Infrared temperature image enhancement method and system based on image processing

Abstract

The present invention proposes an infrared temperature image enhancement method and system based on image processing, including: obtaining an infrared temperature image, applying a multi-scale image processing technique to extract data at different frequencies and scales in the infrared temperature image, normalizing the data respectively, and constructing a neural network model; performing image enhancement at multiple scales, inputting the pre-processed infrared temperature images at multiple scales into the neural network model, designing a Q-learning algorithm, learning and updating the Q value of the infrared temperature image under different enhancement strategies; selecting and executing the best enhancement strategy to train the neural network model, and optimizing the loss function of the neural network model; fusing the enhancement results at multiple scales and performing post-processing to optimize the visual effect of the image; and outputting the enhanced infrared temperature image. The present invention uses a Q-learning algorithm to design a reward function, continuously optimizes and enhances the loss function generated by recognition, and improves the recognition accuracy of image processing.

Description

Technical Field

[0001] The invention belongs to the field of image processing, and in particular relates to an infrared temperature image enhancement method and system based on image processing. Background Art

[0002] The current status of image processing shows a steady growth trend, and the market size is expected to reach a considerable level in the next few years. In the image processing software market, there are many major players, including well-known technology companies such as Adobe, Corel and GIMP, as well as some emerging startups and projects in the open source community.

[0003] With the continuous development of science and technology, people's demand for image processing software is also growing. At present, the market demand is mainly focused on ease of use, versatility, high performance and effects. For example, users hope to be able to use image processing software easily without too much technical knowledge; at the same time, the market demand for software that can process various types of images and provide diversified functions is increasing; in addition, users have higher requirements for the performance and effects of image processing software, such as image quality, speed and accuracy.

[0004] However, although image processing technology has made significant progress, some challenges and shortcomings still exist. Image segmentation is a complex process that requires efficient algorithms and techniques to improve the accuracy of segmentation.

[0005] The current status of external temperature image enhancement is reflected in continuous technological progress and wide application. With the continuous development of infrared technology, people have higher and higher requirements for the clarity of infrared images, prompting researchers to conduct a lot of research on improving the contrast of infrared images and enhancing the detailed pixels of images. These studies cover various algorithms and methods of infrared image enhancement, aiming to improve the visual quality and recognition accuracy of infrared images.

[0006] However, the infrared thermal image represents the temperature distribution of the scene. It is a grayscale image without color or shadow, low resolution and poor resolution potential. This makes the infrared image inferior to the visible light image in detail expression. The infrared image has strong spatial correlation, low contrast and blurred visual effect, which further affects the recognition accuracy and visual effect of the infrared image. The clarity of the infrared image is lower than that of the visible light image, which also limits its application in some scenes that require high-precision recognition.

[0007] Therefore, there is an urgent need for a suitable enhancement method and algorithm to maximize the visual quality and recognition accuracy of infrared images. Summary of the invention

[0008] The purpose of the present invention is to propose an infrared temperature image enhancement method and system based on image processing, combine the image enhancement method of neural network and AI, use Q-learning algorithm to design reward function, continuously optimize training in the originally designed neural network model, and continuously optimize and enhance the loss function generated by recognition, so as to improve the recognition accuracy of image processing.

[0009] In order to achieve the above object, a first aspect of the present invention provides an infrared temperature image enhancement method based on image processing, the method comprising:

[0010] S1. Obtain an infrared temperature image, apply multi-scale image processing technology to extract data at different frequencies and scales in the infrared temperature image, normalize the data respectively, and construct a neural network model; wherein the multi-scale image processing technology includes:

[0011] S11, applying wavelet transform to perform multi-scale decomposition on the image to obtain wavelet coefficients at different scales;

[0012] S12, enhancing the wavelet coefficients at each scale to improve the information expression of the image at different scales;

[0013] S13, reconstructing the enhanced wavelet coefficients into an enhanced multi-scale infrared temperature image through inverse wavelet transform;

[0014] S2. Perform image enhancement at multiple scales, input the preprocessed infrared temperature images at multiple scales into the neural network model, design a Q-learning algorithm, learn and update the Q value of the infrared temperature image under different enhancement strategies;

[0015] S3. According to the Q value learned by the Q-learning algorithm, select and execute the best enhancement strategy to train the neural network model, and optimize the loss function of the neural network model; where the loss function L is expressed as

[0016]

[0017] Where y represents the target Q value, which is calculated based on the reward (r) and the maximum Q value of the next state; Q(s,a) is the model's prediction of the Q value for the current state s and action a;

[0018] S4. The multiple multi-scale infrared temperature images output by the neural network model are post-processed by fusing the enhancement results at multiple scales using image fusion technology to optimize the visual effect of the image; wherein the image fusion technology fuses the enhancement results at different scales according to the weights, which is expressed as follows:

[0019]

[0020] Where F represents the fused image, c represents the number of multi-scale infrared temperature images involved in the fusion, and w k represents the weight of the kth multi-scale infrared temperature image enhancement, E i represents the kth multi-scale infrared temperature enhanced image;

[0021] S5. Output the enhanced infrared temperature image.

