Simulation method for performance evaluation of infrared imaging system based on parallel computing

Abstract

The application discloses an infrared imaging system performance evaluation simulation method based on parallel computing, mainly solves the problems of long time consumption and high cost in the prior art performance evaluation, and an implementation scheme is as follows: acquiring various parameters in the infrared imaging system and GPU configuration of a simulation machine; respectively establishing kernel function families of a modulation transfer function (MTF) and a contrast threshold function (CTF) of the infrared imaging system according to the acquired parameters; a host end performs logical control in performance evaluation and transmits tasks to a device end; the device end runs the kernel function families and parallelly calculates data thereof, and manages and identifies the calculation results; the host end processes the data according to the calculation results and the identification of the device end; a detection probability of the infrared imaging system is calculated by using the processing results, a curve is drawn, corresponding data is given in the form of a report, and the evaluation simulation of the system is completed. The application reduces the evaluation time consumption and the cost, and can be used for infrared imaging system performance evaluation and test verification.

Description

Technical Field

[0001] This invention belongs to the field of infrared imaging system technology, and specifically relates to a simulation method for performance evaluation of infrared imaging systems, which can be used for performance evaluation and testing verification of infrared imaging systems. Background Technology

[0002] With the development of infrared technology, infrared imaging systems are becoming increasingly sophisticated, and the performance of optoelectronic imaging systems is constantly improving, placing higher demands on the evaluation of system imaging performance. Traditional experimental methods and hardware-in-the-loop (HIL) simulations can accurately, reliably, and comprehensively reflect measurement results; however, these two methods have many disadvantages in practical applications. For example, they cannot cope with complex and changing external environments; field tests require significant manpower, material resources, and financial investment; the development cycle is long; and they can only acquire a limited amount of experimental data under specific scene characteristics, time, and weather conditions, failing to meet the practical performance testing requirements of infrared systems across all bands, all weather conditions, and at different distances and angles. More importantly, experimental methods can only evaluate the performance of established imaging systems and cannot perform performance prediction, design parameter adjustments, or analysis for systems that are not yet finalized.

[0003] The performance theoretical modeling method analyzes the physical characteristics of each component of the optoelectronic imaging system based on its overall performance parameters, establishes a physical model corresponding to the system, and utilizes software simulation to fully simulate the optoelectronic imaging system and various complex environments such as targets and backgrounds. This allows for accurate evaluation of the infrared imaging system's performance. The theoretical modeling method can obtain performance evaluation data of the infrared system's imaging at different distances and angles under all weather conditions through computer simulation, and also facilitates the optimized design of the optoelectronic system. By using simulation methods, research expenses are significantly reduced, and the arduous fieldwork of laboratory personnel is avoided, thereby shortening the design cycle and improving work efficiency.

[0004] Traditional approaches focus on the imaging process of infrared imaging systems, simulating the effects of optical systems, detectors, and circuitry to build models and obtain performance evaluation data through extensive computation. In recent years, with the continuous improvement of optoelectronic imaging system performance, higher demands have been placed on system imaging performance evaluation. The scenarios requiring evaluation are becoming increasingly complex, and with the improvement of performance evaluation models, the computational load is also increasing, especially for large-scale scenarios where performance evaluation struggles to meet requirements. Currently, the efficiency of simulation evaluation is mainly improved by enhancing the hardware performance of the simulation platform. However, upgrading the hardware platform requires additional financial resources and is difficult to meet the complex and flexible application scenarios, thus presenting significant limitations. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of the prior art by proposing a simulation method for performance evaluation of infrared imaging systems based on parallel computing. This method aims to reduce the time and financial costs associated with performance evaluation, improve the efficiency of infrared imaging system performance evaluation, and enable the system to adapt to complex and flexible application scenarios.

[0006] To achieve the above objectives, the technical solution of the present invention includes the following steps:

[0007] (1) Obtain the optical system parameters, detector parameters and circuit system parameters in the infrared imaging system, and use the API interface to obtain the GPU configuration information of the simulated machine;

[0008] (2) Establish the kernel function family of the modulation transfer function (MTF):

[0009] (2a) Based on the parameters obtained in (1), a theoretical model of the overall modulation transfer function (MTF) of the infrared imaging system is established through sampling:

[0010] MTF sys =MTF tur (f)·MTF opt (f)·MTF det (f)·MTF disp (f)·MTF eye (f)

[0011] Among them, MTF sys Let MTF be the overall modulation transfer function of the infrared imaging system. tur (f) represents the atmospheric turbulence MTF before sampling. opt (f) is the MTF of the optical system before sampling, which includes aberration, defocus, and diffraction.

