Millimeter wave passive imaging super-resolution reconstruction method based on adjacent signal compensation

Abstract

The invention discloses a millimeter wave passive imaging super-resolution reconstruction method based on adjacent signal compensation, and the method comprises the steps: building a millimeter wave passive imaging receiving link output voltage time domain physical model under an integral effect, defining a system compensation coefficient, and deducing a theoretical steady-state output voltage through the segment end voltage difference of adjacent integral windows, and sub-pixel-level brightness temperature jump details are recovered, and the problem of brightness temperature signal blurring caused by an integral effect is solved. The method disclosed by the invention is applied to the millimeter wave passive imaging system, realizes low-complexity real-time processing under the conditions of not changing the existing hardware architecture and not increasing the system cost, can be quickly deployed to the existing passive imaging system, remarkably improves the resolution ratio, meets the high-precision detection requirements of scenes such as security detection and the like, and is suitable for popularization and application. The method can also be applied to other products with the signal fuzzy problem caused by the low-pass filtering effect in the integral sampling process, and the signal contrast can be remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of focal plane passive imaging, and in particular relates to a millimeter wave passive imaging super-resolution reconstruction method based on adjacent signal compensation. Background Art

[0002] Millimeter-wave focal plane passive imaging technology reconstructs the scene's brightness temperature distribution by receiving the millimeter-wave signals radiated by the target itself, demonstrating significant application value in the field of security detection. Compared to active imaging, it does not require the emission of electromagnetic waves and has the advantages of being contactless, radiation-free, and providing privacy protection. It can penetrate non-metallic materials such as clothing and plastics, and accurately identify concealed dangerous goods such as ceramic knives and explosives. It is suitable for high-traffic scenarios such as airport security and key area monitoring. However, due to the integration time effect of the focal plane array and the low-pass filtering characteristics of the system, the sensor smoothes the brightness temperature signal in the spatial domain, resulting in a decrease in the sharpness of the target edge and a loss of detail in areas with weak temperature differences, resulting in a blurred brightness temperature signal. Especially in complex backgrounds or outside the depth of field, the blurring effect significantly reduces the resolution and false alarm suppression capabilities, seriously restricting the accurate recognition of submillimeter targets.

[0003] To address the brightness temperature ambiguity problem, existing technologies mainly use the following methods:

[0004] 1. Frequency-domain deconvolution methods, such as Wiener filtering and the Richardson-Lucy algorithm, recover high-frequency signals based on frequency-domain inverse filtering of the system transfer function. However, these algorithms rely on accurate prior knowledge of the point spread function. In practical applications, they are prone to ringing artifacts due to model mismatch and are extremely sensitive to noise, making them difficult to meet the robustness requirements of security scenarios.

[0005] 2. Compressed sensing and sparse reconstruction methods: Improve equivalent resolution through sparse array design or compressed sensing theory. These methods require customized hardware, significantly increasing system costs, and are incompatible with existing commercial focal plane architectures, limiting their engineering practicality.

[0006] 3. Data-driven deep learning: This method uses convolutional neural networks to learn super-resolution mapping relationships from blurred images. Although it achieves good simulation results, it relies on large-scale annotated datasets, lacks generalization capabilities for noise distribution and imaging conditions, and has high real-time performance and hardware deployment costs.

[0007] In summary, existing super-resolution methods have defects such as frequency domain deconvolution relying on precise point spread function, compressed sensing requiring customized hardware modification, and insufficient generalization of data-driven deep learning. It is now necessary to study a super-resolution method to address the problem of brightness temperature signal blurring caused by the integration effect in millimeter-wave focal plane passive imaging. Summary of the Invention

[0008] To solve the above technical problems, the present invention provides a millimeter-wave passive imaging super-resolution reconstruction method based on adjacent signal compensation, which can be used in millimeter-wave passive imaging systems to eliminate the signal blurring problem caused by the low-pass filtering effect during the integral sampling process, thereby improving imaging resolution, restoring the image contour of the imaging target, and improving recognition accuracy.

