A millimeter wave radiation image fire point detection and identification method and system

Abstract

The application discloses a millimeter wave radiation image fire point detection and recognition method and system, and belongs to the technical field of millimeter wave imaging. The application improves the wavelet threshold denoising method when the millimeter wave radiation image is pretreated, and has the following advantages: the threshold is continuous, the influence caused by the discontinuity of the image edge part due to local jitter is eliminated, the quality of the image and the accuracy of the fire point discrimination are improved; an adjustment factor is added to adjust the steepness of the threshold function, so that the function reaches the hard threshold function state more quickly after passing the threshold point, the influence of the constant deviation between the processed wavelet coefficient and the original value is minimized, the reconstructed signal is maximally approximated to the real signal, and the accuracy of the fire point discrimination is improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of millimeter wave imaging, and more particularly, relates to a method and system for detecting and identifying fire points in millimeter wave radiation images. BACKGROUND

[0002] Millimeter wave refers to electromagnetic waves with a wavelength between 1mm and 10mm. A passive millimeter wave imaging system forms an image by passively detecting the thermal radiation energy naturally existing from a target scene.

[0003] The resolution of millimeter wave radiation images of fire scenes is low. Common processing methods include removing noise and super-resolution image reconstruction, and then identifying fire points in the image. In general, mean filtering, wavelet denoising and other methods are used to filter millimeter wave images. The smoothing processing of mean filtering easily leads to unclear fire point features at the edges of low-brightness objects, resulting in missed detection and inaccurate fire point detection results. In the wavelet denoising method, the hard threshold filtering of wavelets has jump points, which can lead to edge discontinuity and thus affect image quality and reduce the accuracy of fire point identification. In the wavelet soft threshold function and some existing improved wavelet threshold denoising methods, the former has a constant deviation between the wavelet coefficient and the original value, and the latter still has a long interval after the threshold value where the wavelet coefficient and the original value have an unequal deviation. Too much deviation can lead to low signal accuracy, affecting the approximation degree of the reconstructed signal to the true signal, and also affecting the image quality and the accuracy of fire point identification. At the same time, a two-dimensional turntable is used with a millimeter wave radiometer system. During imaging, similar to the characteristics of fire points, interference points may be generated due to human factors or uncontrollable external interference factors, which can be misjudged as fire points, resulting in inaccurate fire point detection results. SUMMARY

[0004] In view of the above defects or improvement needs of the prior art, the present application provides a method for detecting and identifying fire points in millimeter wave radiation images, which aims to solve the technical problems of missed detection and misjudgment of fire points.

[0005] To achieve the above-mentioned purpose, according to one aspect of the present application, a method for detecting and identifying fire points in millimeter wave radiation images is provided, comprising:

[0006] S1. Obtain a gray value matrix corresponding to the millimeter wave radiation image of the fire area;

[0007] S2. Preprocess the gray value matrix using a wavelet threshold denoising method, and then perform super-resolution reconstruction; wherein the threshold function is:

[0008]

[0009] ε is a threshold value; a, b are adjustment factors; sign(*) is a sign function; e is a constant, and w is an initial wavelet coefficient;

[0010] S3. Traverse each row of pixel gray values of the image, in each row of pixel gray values, according to a first preset threshold value, screen out suspected fire points; according to a second preset threshold value, exclude interference points in the suspected fire points, to obtain confirmed fire points; according to a gradient change relationship of the pixel gray values, determine upper and lower boundary points of the fire points;

[0011] S4. After transposing the gray value matrix, the same row operation processing as in step S3 is performed;

[0012] S5. The confirmed fire point pixel points and the fire point boundary points obtained in steps S3 and S4 are taken respectively, and the union is obtained, to obtain a complete image fire point pixel point set and a fire point boundary pixel point set.

[0013] Further, step S1 comprises:

[0014] The fire field region is scanned by using the calibrated millimeter wave radiometer, to obtain a voltage matrix corresponding to the fire power scene;

[0015] The voltage matrix is converted into a gray value matrix of 0-255.

[0016] Further, step S2 specifically comprises,

[0017] A suitable wavelet basis and decomposition layer number are selected to perform wavelet decomposition on the obtained original image, to obtain initial wavelet coefficients;

[0018] The initial wavelet coefficients are processed by using the threshold value function;

[0019] The processed wavelet coefficients are subjected to wavelet reconstruction, to obtain a denoised image;

[0020] The filtered image is processed by using a bicubic interpolation method, to obtain a high-resolution millimeter wave radiation image.