[0012] MTF det (f) is the modulation transfer function of the detector before sampling, MTF disp (f) is the modulation transfer function of the display after sampling, MTF. eye (f) is the modulation transfer function of the human eye after sampling;

[0013] (2b) Based on the modulation transfer function model established in step (2a), a family of kernel functions for the MTF is established, which includes the family of atmospheric turbulence kernel functions, the family of optical system kernel functions, the family of detector kernel functions, the family of display kernel functions, and the family of human eye kernel functions.

[0014] (3) Establish the kernel function family for the system contrast threshold function CTF:

[0015] (3a) Based on the modulation transfer function (MTF) theoretical model, and considering the system's influence on the naked-eye contrast threshold function, establish the system's contrast threshold function (CTF) theoretical model:

[0016]

[0017] Among them, CTF nake M is the naked-eye contrast threshold function CTF. dsp NF(f) is the glare factor of the display, NF(f) is the noise factor, and f is the spatial frequency. H represents the system amplification. disp H represents the display pixel size. pix L is the detector pixel size. focus L is the system focal length. eye The distance from the human eye to the monitor;

[0018] (3b) Based on the system contrast function model, establish the kernel function family of the contrast function model CTF, which includes the naked-eye CTF kernel function family, the display glare factor kernel function family, and the noise factor kernel function family.

[0019] (4) Introduce a parallel computing model, establish a host side and a device side, and perform task scheduling, distribution and parallel data computation respectively:

[0020] (4a) Considering that CPUs are suitable for control-intensive tasks and GPUs are suitable for data-intensive tasks, a parallel computing model based on the unified computing device architecture CUDA is introduced, establishing host-CPU on the host side and device-GPU on the device side; and grid and block are established according to the architectural characteristics of CUDA.

[0021] (4b) The device manages the memory in the CUDA programming model, that is, it puts the data that needs to be accessed frequently in the model calculation into shared memory, puts ordinary data into registers, and merges the access to global data during the kernel function processing.

[0022] (4c) The host performs task scheduling and distribution:

[0023] The host executes the logic control involved in the performance evaluation and transmits the tasks involving a large number of floating-point calculations, such as vectors, matrices, and integrals, to the device.

[0024] On the device side, the kernel function family of the Modulation Transfer Function (MTF) and Comparison Threshold Function (CTF) models is run in different grids and thread blocks to compute MTF and CTF data in parallel, and these functions are managed and their validity is identified through grids and thread blocks.

[0025] The host receives the calculation results from the device, organizes and merges the data according to the device's identifier, and obtains the MTF data. sys CTF datasys ;

[0026] (5) Based on the CTF data obtained in (4), calculate the detection probability of the infrared imaging system according to the TTP criterion;

[0027] (6) Based on the current scene settings, repeat steps (4b), (4c), and (5) until the detection probability set under the current scene settings is obtained. At the same time, plot the probability set curve and give the corresponding data in the form of a report to complete the performance evaluation simulation of the infrared imaging system.

[0028] Compared with the prior art, the present invention has the following advantages.

[0029] First, this invention introduces a parallel computing architecture. By analyzing the characteristics of CPUs and GPUs, a CUDA-based parallel computing model is established. The host computer schedules and distributes tasks, and tasks involving a large number of floating-point calculations such as vectors, matrices, and integrals are handed over to the device computer for execution. The host computer receives the calculation results from the device computer while mainly performing logic control and task distribution in the performance evaluation, which greatly improves the computing performance and can meet the performance requirements of high-precision simulation in complex scenarios.

[0030] Secondly, this invention proposes a parallel computing method. By establishing a host end and a device end, and creating corresponding kernel function families based on the modulation transfer function model and the contrast threshold function model, the evaluation task is divided into different sub-tasks. Different tasks are completed in their respective kernel functions, which changes the data flow and processing flow. At the same time, the number of grids and the number of thread blocks are used to specify the number of threads to be executed on the device end. Based on the existing hardware performance, an operating scheme that matches the hardware performance is selected, providing a flexible solution for the performance evaluation of actual infrared imaging systems, which can cope with complex and ever-changing application scenarios.

