objective lens (1), the collimating lens (2), the dichroic beam splitting element (3), the coding template (4), the hyperspectral imaging component (5), the piezoelectric ceramic micro-scanning super-resolution device (6) and the multispectral detection component (7) are arranged on the adjustable optical bracket for azimuth adjustment.

7. The spectral imaging system based on adaptive super-resolution according to claim 1, characterized in that: The hyperspectral and super-resolution infrared collaborative detection mode is started, and image acquisition and detection of the current target scene are performed again; the details are as follows: Starting a hyperspectral and super-resolution infrared collaborative detection mode includes controlling the piezoelectric ceramic micro-scanning super-resolution device (6) to be in an operating state; in the operating state, the vibration component of the piezoelectric ceramic micro-scanning super-resolution device (6) operates and causes the output light path to be displaced, so that the pixel points on the multi-channel detector are shifted at the sub-pixel level according to a preset trajectory between each frame interval; For the current target scene: continuously collect L×N frames of multispectral infrared images with displacement information through the mid-infrared beam branch optical path, where N represents the number of multispectral infrared image frames continuously collected for each channel; and synchronously continuously collect M frames of encoded hyperspectral images through the visible light / near-infrared light branch optical path, where M represents the number of encoded hyperspectral image frames continuously collected; Obtain L×N frames of multispectral infrared images with displacement information, and fuse the multispectral infrared images of the same channel to obtain a fused multispectral infrared image; obtain M frames of encoded hyperspectral images and decode them to obtain a super-resolution spectral image; Target detection is performed based on the fusion of multispectral infrared images and super-resolution spectral images, and the target detection results are obtained and output.

8. The spectral imaging system based on adaptive super-resolution according to claim 1, characterized in that: The infrared super-resolution mode is started, and image acquisition and detection of the current target scene are performed again; the details are as follows: Starting the infrared super-resolution mode includes controlling the piezoelectric ceramic micro-scanning super-resolution device (6) to be in an operating state; in the operating state, the vibration component of the piezoelectric ceramic micro-scanning super-resolution device (6) operates and causes the output light path to be displaced, so that the pixel points on the multi-channel detector are shifted at the sub-pixel level according to a preset trajectory during each frame interval; For the current target scene, L×N frames of multispectral infrared images with displacement information are continuously collected through the mid-infrared beam branching optical path, where N represents the number of multispectral infrared image frames continuously collected for each channel; Obtain L×N frames of multispectral infrared images with displacement information, and fuse the multispectral infrared images of the same channel to obtain a fused multispectral infrared image; Target detection is performed based on the fused multispectral infrared image, and the target detection results are obtained and output.

9. The spectral imaging system based on adaptive super-resolution according to claim 1, characterized in that: The first confidence threshold is set to 30%; the second confidence threshold is set to 80%.

Images