[0021] The application further provides a system for detecting and identifying fire points in a millimeter wave radiation image, comprising:

[0022] A data acquisition module acquires a gray value matrix corresponding to a millimeter wave radiation image of a fire field region;

[0023] A wavelet threshold denoising module adopts a wavelet threshold denoising method to pre-process the gray value matrix, and then performs super-resolution reconstruction; wherein, a threshold value function is:

[0024]

[0025] ε is a threshold value; a, b are adjustment factors; sign(*) is a sign function; e is a constant, and w is an initial wavelet coefficient;

[0026] The fire point boundary preliminary screening module traverses each row of pixel gray values of the image, and in each row of pixel gray values, according to a first preset threshold value, screens out suspected fire points; according to a second preset threshold value, excludes interference points in the suspected fire points, to obtain confirmed fire points; according to a gradient change relationship of the pixel gray values, determines upper and lower boundary points of the fire points;

[0027] The fire point boundary secondary screening module performs the same row operation processing as in the fire point boundary preliminary screening module after transposing the gray value matrix;

[0028] The fire point boundary acquisition module takes the union of the confirmed fire point pixel points and the fire point boundary points obtained by the fire point boundary preliminary screening module and the fire point boundary secondary screening module, to obtain a complete image fire point pixel point set and a fire point boundary pixel point set.

[0029] Further, the data acquisition module execution process includes:

[0030] The fire field area is scanned by using the calibrated millimeter wave radiometer, to obtain a voltage matrix corresponding to the fire power scene;

[0031] The voltage matrix is converted into a gray value matrix of 0-255.

[0032] Further, the wavelet threshold denoising module execution process includes,

[0033] A suitable wavelet basis and decomposition layer number are selected to perform wavelet decomposition on the obtained original image, to obtain initial wavelet coefficients;

[0034] The threshold function is used to process the initial wavelet coefficients;

[0035] The wavelet coefficients after processing are subjected to wavelet reconstruction, to obtain a denoised image;

[0036] The filtered image is processed by using a bicubic interpolation method, to obtain a high-resolution millimeter wave radiation image.

[0037] Overall, compared with the prior art, the above technical solutions conceived by the present application can achieve the following beneficial effects.

[0038] (1) The improved wavelet threshold denoising method is adopted when the millimeter wave radiation image is preprocessed, the improved threshold function has the advantages that: continuous at the threshold, the influence caused by the discontinuity of the image edge part due to local jitter is eliminated, the quality of the image and the accuracy of the fire point discrimination are improved; the steepness of the threshold function is adjusted by increasing the adjustment factor, so that the function reaches the hard threshold function state more quickly after passing the threshold point, the influence of the constant deviation between the processed wavelet coefficient and the original value is minimized, the reconstructed signal is maximized to approximate the real signal, and the accuracy of the fire point discrimination is improved.

[0039] (2) The image is traversed row by row, and the suspected fire point is obtained by comparing with the gray value threshold, and then the interference point is excluded by the gray value gradient threshold, so that the accuracy of fire point recognition is improved; the image is also traversed column by column, specifically, the gray value matrix is transposed, and then the same row operation processing is performed, the same function can be called for operation, and no additional function needs to be established; the row traversal and column traversal are processed together, which can make up for the problem that the incomplete fire point cannot detect the fire point boundary due to the image edge when the row traversal or column traversal is processed alone, the row and column processing can improve the accuracy of fire point boundary detection and reduce the possibility of missed detection. The accurate fire point and image boundary are obtained by processing and marked in the millimeter wave radiation image. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 is a flowchart of a millimeter wave radiation image fire point detection and recognition method provided by an embodiment of the present application;

[0041] Figure 2 is a comparison diagram of a hard threshold function, a soft threshold function and an improved threshold function curve provided by an embodiment of the present application;

[0042] Figure 3 is an initial millimeter wave radiation image F(i,j) provided by an embodiment of the present application;

[0043] Figure 4 is a filtered image obtained by filtering the initial millimeter wave radiation image F(i,j) by the improved wavelet threshold function provided by an embodiment of the present application;

[0044] Figure 5 is a high-resolution millimeter wave fire point image F(i',j') after preprocessing provided by an embodiment of the present application.