[0009] The technical solution adopted by the present invention is: a millimeter wave passive imaging super-resolution reconstruction method based on adjacent signal compensation, the specific steps are as follows:

[0010] S1. Construct a physical model of the output voltage of the millimeter wave passive imaging receiving link based on the circuit composition;

[0011] S11. Establishing a physical model;

[0012] For the millimeter-wave focal plane passive imaging system, the original signal is first amplified by a low-noise amplifier, then subjected to envelope detection processing, further amplified, and an integrator is used to filter out high-frequency noise.

[0013] For a target point with constant brightness temperature, the radiated signal is approximately considered to contain only DC component after detection and amplification. Suppose the input signal of the RC integrator is U i (t), the output signal is U o (t), according to Kirchhoff’s voltage law KVL, the expression can be obtained as follows:

[0014] U i (t) = U R (t)+U o (t) (1)

[0015] Where t represents time, and the voltage drop across the resistor is U R (t)=I(t)R, capacitor current Substituting into formula (1), we can get the following expression:

[0016]

[0017] Where R represents the resistance of the RC circuit's current-limiting resistor, and C represents the capacitance of the RC circuit's filter capacitor. Letting the time constant τ = RC, Equation (2) can be reorganized into a standard first-order linear differential equation, as follows:

[0018]

[0019] S12, setting the pixel integration time, solving the RC model based on the initial conditions, and obtaining the mathematical representation of the output voltage at the end of the segment;

[0020] Set the voltage corresponding to the nth pixel point of the vertical scan of a receiving channel to A n , the pixel integration time is Tint , the expression is as follows:

[0021] U i (t) = A n , t∈[t n-1 ,t n ] (4)

[0022] Among them, t n-1 Indicates the end time of the segment corresponding to the n-1th pixel point scanned, t n Indicates the end time of the segment corresponding to the nth pixel scanned.

[0023] In t n-1 There is an initial condition at the moment, that is, the output voltage U corresponding to the n-1th pixel point o (t n-1 )=V n-1 , substitute into differential equation (3), solve the RC model, and get the output voltage U across the capacitor o (t) is expressed as follows:

[0024]

[0025] Then at the end of the segment, time t = t n , that is, the integration time T int After that, the output voltage V n The expression is as follows:

[0026]

[0027] S13, define the compensation coefficient, substitute it into the output voltage expression obtained in step S12, and inversely derive the recursive explicit expression of the theoretical steady-state output voltage;

[0028] Define the compensation coefficient α, the expression is as follows:

[0029]

[0030] Substituting α into formula (6) yields the following expression:

[0031] V n =αA n +(1-α)V n-1 (8)

[0032] Then use t=t n and t=t n-1 The sampling value V at the moment n and V n-1 Inverse its theoretical steady-state output voltage Y n =A n , according to formula (8), the expression is as follows:

[0033]

[0034] Among them, when α=1, that is, T int →∞, Y n =V n , which is consistent with the steady-state characteristics.

[0035] S2. Based on the output voltage physical model constructed in step S1, the imaging target is scanned and sampling is performed at the end of each integration window to obtain an initial data matrix of the frame image;

[0036] The specific expression of the initial data matrix V is as follows:

[0037]

[0038] Among them, V ii Represents the signal intensity of the pixel at row i and column i of the initial data matrix V, and 1≤i≤N.

[0039] S3. Calculate the time constant based on the circuit composition and calculate the compensation coefficient based on the integration time. Then perform differential processing on the data at the end of the previous and next segments to obtain a near differential data matrix.

[0040] Calculate the time constant τ based on the circuit composition and combine it with the set integration time T for each pixel int , calculate the compensation coefficient α according to formula (7).

[0041] The millimeter-wave focal plane passive imaging system performs longitudinal scanning and inverts each column of data. First, each column of data is subtracted to obtain the adjacent differential data matrix ΔV, which is expressed as follows:

[0042]

[0043] Where, ΔV ii It represents the difference between the signal intensity of the pixel in the i-th row and i-th column of the initial data matrix V and the signal intensity of the pixel in the i-th row and i-th column, and 1≤i≤N.