[0022] Furthermore, the Q-learning algorithm includes:

[0023] A. Initialize the Q value table and assign an initial Q value to each scale infrared temperature image state-action pair, as shown below:

[0024] Q(s,a)=E[R_ t+1 +γR_ t+2 +γ^2R_ t+3 +…|S_t=s,A_t=a], where s represents the current state, a represents the action taken, R_t represents the immediate reward obtained at time t, γ is the discount factor, S_t represents the current state, and the action A_t is executed;

[0025] B. Select and execute an enhancement strategy based on the state of the current scale infrared temperature image;

[0026] C. Observe the result after executing the enhancement strategy, that is, the state of the next pixel or pixel block and the corresponding reward value;

[0027] D. Fusion evaluation results of the image T otal As a reward, update the Q value table, update the formula V(s) = max[R(s,a)+γ*V(s')], where V(s) represents the value function of state s, R(s,a) represents the reward obtained by taking action a in state s, γ is a discount factor used to weigh the importance of current rewards and future rewards, and V(s') represents the value function of the subsequent state s';

[0028] E. Repeat steps B to D until the Q value table converges or the preset number of iterations is reached.

[0029] Furthermore, the process of reconstructing the enhanced wavelet coefficients into an enhanced multi-scale infrared temperature image through inverse wavelet transform specifically includes:

[0030] a. Calculate the regional energy required for decomposition of the multi-scale infrared temperature image, expressed as follows:

[0031]

[0032] in, represents the regional energy centered at m*n, σ=1,2,3 represent the horizontal, vertical and diagonal directions respectively, i represents the i-th layer, Represents the corresponding wavelet coefficient matrix L i The weight coefficient, L i represents the wavelet coefficient matrix corresponding to the i-th layer, (m', n') represents the corresponding matrix, and K and J represent the range of variation corresponding to m' and n' respectively;

[0033] b. Calculate the matching degree Match between the multi-scale infrared temperature image and the infrared temperature image i (m, n), expressed as follows:

[0034]

[0035] Among them, L i,s represents the wavelet coefficient matrix of the multi-scale infrared temperature image corresponding to the i-th layer, L i,f Represents the wavelet coefficient matrix of the infrared temperature image corresponding to the i-th layer, Represents the regional energy of the multi-scale infrared temperature image centered at m*n, Indicates the energy of the area centered at m*n in the infrared temperature image;

[0036] c. Determine the fusion operator to calculate the wavelet coefficients, and then reconstruct the wavelet coefficients into an enhanced multi-scale infrared temperature image through inverse wavelet transform.

[0037] Furthermore, the determination fusion operator calculates the wavelet coefficients, which are expressed as follows:

[0038] Determine the matching threshold T, if Match i (m, n)<T, then it is expressed as

[0039]

[0040] If Match i (m, n) ≥ T, which is expressed as follows:

[0041]

[0042] Then there is,

[0043]

[0044] in, represents the wavelet coefficients of the fused image, The wavelet coefficients representing the multi-scale infrared temperature image, Represents the wavelet coefficients of the infrared temperature image, Represents the maximum weight coefficient of the corresponding wavelet coefficient matrix, Represents the minimum weight coefficient of the corresponding wavelet coefficient matrix.

[0045] Furthermore, the range of T includes 0.5~1.

[0046] Furthermore, m*n includes 3×3, 5×5 and 7×7.

[0047] Furthermore, the Q value table is updated according to the result of the image fusion evaluation, which is expressed as follows:

[0048] The state S represents the characteristics of the infrared temperature image or the multi-size infrared temperature image, and the action A represents the adjustment of the weight coefficient in the wavelet coefficient. The relevant actions are performed according to the characteristics of the infrared temperature image or the multi-size infrared temperature image, and the Q value table is updated, which is expressed as follows:

[0049] Q[s, a]=(1-β)*Q[s, a]+β*(R+γ*max(Q[s′,a']))

[0050] Among them, β represents the learning rate, which controls the influence of the new image fusion result on the old Q value; γ represents the discount factor, which determines the importance of future rewards; s' represents the new state after executing action a, and max(Q[s',a']) represents the Q value of the action with the largest expected return among all possible actions a' taken in the new state.

[0051] Furthermore, the learning rate β also follows the greedy strategy, which is expressed as follows:

[0052]

[0053] Among them, P represents probability, ∈ represents the selection of RandomAction random action with probability ∈, and argmaxQ(a) represents the selection of the action that maximizes the Q value within a certain range with probability 1-∈.