[0045] Figure 6 is a pixel gray value curve diagram of the millimeter wave radiation image F(i',j') provided by an embodiment of the present application.

[0046] Figure 7 is a millimeter wave radiation image of a fire point with interference points provided by the embodiment.

[0047] Figure 8 The 53rd row pixel gray value curve and the pixel gray gradient curve chart of the millimeter wave radiation image with the interference point provided in the embodiment are shown in the following figure.

[0048] Figure 9 The fire point detection and recognition chart of the preprocessed high-resolution millimeter wave radiation image F'(i',j') provided in the embodiment is shown in the following figure.

[0049] Figure 10 The fire point detection and recognition chart of the millimeter wave radiation image with the interference point provided in the embodiment is shown in the following figure. DETAILED DESCRIPTION

[0050] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0051] As shown in the following figure, a millimeter wave radiation image fire point detection and recognition method includes the following steps: Figure 1

[0052] S1. Obtain the gray value matrix corresponding to the millimeter wave radiation image of the fire area;

[0053] A standard radiation source is selected to calibrate the millimeter wave radiometer to determine the relationship between the output voltage of the radiometer and the object radiation brightness temperature.

[0054] The millimeter wave radiometer calibration method includes single-point calibration method, two-point calibration method, multi-point calibration method, noise injection calibration method, and tilt curve calibration method. The relationship between the output voltage of the radiometer and the object radiation brightness temperature is determined by calibration, and the object radiation brightness temperature is inverted according to the obtained voltage data. In the embodiment, the receiver channel response of the radiometer is linear, and the two-point calibration method is used to calibrate the output voltage of the receiver of the radiometer.

[0055] First, the output voltage of each channel when aligned with the cold and hot source is obtained, and the linear relationship between the output voltage of each channel and the object radiation brightness temperature is obtained.

[0056] The millimeter wave radiometer is aligned with the standard cold source T1, and the output voltage of the radiometer receiver is obtained as V1.

[0057] The millimeter wave radiometer is aligned with the standard hot source T2, and the output voltage of the radiometer receiver is obtained as V2.

[0058] ​The linear relationship between the output voltage U of the computing radiometer receiver and the object radiation brightness temperature T is calculated: U=A*T+B, wherein The gain coefficient representing the channel response of the radiometer receiver; The bias coefficient representing the channel response of the radiometer receiver.

[0059] After the millimeter wave radiometer calibration is completed, the passive millimeter wave radiometer system is used to scan the fire field region combined with a two-dimensional turntable to obtain m*n fire point scene voltage data, m is the number of scanning rows, and n is the number of points scanned in each row.

[0060] According to the pixel value corresponding to the voltage size U at any position (i, j) in the initial millimeter wave radiation image F(i, j) of the fire point scene i,j , the elements in the voltage data set are arranged in the form of a matrix according to their subscripts, and the voltage data set corresponding to the scanning scene region radiation brightness temperature is: [U 1,1 ,…,U 1,n ; U 2,1 ,…,U 2,n ;…; U m,1 ,…,U m,n ], wherein 1≤i≤m, 1≤j≤n.

[0061] When the voltage matrix of the image is converted into a 0-255 grayscale value matrix, the data with higher voltage values are set to correspond to higher grayscale values, and the data with lower voltage values are set to correspond to lower grayscale values. The conversion processing formula is f(i, j)=(U i,j -U min )×255 / (U max -U min ), wherein 1≤i≤m, 1≤j≤n, f(i, j) and U i,j are the grayscale value size and the voltage value size corresponding to the data at any position (i, j) in the initial millimeter wave radiation image F(i, j), U max is the maximum value in the voltage matrix, and U min is the minimum value in the voltage matrix. Figure 3 is the initial millimeter wave radiation image F(i, j) with a resolution size of 36*48, and the corresponding image grayscale value matrix F=[f(1, 1), …, f(1, n); f(2, 1), …, f(2, n); …; f(m, 1), …, f(m, n)];

[0062] S2. A wavelet threshold denoising method is used to pre-process the grayscale value matrix, and a high-resolution millimeter wave radiation image is obtained by performing super-resolution reconstruction on the filtered image.