[0044] S4, performing theoretical derivation based on the output voltage physical model described in step S1, inverting and reconstructing the original signal corresponding to the current frame image, and completing super-resolution reconstruction;

[0045] The initial data is inverted by the compensation coefficient to obtain the original signal data Y, which is expressed as follows:

[0046]

[0047] Beneficial effects of the present invention: The method of the present invention establishes a time-domain physical model of the output voltage of the millimeter-wave passive imaging receiving link under the integration effect, defines the system compensation coefficient, and directly derives the theoretical steady-state output voltage using the end-of-segment voltage difference of adjacent integration windows, thereby restoring the sub-pixel level brightness temperature jump details and solving the brightness temperature signal blurring problem caused by the integration effect. The method of the present invention is applied to the millimeter-wave passive imaging system, and realizes low-complexity real-time processing without changing the existing hardware architecture and without increasing the system cost. It can be quickly deployed to the existing passive imaging system, significantly improving the resolution, meeting the high-precision detection requirements of scenarios such as security detection, and can also be applied to other products that have signal blurring problems due to the low-pass filtering effect during the integration sampling process, and can significantly improve the signal contrast. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 The present invention is a flow chart of a millimeter wave passive imaging super-resolution reconstruction method based on adjacent signal compensation.

[0049] Figure 2 This is a flow chart for deducing a physical model for the output voltage of a millimeter-wave passive imaging receiving link in an embodiment of the present invention.

[0050] Figure 3 Schematic diagram of the physical model of the receiving channel link in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The method of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0052] like Figure 1 As shown in FIG, a flow chart of a millimeter wave passive imaging super-resolution reconstruction method based on adjacent signal compensation of the present invention is shown, and the specific steps are as follows:

[0053] S1. Construct a physical model of the output voltage of the millimeter wave passive imaging receiving link based on the circuit composition;

[0054] S11. Establishing a physical model;

[0055] The derivation process of constructing the physical model of the output voltage of the millimeter wave passive imaging receiving link is as follows: Figure 2 As shown, for the millimeter wave focal plane passive imaging system, the receiving channel link physical model is as follows: Figure 3 The original signal is first amplified by a low-noise amplifier, then subjected to envelope detection, further amplified, and an integrator is used to filter out high-frequency noise.

[0056] For a target point with constant brightness temperature, the signal radiated by it can be approximately considered to contain only DC components after detection and amplification. Suppose the input signal of the RC integrator is U i (t), the output signal is U o(t), according to the Kirchhoff voltage law KVL, the expression is as follows:

[0057] U i (t) = U R (t) + U o (t) (1)

[0058] Wherein, t represents time, the voltage drop across the resistor is U R (t) = I(t)R, the capacitor current Substitute equation (1) into the expression is as follows:

[0059]

[0060] Wherein, R represents the resistance value of the RC circuit current limiting resistor, C represents the capacitance value of the RC circuit filter capacitor. Let the time constant τ = RC, then equation (2) can be arranged into a standard first-order linear differential equation, the expression is as follows:

[0061]

[0062] S12, set the pixel integration time, combined with the initial condition to solve the RC model, get the end of time output voltage mathematical expression;

[0063] Set the voltage corresponding to the nth pixel point of the longitudinal scanning of a receiving channel as A n , the integration time of the pixel point is T int , the expression is as follows:

[0064] U i (t) = A n , t ∈ [t n-1 , t n ] (4)

[0065] Wherein, t n-1 represents the end of time corresponding to the scanning of the nth-1 pixel point, t n represents the end of time corresponding to the scanning of the nth pixel point.

[0066] At t n-1 time, there is an initial condition, that is, the output voltage U o (t n-1 ) corresponding to the nth-1 pixel point V n-1 , substitute the differential equation (3), solve the RC model, and the expression of the output voltage U o (t) across the capacitor is as follows:

[0067]

[0068] Then at the end of time t = t n , that is, the integration time T intAfterwards, output voltage V n The expression is as follows:

[0069]

[0070] S13, define the compensation coefficient, substitute the output voltage expression obtained in step S12, and inverse the theoretical steady-state output voltage recursive explicit expression;

[0071] Define the compensation coefficient α, and the expression is as follows:

[0072]

[0073] Substitute α into formula (6), and the expression is as follows:

[0074] V n = αA n +(1-α)V n-1 (8)

[0075] Then, the sampling values V n and V n-1 at t=t n and t=t n-1 are used to inverse the theoretical steady-state output voltage Y n =A n , and the expression is as follows according to formula (8):

[0076]

[0077] When α=1, that is, T int →∞, Y n =V n , which is consistent with the steady-state characteristic.