[0031] Third, due to its advantages of high performance and high flexibility, the present invention can maximize resource utilization, thereby reducing evaluation costs and achieving high economic benefits. Attached Figure Description

[0032] Figure 1 This is a flowchart illustrating the implementation of the present invention;

[0033] Figure 2 This is a schematic diagram illustrating the interaction between each thread block in the host and device sides of this invention. Detailed implementation method:

[0034] The embodiments and effects of the present invention will be further described in detail below with reference to the accompanying drawings.

[0035] Reference Figure 1 The implementation steps for this example are as follows:

[0036] Step 1: Obtain the parameters of the infrared imaging system and the GPU configuration information of the simulated machine.

[0037] 1.1) Obtain various parameters of the infrared imaging system

[0038] The parameters of an infrared imaging system include optical system parameters, detector parameters, and circuit system parameters.

[0039] This step involves obtaining various parameters of the infrared imaging system through experiments or instrument measurements, including:

[0040] The optical system parameters include center wavelength, circular aperture, square aperture width, square aperture length, focal length, average transmittance, blocking ratio, wavefront aberration root mean square, defocusing amount, and optical system temperature.

[0041] The detector parameters include horizontal spacing, vertical spacing, number of horizontal detectors, number of vertical detectors, readout time, horizontal size, vertical size, ideal potential well fill rate, potential well capacity (number of electrons), spectral quantum efficiency, conversion gain, number of field samples per instant, and frame rate.

[0042] The circuit system parameters include low-pass filter cutoff frequency, high-pass filter cutoff frequency, high-frequency lift-up frequency, high-frequency lift-up order, quantization maximum voltage, vertical charge transfer efficiency, horizontal charge transfer efficiency, low-pass filter order, high-pass filter order, and system signal-to-noise ratio threshold.

[0043] 1.2) Obtain the GPU configuration information of the simulated machine:

[0044] The configuration information of the machine can be obtained by using the API interface functions provided by the CUDA library, including: device name, number of stream processor groups (SMs), shared memory size of each thread block, maximum number of threads per thread block, maximum number of threads per stream processor EM, and maximum number of thread bundles per stream processor group (SM).

[0045] Step 2: Establish the kernel function family of the modulation transfer function (MTF).

[0046] 2.1) Based on the parameters obtained in step 1, a theoretical model of the overall modulation transfer function (MTF) of the infrared imaging system is established by sampling a subset of the parameters:

[0047] This example only samples the optical system, detector, and display, and calculates the modulation transfer functions of these sampled and unsampled parameters respectively. The specific implementation is as follows:

[0048] 2.1.1) Calculate the atmospheric turbulence modulation transfer function MTFt before sampling. ur (f):

[0049]

[0050] Where a is a calculation constant, with a value of 3 / 8; f is the spatial frequency, in cys / degree; λ is the wavelength; R is the target distance; and D is the detector entrance pupil diameter, in meters.

[0051] 2.1.2) Calculate the modulation transfer function (MTF) of the optical system before sampling. opt (f):

[0052] MTF opt (f) = MTF aber (f)·MTF defocus (f)·MTF diff (f)

[0053] Among them, MTF aber (f) is the aberration modulation transfer function, MTF defocus (f) is the defocus modulation transfer function, MTF. diff (f) is the diffraction modulation transfer function, and these three functions are expressed as follows:

[0054]

[0055]

[0056]

[0057] In the formula, J1 is the first-order Bessel function, and f0 is the diffraction cutoff frequency. λ1 is the center wavelength; W rms is the mean squared error of aberrations, with a value between 0 and 0.2; f is the spatial frequency, in cys / degree. R represents the amount of defocus. f R is the focusing distance, D is the distance from the viewpoint to the target, and R is the entrance pupil diameter. The units of R and D are meters.