[0063] In the process of scanning imaging to obtain data by the millimeter wave radiometer system, noise interference exists, which makes the voltage data fluctuate, so that the initial data needs to be denoised. Common denoising methods include Gaussian filtering, median filtering, Wiener filtering, wavelet denoising, etc. In the embodiment, the initial data is first processed by using a wavelet threshold denoising method, which mainly includes three steps:

[0064] (1) Wavelet decomposition: select a suitable wavelet basis and decomposition level to perform wavelet decomposition on the obtained initial millimeter wave radiation image to obtain initial wavelet coefficients w. In the embodiment, the Coiflet wavelet is selected, and the decomposition level is 2.

[0065] (2) Threshold processing: compare and process the wavelet coefficients w after wavelet decomposition with the threshold value ε to obtain quantized wavelet coefficients w ε .

[0066] The selected threshold value is used as a demarcation point to distinguish the wavelet coefficients generated by noise from the wavelet coefficients of the image itself. The selection of the threshold value affects the accuracy and continuity of the reconstructed signal, and therefore has a great influence on the denoising effect of the image. In the wavelet domain, the coefficients corresponding to the effective signal are large, while the coefficients corresponding to the noise are small and still satisfy the Gaussian distribution. Therefore, by setting the threshold value to zero the coefficients of the signal in a certain interval in the wavelet domain, the noise can be suppressed to the greatest extent, while only slightly damaging the effective signal.

[0067] Fixed threshold estimation is more effective in denoising, but it is also easy to mistake useful signals for noise and remove them. The calculation formula of the fixed threshold value is:

[0068]

[0069] Where median(*) is the median function, and MxN is the image resolution size. Since the resolution of the millimeter wave image is generally low, the threshold value ε obtained by using the fixed threshold estimation method will be relatively small, reducing the zero interval and reducing the damage to the effective signal.

[0070] Traditional threshold functions include hard threshold function and soft threshold function. When the absolute value of a wavelet coefficient is less than a threshold value, it is considered as noise and is filtered out. When the absolute value of a wavelet coefficient is greater than a threshold value, the hard threshold function keeps the coefficient unchanged, and the soft threshold function subtracts the threshold value from the wavelet coefficient. The hard threshold function retains the local characteristics of the signal, and the hard threshold function is closer to the original signal than the soft threshold function. However, the hard threshold function is discontinuous at the threshold value, which may cause local jitter in the denoised signal and produce jumping points, and the hard threshold function does not have the smoothness of the original signal, resulting in discontinuity of the edge part and affecting the quality of the image and the accuracy of the fire point discrimination. The soft threshold function is continuous at the threshold value, and the processed signal is relatively smooth. However, the wavelet coefficient has a constant deviation from the original value, which may result in low signal accuracy and even distortion, directly affecting the approximation degree of the reconstructed signal to the true signal, and also affecting the quality of the image and the accuracy of the fire point discrimination.

[0071] In view of the shortcomings of the traditional hard threshold function and the soft threshold function, the threshold function in the embodiment adjusts the steepness of the threshold function by increasing the adjustment factor, so as to have the advantages of the hard threshold function and the soft threshold function, and to compensate for the shortcomings of the two functions to the greatest extent. The improved threshold function in the embodiment has the following advantages: continuous at the threshold value, eliminating the influence of local jitter; after passing the threshold point, it reaches the state of the hard threshold function faster, thereby minimizing the constant deviation of the processed wavelet coefficient from the original value, and maximizing the approximation of the reconstructed signal to the true signal, thereby improving the accuracy of fire point discrimination. The initial model of the improved threshold function is as follows:

[0072]

[0073] wherein sign(*) is a sign function, and g(w) is a function of w. Since f(w) is continuous at the threshold value and reaches the state of the hard threshold function faster after passing the threshold point, the following conditions should be met:

[0074]

[0075] wherein f'(w) is the derivative of f(w), and ε + is a value slightly larger than ε, and (-ε) - is a value slightly smaller than -ε. According to the property conditions met by f(w) and f'(w), the function g(w) should meet the following conditions:

[0076]