[0078] S2, based on the output voltage physical model constructed in step S1, the imaging target is scanned, and the sampling is performed at the end of each integration window to obtain an initial data matrix of a frame image;

[0079] The specific expression of the initial data matrix V is as follows:

[0080]

[0081] Wherein, V ii represents the signal intensity of the i-th row i-th column pixel point of the initial data matrix V, and 1≤i≤N.

[0082] S3, calculate the time constant according to the circuit composition, combine the compensation coefficient according to the integration time, and then perform difference processing on the data at the end of the front and rear segments to obtain a near-difference data matrix;

[0083] Calculate the time constant τ according to the circuit composition, combine the set integration time Tint , calculate the compensation coefficient α according to formula (7).

[0084] The ambiguity of the brightness temperature signal is mainly reflected in the scanning direction. The millimeter-wave focal plane passive imaging system performs longitudinal scanning, and each column of data needs to be inverted. First, each column of data is subtracted to obtain the adjacent differential data matrix ΔV. The expression is as follows:

[0085]

[0086] Where, ΔV ii It represents the difference between the signal intensity of the pixel in the i-th row and i-th column of the initial data matrix V and the signal intensity of the pixel in the i-th row and i-th column, and 1≤i≤N.

[0087] S4, performing theoretical derivation based on the output voltage physical model described in step S1, inverting and reconstructing the original signal corresponding to the current frame image, and completing super-resolution reconstruction;

[0088] The initial data is inverted by the compensation coefficient to obtain the original signal data Y, which is expressed as follows:

[0089]

[0090] This embodiment further conducts simulation verification. In the millimeter-wave focal plane passive imaging system used in this embodiment, the integrator input signal can be approximately considered as the superposition of DC signal and noise. Assume that the DC sequence is U i =[1120.5311.50.512], the overall signal-to-noise ratio is 10dB.

[0091] In this embodiment, the integrator time constant is set to 1.5ms, and the integration time of each pixel is 1ms. Based on the output voltage physical model established in the principle derivation, sampling is performed at the end of each integration period to obtain the initial data sequence, which is expressed as follows:

[0092] V=[0.50970.74801.35970.92641.94411.50551.49050.99610.99291.4864](13)

[0093] The method of the present invention can be used to invert the initial data sequence V into the DC sequence U i , as follows:

[0094] In this embodiment, the integration time T int =1ms, time constant τ =1.5ms, then the compensation coefficient α can be calculated as follows:

[0095]

[0096] Then, the sampling data at the end of adjacent segments are differentially processed to obtain the adjacent differential sequence ΔV, which is expressed as follows:

[0097]

[0098] Finally, the adjacent difference sequence, compensation coefficient and sampling sequence are substituted into formula (12) to invert and reconstruct the original signal sequence Y, which is expressed as follows:

[0099]

[0100] The simulation results of this embodiment show that the method of the present invention can accurately invert and reconstruct the original signal corresponding to the target. The results show that the signal reconstruction accuracy is greatly affected by noise. int =1ms, time constant τ =1.5ms. When the signal-to-noise ratio is 4dB, the average absolute percentage error between the end-segment sampling value and the input DC component is 48.91%, and the average absolute percentage error between the reconstructed signal and the input DC component is significantly reduced to 4.66%. When the signal-to-noise ratio is 15dB, the average absolute percentage error between the end-segment sampling value and the input DC component is 48.90%, and the average absolute percentage error between the reconstructed signal and the input DC component is further reduced to 0.97%.