[0058] 2.1.3) Calculate the detector modulation transfer function (MTF) before sampling. det (f):

[0059] MTF det (f)=sinc(f·w det )

[0060] in, The angle subtended by one pixel of the detector, measured in mrad, H pix L represents the pixel size of the detector, measured in micrometers. foous The focal length of the optical system;

[0061] 2.1.4) Calculate the modulation transfer function (MTF) of the sampled display. disp (f):

[0062] MTF disp (f)=sin c(f / SMAG·w disp )

[0063] in, H represents the angle subtended by one pixel of the display at the human eye. disp The pixel size of the display is in millimeters (L). eye The distance from the human eye to the detector, expressed in meters;

[0064] 2.1.5) Based on the influence of human eye optics, retinal and ocular nystagmus on the modulation transfer function (MTF) of the human eye, calculate the MTF of the sampled human eye. eye (f):

[0065] MTF eye (f) = MTF opt (f)·MTF retina (f)·MTF tre (f)

[0066]

[0067] MTF retina (f)=exp(-0.4441f 1.21 )

[0068] MTF tre (f)=exp(-0.375f 2 )

[0069] D pup = -9.011 + 13.23exp(-log 10 (L0) / 21.082)-0.5(n eye -1)

[0070]

[0071]

[0072] Among them, MTF opt (f) is the modulation transfer function of the pupil, MTF retina (f) is the modulation transfer function of the retina, MTFt re (f) is the modulation transfer function of nystagmus, L0 is the average display brightness of the display in fL, and n eye The number of human eyes being observed;

[0073] 2.1.6) Based on the above calculation results, the theoretical model of the overall modulation transfer function (MTF) of the infrared imaging system is obtained:

[0074] MTF sys =MTF tur (f)·MTF opt (f)·MTF det (f)·MTF disp (f)·MTF eye (f)

[0075] Among them, MTF sys is the overall modulation transfer function (MTF) of the infrared imaging system.

[0076] 2.2) Based on the modulation transfer function model established in step (2.1), a family of kernel functions for the MTF is established, including the atmospheric turbulence kernel function family, the optical system kernel function family, the detector kernel function family, the display kernel function family, and the human eye kernel function family:

[0077] Each kernel function must be written according to the set rules, that is, it must be written in a *.cu file. In this example, it is written in the mtf.cu file, and is declared and defined with the __global__ qualifier, with void indicating the return type, and does not support a variable number of parameters;

[0078] Each kernel function has four very useful built-in variables: threadIdx, lockIdx, blockDim, and gridDim.

[0079] The thread block is indexed by blockIdx, the thread within a block is indexed by threadIdx, the total number of threads within a block is obtained by blockDim, and the total number of blocks within a grid is obtained by gridDim.

[0080] Step 3: Establish the kernel function family of the system contrast threshold function CTF.

[0081] 3.1) Based on the parameters obtained in step 1, establish the CTF theoretical model for the contrast threshold function of the infrared imaging system:

[0082] 3.1.1) Calculate the uncorrected visual contrast threshold function (CTF) nake :

[0083]

[0084] in, L represents the average brightness of the display, measured in fL, and f represents the spatial frequency, measured in cys / degree. eyeHere, is the number of eyes, 'a' is the size of the target in meters, 'R' is the distance from the viewpoint to the target in kilometers, and 'c' is a constant with a value of 0.06.

[0085] 3.1.2) Calculate the glare factor M of the display screen respectively. dsp And noise factor NF(f):

[0086]

[0087]

[0088] Among them, L glare Γ is the glare signal of the display, SCT is the scene contrast temperature difference in K; α is the scaling factor in √-Hertz; Γ is the noise power spectral density in K·mrad·Hz; QH(f) is the horizontal noise equivalent bandwidth related to the spatial frequency f; QV(f) is the vertical noise equivalent bandwidth related to the spatial frequency f.

[0089] 3.1.3) Based on the above calculation results, the theoretical model of the contrast threshold function (CTF) for the infrared imaging system is obtained:

[0090]

[0091] Among them, CTF nake M is the naked-eye contrast threshold function. dsp NF(f) is the glare factor of the display, NF(f) is the noise factor, and f is the spatial frequency. H represents the system amplification. disp H represents the display pixel size. pix L is the detector pixel size. focus L is the system focal length. eye The distance from the human eye to the monitor;

[0092] 3.2) Based on the contrast threshold function model established in step (3.1), the kernel function family of CTF is established according to the set writing rules. It includes the naked eye contrast threshold function kernel function family, the display glare factor kernel function family and the noise factor kernel function family. The writing rules are the same as in step 2.2).

[0093] Step 4: Introduce a parallel computing model, establish host and device sides, and perform task scheduling, distribution and parallel data computation respectively.