[0077] where g'(w) is the derivative of g(w), according to the conditions that g(w) satisfies, the exponential function is selected to construct g(w). The exponential function has the characteristics of steep increase, which can make the improved threshold function reach the hard threshold state faster after the threshold value; and compared with the logarithmic function, the exponential function has a wide range of independent variable values, which can take 0 value. Using the exponential function as the basic model, the expression of g(w) that satisfies the conditions is constructed as:

[0078]

[0079] In order to make the threshold function continuous at the threshold value, and can be adjusted to a hard threshold function, a regulating factor a is added for adjustment. First, it needs to satisfy that when a takes a certain value, g'(w) = 0. An item is 0 at w = ±ε, since the denominator cannot be 0, only the case of adding a to make the numerator 0 is considered. Multiply the numerator by a coefficient (a + 1) to become (a + 1)ε, when a = -1, the numerator satisfies 0; At the same time, it also needs to satisfy g(±ε) = 0, at this time the exponential term is 1, and the denominator should be 1 + ae |w|-ε . Then the second condition needs to be satisfied, when w→±∞ . According to the formula derived from the first condition to verify the second condition, it can be found that satisfies, then it is derived that the g1(w) added with the regulating factor a satisfies that the threshold function is continuous at the threshold value, and can adjust f(w) to a hard threshold function:

[0080]

[0081] In order to make the threshold function f(w) reach the hard threshold state faster after the threshold value, and minimize the influence of the constant deviation between the processed wavelet coefficient and the original value, it is necessary to make g'(ε + )→+∞ and g'((-ε) - )→-∞, calculate the derivative of g(w) at w = ε + and w = (-ε) - , which are and It can be seen that actually 1 + ε cannot directly reach +∞, and -1 - ε cannot directly reach -∞. Therefore, a multiple is added to make the modulus value become larger, which has better steepness, here a second regulating factor b is added, g'(ε + ) = 1 + bε, g'((-ε) - ) = -1 - bε, and the improved g2(w) is obtained by backstepping as:

[0082]

[0083] Combined with g1(w) and g2(w), the improved g(w) expression is obtained as:

[0084]

[0085] The improved threshold function is then obtained as follows:

[0086]

[0087] Where ε is the threshold; a and b are adjustment factors; sign(*) is the sign function; and e is a constant with the natural base.

[0088] Comparison of hard threshold function, soft threshold function, and improved threshold function curves, for example Figure 2 As shown, curve h is the soft threshold function, curve i is the hard threshold function, curve j is the improved threshold function in this embodiment (a = 0.5, b = 1), and curve k is also the improved threshold function in this embodiment (a = 0.5, b = 10). Curve k reaches the hard threshold state faster than curve j after passing the threshold point. When b is 0, the function becomes a soft threshold function; when a is -1, it becomes a hard threshold function. By changing the values ​​of a and b, the tendency of the function can be changed, improving the adaptability, ensuring continuity at the threshold, and eliminating the influence of local jitter. Comparing curves k and j, it can be seen that after passing the threshold point, increasing the value of the adjustment factor b can reach the hard threshold function state faster, minimizing the influence of the constant deviation between the processed wavelet coefficients and the original values, making the reconstructed signal approximate the real signal as closely as possible, and improving the accuracy of fire point detection.

[0089] (3) Wavelet reconstruction: The processed wavelet coefficients w ε Perform an inverse transform to calculate the wavelet reconstruction of the two-dimensional signal, and obtain the denoised image. Figure 4 The filtered image is obtained by filtering the initial millimeter-wave radiation image F(i,j) using an improved wavelet threshold function. (Comparison) Figure 4 and Figure 3 As can be seen, the improved wavelet thresholding denoising method has a good effect, removing a lot of noise and making the outline of the fire point clearer.

[0090] After filtering and denoising, a higher-resolution millimeter-wave radiation image is obtained using a super-resolution reconstruction algorithm. Super-resolution algorithms based on interpolation reconstruction include nearest-neighbor interpolation, bilinear interpolation, and bicubic interpolation. In this embodiment, bicubic interpolation is used for interpolation reconstruction. This algorithm uses the gray values ​​of 16 points surrounding the sampling point for cubic interpolation, taking into account not only the gray value influence of the four directly adjacent points but also the influence of the rate of change of gray values ​​between neighboring points. After processing, the image resolution is improved, and the image becomes clearer. The preprocessed high-resolution millimeter-wave radiation image is denoted as F′(i′,j′), with a resolution of 72×96. Figure 5The corresponding gray value matrix F' = [f'(1,1), f'(1,2), f'(1,3), f'(1,4), f'(1,5), f'(1,6), f'(1,7), f'(1,8); f'(2,1), f'(2,2), f'(2,3), f'(2,4), f'(2,5), f'(2,6), f'(2,7), f'(2,8);...; f'(2m,1), f'(2m,2), f'(2m,3), f'(2m,4), f'(2m,5), f'(2m,6), f'(2m,7), f'(2m,8)] is shown.