[0101] In summary, the method of the present invention is applied to millimeter-wave passive imaging systems, achieving low-complexity real-time processing without changing the existing hardware architecture and without increasing system costs. It can be quickly deployed to existing passive imaging systems, significantly improving resolution and meeting the high-precision detection requirements of scenarios such as security detection. It can also be applied to other products that have signal blurring problems due to low-pass filtering effects during the integral sampling process, and can significantly improve signal contrast.

[0102] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A millimeter wave passive imaging super-resolution reconstruction method based on adjacent signal compensation, the specific steps are as follows: S1. Construct a physical model of the output voltage of the millimeter wave passive imaging receiving link based on the circuit composition; S11. Establishing a physical model; For the millimeter-wave focal plane passive imaging system, the original signal is first amplified by a low-noise amplifier, then subjected to envelope detection processing, further amplified, and an integrator is used to filter out high-frequency noise; For a target point with constant brightness temperature, the radiated signal is approximately considered to contain only DC component after detection and amplification. Suppose the input signal of the RC integrator is U i (t), the output signal is U o (t), according to Kirchhoff’s voltage law KVL, the expression can be obtained as follows: U i (t)=U R (t)+U o (t) (1) in, t represents time, and the voltage drop across the resistor is U R (t)=I(t)R, capacitor current Substituting into formula (1), we can get the following expression: Where R represents the resistance of the RC circuit current limiting resistor, and C represents the capacitance of the RC circuit filter capacitor. Let the time constant τ = RC, then Equation (2) can be organized into a standard first-order linear differential equation as follows: S12, setting the pixel integration time, solving the RC model based on the initial conditions, and obtaining the mathematical representation of the output voltage at the end of the segment; Set the voltage corresponding to the nth pixel point of the vertical scan of a receiving channel to A n , the pixel integration time is T int , the expression is as follows: U i (t)=A n ,t∈[t n-1 ,t n ] (4) Among them, t n-1 Indicates the end time of the segment corresponding to the n-1th pixel point scanned, t n Indicates the end time of the segment corresponding to the nth pixel point scanned; In t n-1 There is an initial condition at the moment, that is, the output voltage U corresponding to the n-1th pixel point o (t n-1 )=V n-1 , substitute into differential equation (3), solve the RC model, and get the output voltage U across the capacitor o (t) is expressed as follows: Then at the end of the segment, time t = t n , that is, the integration time T int After that, the output voltage V n The expression is as follows: S13, define the compensation coefficient, substitute it into the output voltage expression obtained in step S12, and inversely derive the recursive explicit expression of the theoretical steady-state output voltage; Define the compensation coefficient α, the expression is as follows: Substituting α into formula (6) yields the following expression: V n =αA n +(1-α)V n-1 (8) Then use t=t n and t=t n-1 The sampling value V at the moment n and V n-1 Inverse its theoretical steady-state output voltage Y n =A n , according to formula (8), the expression is as follows: Among them, when α=1, that is, T int →∞, Y n =V n , which is consistent with the steady-state characteristics; S2. Based on the output voltage physical model constructed in step S1, the imaging target is scanned and sampling is performed at the end of each integration window to obtain an initial data matrix of the frame image; The specific expression of the initial data matrix V is as follows: Among them, V ii represents the signal intensity of the pixel at row i and column i of the initial data matrix V, and 1≤i≤N; S3. Calculate the time constant based on the circuit composition and calculate the compensation coefficient based on the integration time. Then perform differential processing on the data at the end of the previous and next segments to obtain a near differential data matrix. Calculate the time constant τ based on the circuit composition and combine it with the set integration time T for each pixel int , calculate the compensation coefficient α according to formula (7); The millimeter-wave focal plane passive imaging system performs longitudinal scanning and inverts each column of data. First, each column of data is subtracted to obtain the adjacent differential data matrix ΔV, which is expressed as follows: Where, ΔV ii represents the difference between the signal intensity of the pixel at the i-th row and i-th column of the initial data matrix V and the signal intensity of the pixel at the i-th row and i-th column, and 1≤i≤N; S4, performing theoretical derivation based on the output voltage physical model described in step S1, inverting and reconstructing the original signal corresponding to the current frame image, and completing super-resolution reconstruction; The initial data is inverted by the compensation coefficient to obtain the original signal data Y, which is expressed as follows:

Images