[0094] 4.1) Based on the GPU information obtained in step 1.2), set the number of resources that the host can access from the device:

[0095] CUDA logically divides GPU threads into three levels: grid, block, and thread. Each kernel function corresponds to a grid, a grid contains one or more blocks, and a block contains one or more threads. The relationship between the three is as follows: Figure 1 The device-side section is shown. The three-layer structure of threads is related to the hardware composition of the GPU. Since the GPU hardware itself contains three layers—cores, stream processor groups, and devices—it forms a hierarchical structure similar to a computer cluster. Therefore, the previously established kernel functions can run on this hierarchical structure, and tasks can be smoothly passed sequentially during runtime. The startup of a kernel function depends on hardware configuration information, usually given in angle brackets <<<>>>. These angle brackets include four types of information: the number of blocks, the number of threads, dynamically allocated shared memory, and streams. Dynamically allocated shared memory and streams are not mandatory options.

[0096] The number of resources that the host can access from the device includes the number of threads and the number of blocks, and their settings are as follows:

[0097] 4.1.1) Determine the number of threads (threads) within a block based on the hardware resources of the GPU device, i.e., obtain this information using the API interface functions provided by the CUDA library.

[0098] 4.1.2) The number of blocks (blocks) is determined based on the data size and the number of data items processed by each thread:

[0099] blocks=(dataNum+threads-1) / threads / taskPerThread

[0100] Where dataNum is the amount of data to be processed, and taskPerThread is the amount of data processed by each thread.

[0101] By calculating the appropriate number of blocks and threads, a suitable scheme for the operation of the kernel function family is determined.

[0102] 4.2) Considering that CPUs are suitable for control-intensive tasks and GPUs are suitable for data-intensive tasks, a parallel computing model based on the unified computing device architecture CUDA is introduced, establishing a host-CPU and a device-GPU; and a grid and a block are established according to the architectural characteristics of CUDA.

[0103] 4.3) Device-side memory management in the CUDA programming model:

[0104] Place the data that needs to be accessed frequently in the model calculation in shared memory, and store ordinary data in registers;

[0105] Global data is accessed in a merged manner during kernel function processing. That is, the data to be used is first packaged to reduce the number of access operations of the device to the memory subsystem. Then, the number of grids and thread blocks are used to specify the number of threads to be executed on the device side. A unique thread ID is obtained through the built-in variable threadIdx of the kernel function. Each thread executes the corresponding kernel function according to the thread ID.

[0106] 4.4) Data transfer between the host and device:

[0107] Reference Figure 2 The specific implementation of this step is as follows:

[0108] The host executes the logic control involved in the performance evaluation and transmits the tasks involving a large number of floating-point calculations, such as vectors, matrices, and integrals, to the device.

[0109] On the device side, the kernel function families of the Modulation Transfer Function (MTF) and Contrast Threshold Function (CTF) models are run in different grids and thread blocks, such as... Figure 2 The diagram shows a thread block in the device side. The thread block manages multiple threads in it, calculates MTF and CTF data in parallel, and marks the validity of these functions.

[0110] The host allocates host memory and initializes data.

[0111] The device allocates device memory and copies data from host memory to the device; then it calls the kernel function of the modulation transfer function (MTF) and contrast threshold function (CTF) model to perform the specified calculations on the device, and marks the results to distinguish the loop conditions of these results; then it copies the calculation results to host memory on the host side, and releases the memory it allocated.

[0112] The host receives the calculation results from the device, organizes and merges the data according to the device's identifier, and obtains the modulation transfer function (MTF) data. sys Data CTF with contrast threshold function sys .

[0113] Step 5, based on the contrast threshold function data CTF obtained in (4) sys Based on the TTP criterion, the detection probability of the infrared imaging system is calculated.

[0114] 5.1) Based on the target contrast C tgt Data CTF with contrast threshold function sys Calculate the target task performance (TTP):

[0115]

[0116] Wherein, the upper and lower limits of integration f low f high These represent the start and end points of the spatial frequency range where the target contrast is greater than the modulation transfer function, respectively.

[0117] 5.2) Calculate the number of stripes V based on the target task performance (TTP):

[0118]

[0119] Where a is the size of the target in meters; R is the distance to the target in kilometers.

[0120] 5.3) Calculate the detection probability P based on the number of stripes V:

[0121]

[0122] in, V 50 The value input by the user is the V value when the detection probability is 50%.

[0123] Step 6: Complete the performance evaluation simulation of the infrared imaging system for the entire current scene.