[0091] [f'(1,1), f'(1,2), f'(1,3), f'(1,4), f'(1,5), f'(1,6), f'(1,7), f'(1,8); f'(2,1), f'(2,2), f'(2,3), f'(2,4), f'(2,5), f'(2,6), f'(2,7), f'(2,8);...; f'(2m,1), f'(2m,2), f'(2m,3), f'(2m,4), f'(2m,5), f'(2m,6), f'(2m,7), f'(2m,8)], where 1≤i'≤2m, 1≤j'≤2n.

[0092] S3. Traversing each row of pixel gray values of the image, in each row of pixel gray values, according to a first preset threshold value, screening out suspected fire points; according to a second preset threshold value, excluding interference points in the suspected fire points to obtain confirmed fire points; according to a pixel gray value gradient change relationship, determining upper and lower boundary points of the fire points;

[0093] According to the feature that the gray value of the fire point in the scanned scene millimeter wave radiation image is higher than the surrounding background gray value, the suspected fire points are screened out. In the actual scene, the temperature of the fire area is not uniform and shows a certain rule, the pixel gray value of the fire point shows an increasing rule from the edge to the center, according to this feature, commonly used methods for fire point detection include the window splitting method, image edge detection technology, binocular stereo vision and other methods.

[0094] In the embodiment, the fire point boundary is confirmed by judging and comparing the gray value and the gray gradient value.

[0095] First, each row of the gray value matrix F' is traversed to obtain a pixel gray value curve function of each row, for example, the pixel gray value curve f'(i', :) = {f'(i', 1), f'(i', 2), f'(i', 3), f'(i', 4), f'(i', 5), f'(i', 6), f'(i', 7), f'(i', 8)} of the i'th row, where 1≤i'≤2m, 1≤j'≤2n.

[0096] The pixel point (i', j') in the i'th row that satisfies f'(i', j') > η1 is determined as a suspected fire point and recorded to the suspected fire point set φ A According to this threshold condition, the range of the suspected fire points can be reduced; Figure 6 The pixel gray value size curve of the 43rd, 48th and 53rd rows of the image F(i', j') in the embodiment, the dashed line A in the figure represents the set pixel gray value threshold value η1 = 200, the 53rd row is determined as having a suspected fire point, and the 43rd and 48th rows are determined as not having a suspected fire point.

[0097] When the millimeter wave radiometer scans the fire point scene, due to improper operation or uncontrollable external factors, the area in the image where no fire occurs may generate small area pixel gray value jitter points that are slightly high due to the interference of adjacent fire points. Figure 7The fire point microwave image provided in the embodiment has an interference point. The area size where the interference point appears and the sharp change in the boundary thereof are considered. While traversing each row of the image gray value matrix F', the gray gradient curve function df'(i', :) of each row of pixels is compared with the set gray value gradient threshold value η2, and the jitter point in the suspected fire point pixel is excluded, so as to improve the accuracy of fire point identification.

[0098] The gray gradient curve function df'(i', :) of the i'th row of pixels is df'(i', :) = {df'(i', 1), df'(i', 2),..., df'(i', 2n)}, where 1≤i'≤2m, df'(i', j') = f'(i', j') - f'(i', j'-1), 1<j'≤2n, and df'(i', 1) = 0.

[0099] The jitter interference point has a sharp change in the pixel gray value, and the corresponding gray value gradient df'(i', j') is N . N The maximum value of the module |df'(i', j')| is greater than the set gray value gradient threshold value η2, and the corresponding pixel is the jitter point. A The relationship between the gray value gradient module |df'(i', j')| of each pixel point and the threshold value η2 is compared in the suspected fire point set φ A .

[0100] Figure 8 The 53rd row of the fire point millimeter wave radiation image provided in the embodiment has a pixel gray value curve and a pixel gray gradient curve. The dotted line A represents the set pixel gray value threshold value η1 = 200, the dotted line B represents the set pixel gray gradient threshold value η2 = 40, and the dotted line B' represents the negative value of the set pixel gray gradient threshold value, which is -η2 = -40. The pixel gray value of the interference point is greater than the pixel gray value threshold value η1 = 200, and it is determined as a suspected fire point. Further, the pixel gray gradient is used for judgment. Here, the gray value gradient threshold value η2 = 40, the area where the interference point appears is very small, the module value of the pixel gray gradient is large, and it is greater than the threshold value η2 = 40. It is determined as an interference point. The point is excluded from the suspected fire point.

[0101] Subsequently, in the confirmed fire point pixel in each row, the fire point center and the boundary pixel position are identified according to the characteristics of the fire point pixel value gray and the characteristics of the fire point edge pixel value gradient.

[0102] Taking the i′ row as an example, when (i′,j′), (i′,j′-1), (i′,j′+1)∈φ′, and satisfy the two conditions f′(i′,j′)≥f′(i′,j′-1) and f′(i′,j′)≥f′(i′,j′+1), i.e., f′(i′,j′) is a maximum point, the value of the pixel gray-level gradient curve function df′(i′,:) at (i′,j′) in the i′ row tends to 0, then the pixel (i′,j′) is determined to be the center of the fire point in the environment.

[0103] The change in pixel grayscale gradient values ​​from the center of the fire point to its left boundary is as follows: df′(i′,j′)→0, increases to its maximum value, and then decreases back to df′(i′,j′). L )→0, which means the first leftmost df′(i′,j′) closest to the center of the fire point. L The pixel position corresponding to 0 is the left boundary of the fire point. Based on this characteristic, the left boundary pixel (i′, j′) of the fire point is found. L The change in pixel grayscale gradient values ​​from the center of the fire point to its right boundary is as follows: df′(i′,j′)→0, decreasing to the minimum value, and then increasing to df′(i′,j′). R )→0, which means the first rightmost df′(i′,j′) closest to the center of the fire point. L The pixel position corresponding to 0 is the right boundary of the fire point. Based on this characteristic, the right boundary pixel (i′, j′) of the fire point is found. R ).

[0104] Perform the same processing on all rows, and record the pixels (i′, j′) that satisfy the fire point boundary determination condition into set φ. edge middle.

[0105] S4. After transposing the grayscale matrix, perform the same row operation as in step S3;

[0106] Transposed image F T The gray value matrix corresponding to (j′,i′) is:

[0107] F ′T =[f′(1,1),…,f′(1,2m); f′(2,1),…,f′(2,2m);…;f′(2n,1),…,f′(2n,2m)], 1≤i′≤2m, 1≤j′≤2n.

[0108] Firstly, it is judged whether the pixel point in each row satisfies f'(j', i')>η1, and the pixel point (j', i') satisfying the condition is determined as a suspected fire point; then it is judged whether the pixel gray gradient corresponding to the suspected fire point satisfies |df'|<η2, and the pixel point (j', i') satisfying the condition is determined as a fire point pixel; subsequently, in the confirmed fire point pixel in each row, the center of the fire point is found and the position of the fire point boundary pixel is identified according to the characteristics of the fire point pixel value gray and the characteristics of the fire point edge pixel value gradient by using the fire point center determination method and the fire point boundary determination method in step S3.

[0109] Image F T (j', i'), corresponding to the upper and lower boundary pixel points (i', j') of the fire point in the millimeter wave radiation image F(i', j'), after the boundary pixel points (j', i') of the transposed image are changed to (i', j'), the changed boundary pixel points (i', j') are recorded to the set φ' edge .

[0110] S5. The confirmed fire point pixel and the fire point boundary point obtained in steps S3 and S4 are taken as a union respectively, and the complete image fire point pixel and the fire point boundary pixel are obtained.

[0111] Figure 9 is the fire point detection and identification image of the preprocessed high-resolution millimeter wave radiation image F'(i', j') provided by the embodiment, Figure 10 is the fire point detection and identification image of the millimeter wave radiation image with interference points provided by the embodiment. The boundary pixel point sets of the transposed and non-transposed images are taken as a union, and the complete image boundary pixel point set φ is obtained, and the calculation formula is φ=φ edge ∪φ' edge , and the fire point boundary is accurately marked in the millimeter wave radiation image F'(i', j') according to the set φ.