[0124] 6.1) Based on the current scene conditions, calculate the detection probability at the current distance and determine whether the detection probability P at all distances has been completed:

[0125] If completed, proceed to step 6.2);

[0126] If not completed, return to step 4.3) until the detection probability set under the current conditions of the entire scene is obtained, and then execute step 6.2);

[0127] 6.2) Plot a probability curve, which is based on different V values ​​input by the user. 50 The numerical values ​​can be detection curves, identification curves, or recognition curves, and the corresponding data are given in the form of a report to complete the performance evaluation and simulation of the infrared imaging system.

[0128] The above description is merely a specific example of the present invention and does not constitute any limitation on the present invention. Obviously, those skilled in the art, after understanding the content and principles of the present invention, may make various modifications and changes in form and details without departing from the principles and structure of the present invention. However, these modifications and changes based on the ideas of the present invention are still within the scope of protection of the claims of the present invention.

Claims

1. A simulation method for performance evaluation of an infrared imaging system based on parallel computing, characterized in that, Including the following: (1) Obtain the optical system parameters, detector parameters and circuit system parameters in the infrared imaging system, and use the API interface to obtain the GPU configuration information of the simulated machine; (2) Establish the kernel function family of the modulation transfer function (MTF): (2a) Based on the parameters obtained in (1), a theoretical model of the overall modulation transfer function (MTF) of the infrared imaging system is established through sampling: ， in, The overall modulation transfer function (MTF) of the infrared imaging system. The atmospheric turbulence MTF before sampling The MTF of the optical system before sampling includes three parts: aberration, defocus, and diffraction. The modulation transfer function (MTF) of the detector before sampling. The modulation transfer function (MTF) of the display after sampling. The modulation transfer function (MTF) of the human eye after sampling; (2b) Based on the modulation transfer function model established in step (2a), establish the kernel function family of MTF, which includes the atmospheric turbulence kernel function family, the optical system kernel function family, the detector kernel function family, the display kernel function family, and the human eye kernel function family; (3) Establish the kernel function family of the system contrast threshold function CTF: (3a) Based on the modulation transfer function (MTF) theoretical model and the influence of the system on the naked-eye contrast threshold function, establish the system's contrast threshold function (CTF) theoretical model: ， Among them, CTF nake M is the naked-eye contrast threshold function CTF. dsp Let NF(f) be the glare factor of the display, NF(f) be the noise factor, and f be the spatial frequency. H represents the system amplification. disp H represents the display pixel size. pix L is the detector pixel size. focus L is the system focal length. eye The distance from the human eye to the monitor; (3b) Based on the system contrast threshold function model, establish the kernel function family of the contrast threshold function model CTF, which includes the naked-eye CTF kernel function family, the display glare factor kernel function family and the noise factor kernel function family; (4) Introduce a parallel computing model, establish host and device sides, and perform task scheduling, distribution and parallel data computation respectively: (4a) Considering that CPUs are suitable for control-intensive tasks and GPUs are suitable for data-intensive tasks, a parallel computing model based on the unified computing device architecture CUDA is introduced, establishing host-CPU on the host side and device-GPU on the device side; and grid and block are established according to the architectural characteristics of CUDA. (4b) The device manages the memory in the CUDA programming model, that is, it puts the data that needs to be accessed frequently in the model calculation in shared memory, puts ordinary data in registers, and merges the access to global data during the kernel function processing. (4c) The host performs task scheduling and distribution: The host executes the logic control involved in the performance evaluation and transmits the tasks involving a large number of floating-point calculations, such as vectors, matrices, and integrals, to the device. On the device side, the kernel function family of the modulation transfer function (MTF) and contrast threshold function (CTF) models is run in different grids and thread blocks to compute MTF and CTF data in parallel, and these functions are managed and their validity is identified through grids and thread blocks. The host receives the calculation results from the device, organizes and merges the data according to the device's identifier, and obtains the MTF data. and CTF data ; (5) Based on the CTF data obtained in (4), calculate the detection probability of the infrared imaging system according to the TTP criterion; (6) Repeat steps (4b), (4c), and (5) according to the current scene settings until the detection probability set under the current scene settings is obtained. At the same time, plot the probability set curve and give the corresponding data in the form of a report to complete the performance evaluation simulation of the infrared imaging system.