[0112] Those skilled in the art can easily understand that the above description is only the preferred embodiment of the present application, and is not used to limit the present application, and any modification, equivalent replacement and improvement made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for millimeter wave radiation image fire point detection and identification, characterized in that, The method comprises the following steps: S1. Obtain a gray value matrix corresponding to a millimeter wave radiation image of a fire area; S2. Preprocess the gray value matrix by using a wavelet threshold denoising method, and then perform super-resolution reconstruction; wherein the threshold function is: ε is a threshold value; a and b are adjustment factors; sign(*) is a sign function; e is a constant, and w is an initial wavelet coefficient; S3. Traverse each row of pixel gray values of the image, and in each row of pixel gray values, according to a first preset threshold value, screen out suspected fire points; according to a second preset threshold value, exclude interference points in the suspected fire points to obtain confirmed fire points; and according to a gradient change relationship of the pixel gray values, determine upper and lower boundary points of the fire points; S4. After transposing the gray value matrix, perform the same row operation processing as in step S3; S5. Take the union of the confirmed fire point pixel points and the fire point boundary points obtained in steps S3 and S4 respectively to obtain a complete image fire point pixel point set and a fire point boundary pixel point set.

2. The method of claim 1, wherein, Step S1 comprises: Scanning a fire area by using a calibrated millimeter wave radiometer to obtain a voltage matrix corresponding to a fire point scene; Convert the voltage matrix into a gray value matrix of 0-255.

3. The method of claim 1, wherein, Step S2 specifically comprises: Selecting a suitable wavelet basis and decomposition level to perform wavelet decomposition on the obtained original image to obtain an initial wavelet coefficient; Processing the initial wavelet coefficient by using the threshold function; Performing wavelet reconstruction on the processed wavelet coefficient to obtain a denoised image; Processing the filtered image by using a bicubic interpolation method to obtain a high-resolution millimeter wave radiation image.

4. A system for millimeter wave radiation image fire point detection and identification, characterized by, The method comprises the following steps: A data acquisition module is configured to obtain a gray value matrix corresponding to a millimeter wave radiation image of a fire area; A wavelet threshold denoising module is configured to preprocess the gray value matrix by using a wavelet threshold denoising method, and then perform super-resolution reconstruction; wherein the threshold function is: ε is a threshold value; a and b are adjustment factors; sign(*) is a sign function; e is a constant, and w is an initial wavelet coefficient; A fire point boundary primary screening module is configured to traverse each row of pixel gray values of the image, and in each row of pixel gray values, according to a first preset threshold value, screen out suspected fire points; according to a second preset threshold value, exclude interference points in the suspected fire points to obtain confirmed fire points; and according to a gradient change relationship of the pixel gray values, determine upper and lower boundary points of the fire points; A fire point boundary secondary screening module is configured to perform the same row operation processing as in the fire point boundary primary screening module after transposing the gray value matrix; A fire point boundary acquisition module is configured to take the union of the confirmed fire point pixel points and the fire point boundary points obtained by the fire point boundary primary screening module and the fire point boundary secondary screening module respectively to obtain a complete image fire point pixel point set and a fire point boundary pixel point set.

5. The system for millimeter wave radiation image fire point detection and identification according to claim 4, characterized in that, The execution process of the data acquisition module comprises: Scanning a fire area by using a calibrated millimeter wave radiometer to obtain a voltage matrix corresponding to a fire point scene; Convert the voltage matrix into a gray value matrix of 0-255.

6. The system for millimeter wave radiation image fire point detection and identification according to claim 4, wherein, The execution process of the wavelet threshold denoising module comprises: Selecting a suitable wavelet basis and decomposition level to perform wavelet decomposition on the obtained original image to obtain an initial wavelet coefficient; Processing the initial wavelet coefficient by using the threshold function; Wavelet reconstruction is performed on the processed wavelet coefficients to obtain the denoised image; The filtered image is processed by using a bicubic interpolation method to obtain a high-resolution millimeter wave radiation image.

7. A computer readable medium having stored thereon a computer program which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 3.

Images