2. The method according to claim 1, characterized in that, The atmospheric turbulence modulation transfer function before sampling in step (2a) The calculation formula is as follows: ， Where a is a calculation constant, with a value of 3 / 8; f is the spatial frequency; λ is the wavelength; R is the distance from the viewpoint to the target; and D is the entrance pupil diameter of the detector.

3. The method according to claim 1, characterized in that, The modulation transfer function of the optical system before sampling in step (2a) The calculation formula is as follows: ， Among them, MTF aber (f) is the aberration modulation transfer function, MTF defocus (f) is the defocus modulation transfer function, MTF. diff (f) is the diffraction modulation transfer function, which is expressed as follows: ， ， ， In the formula, J1 is the first-order Bessel function, and f0 is the diffraction cutoff frequency. , The center wavelength; W rms is the mean square error of the aberration, with a value between 0 and 0.2; f is the spatial frequency, W R represents the amount of defocus. f R is the focusing distance, R is the distance from the viewpoint to the target, and D is the entrance pupil diameter of the detector.

4. The method according to claim 1, characterized in that, The pre-sampling detector modulation transfer function in step (2a) The calculation formula is as follows: ， Where f is the spatial frequency. H represents the angle subtended by one pixel of the detector. pix L is the pixel size of the detector. focus This refers to the focal length of the optical system.

5. The method according to claim 1, characterized in that, The sampled display modulation transfer function in step (2a) The calculation formula is as follows: ， Where f is the spatial frequency. H represents the angle subtended by one pixel of the display at the human eye. disp L is the pixel size of the display. eye The distance from the human eye to the detector.

6. The method according to claim 1, characterized in that, The sampling human eye modulation transfer function in step (2a) The calculation formula is as follows: ， in, The modulation transfer function of the pupil. This is the modulation transfer function of the retina. is the modulation transfer function for nystagmus.

7. The method according to claim 1, characterized in that, The naked-eye contrast threshold function CTF in step (3a) nake The calculation formula is as follows: ， in, L is the brightness of the display, f is the spatial frequency, Neye is the number of eyes, a is the size of the target, R is the distance from the viewpoint to the target, and c is a constant with a value of 0.

06.

8. The method according to claim 1, characterized in that, The display glare factor M in step (3a) dsp The formulas for calculating the noise factor NF(f) are as follows: ， ， Where L is the average brightness of the display, L glare Here, SCT represents the glare signal of the display, α represents the scene contrast temperature difference, and α is the scaling factor. Let QH(f) be the noise power spectral density, QV(f) be the noise equivalent bandwidth in the horizontal direction related to the spatial frequency f, and QV(f) be the noise equivalent bandwidth in the vertical direction related to the spatial frequency f.

9. The method according to claim 1, characterized in that, In step (4b), the global data is merged during the kernel function processing. This involves packaging the data to be used to reduce the number of access operations the device performs on the memory subsystem. Then, the number of grids and the number of thread blocks are used to specify the number of threads to be executed on the device side, and a unique thread ID is obtained through the built-in variable threadIdx of the kernel function; Each thread executes the corresponding kernel function based on its thread ID, thereby effectively improving the speed of parallel computing.

10. The method according to claim 1, characterized in that, In step (4c), the device manages and identifies the validity of the kernel functions of the Modulation Transfer Function (MTF) and Contrast Threshold Function (CTF) models through grids and thread blocks, as follows: The host allocates host memory and initializes data. The device allocates memory on the device and copies data from the host memory to the device; The device calls the kernel functions of the modulation transfer function (MTF) and contrast threshold function (CTF) model to perform the specified calculations on the device, identifies the results to distinguish the loop conditions of these results, copies the calculation results to the host memory on the host side, and releases the memory allocated to itself. The host machine aggregates the data obtained from the device.

11. The method according to claim 1, characterized in that, In step (5), the detection probability P of the infrared imaging system is calculated according to the TTP criterion, as follows: (5a) Based on the target contrast C tgt and Calculate the target task performance (TTP): ， Wherein, the upper and lower limits of integration f low、 f high These represent the start and end points of the spatial frequency range where the target contrast is greater than the modulation transfer function, respectively. (5b) Calculate the number of stripes V based on the target task performance (TTP): ， in, R represents the size of the target, and R represents the distance from the viewpoint to the target. (5c) Calculate the detection probability P based on the number of stripes V: ， in, V 50 The value input by the user is the V value when the detection probability is 50%.

Images