[0054] Furthermore, the image fusion evaluation indicators include entropy, mean, standard deviation, and visibility;

[0055] The image fusion evaluation is calculated as follows:

[0056] The entropy Q is calculated as follows:

[0057]

[0058] in, The grayscale value of the current multi-size infrared temperature image, Indicates the ratio of the number of pixels of the multi-size infrared temperature image to the total number of pixels;

[0059] The mean MEAN is calculated as follows:

[0060]

[0061] Wherein, f(i, j) represents the region coordinates of the multi-scale infrared temperature image, and m×n represents the size of the multi-scale infrared temperature image;

[0062] The standard deviation STD is calculated as follows:

[0063]

[0064] The Visibility VI is calculated as follows:

[0065]

[0066] Among them, μ represents the visual constant;

[0067] The image fusion evaluation index is expressed as follows:

[0068]

[0069] In a second aspect of the present invention, an infrared temperature image enhancement system based on image processing is provided, the system comprising:

[0070] The image acquisition unit is used to acquire an infrared temperature image, apply a multi-scale image processing technique to extract data at different frequencies and scales in the infrared temperature image, normalize the data respectively, and construct a neural network model; wherein the multi-scale image processing technique includes:

[0071] S11, applying wavelet transform to perform multi-scale decomposition on the image to obtain wavelet coefficients at different scales;

[0072] S12, enhancing the wavelet coefficients at each scale to improve the information expression of the image at different scales;

[0073] S13, reconstructing the enhanced wavelet coefficients into an enhanced multi-scale infrared temperature image through inverse wavelet transform;

[0074] The image enhancement unit is used to perform image enhancement at multiple scales. The preprocessed infrared temperature images at multiple scales are input into the neural network model, and a Q-learning algorithm is designed to learn and update the Q value of the infrared temperature image under different enhancement strategies.

[0075] The model updating unit is used to select and execute the best enhancement strategy to train the neural network model according to the Q value learned by the Q-learning algorithm, and optimize the loss function of the neural network model; the loss function L is expressed as

[0076]

[0077] Where y represents the target Q value, which is calculated based on the reward (r) and the maximum Q value of the next state; Q(s,a) is the model's prediction of the Q value for the current state s and action a;

[0078] The image fusion unit is used to use the image fusion technology to fuse the enhancement results at multiple scales of the multiple multi-scale infrared temperature images output by the neural network model and then perform post-processing to optimize the visual effect of the image; wherein the image fusion technology fuses the enhancement results at different scales according to the weights, which is expressed as follows:

[0079]

[0080] Where F represents the fused image, c represents the number of multi-scale infrared temperature images involved in the fusion, and w k represents the weight of the kth multi-scale infrared temperature image enhancement, E i represents the kth multi-scale infrared temperature enhanced image;

[0081] The image output unit is used to output the enhanced infrared temperature image.

[0082] The beneficial technical effects of the present invention are at least as follows:

[0083] (1) The application of the neural network of the present invention provides a powerful learning ability for image enhancement. By training the neural network, it can learn various complex mapping relationships required for image enhancement, thereby realizing automatic enhancement of the image. Compared with traditional image enhancement methods, this method has higher flexibility and adaptability and can handle various complex image enhancement tasks.

[0084] (2) The application of the Q-learning algorithm makes the training process of the neural network model more efficient and accurate. By designing a reward function, the Q-learning algorithm can guide the neural network model to gradually optimize its parameters during the training process, so that the performance of the model in image enhancement tasks is continuously improved. This method can accelerate the convergence speed of the model and improve the generalization ability of the model, so that the model can maintain high performance when processing new data.

[0085] (3) Setting up comprehensive evaluation of entropy, mean, standard deviation, and visibility to update the Q-learning algorithm. Entropy, as an indicator of information richness, can help the algorithm identify the most information-rich parts of the image, thereby guiding the algorithm to perform more refined processing in these areas. Standard deviation reflects the dispersion of the grayscale of the image. When the image is subject to noise or other interference, changes in the standard deviation can help the algorithm detect these changes and adjust its strategy accordingly. By comprehensively evaluating these four indicators, the algorithm can more quickly identify the factors that have the greatest impact on the image processing effect, thereby conducting more targeted learning. This helps to accelerate the convergence speed of the algorithm, reduce training time, and improve learning efficiency. Visibility, as an indicator of image clarity, can guide the algorithm to maintain its clarity and recognizability when enhancing the image. At the same time, indicators such as mean and standard deviation can also help the algorithm adjust the enhancement strategy to maximize the improvement of image quality. These factors work together to enable the algorithm to generate higher quality enhanced images to meet the needs of various application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0087] Figure 1 This is a flow chart of the infrared temperature image enhancement method based on image processing of the present invention.

[0088] Figure 2 This is a framework diagram of the infrared temperature image enhancement system based on image processing of the present invention. DETAILED DESCRIPTION

[0089] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0090] like Figure 1 As shown, the infrared temperature image enhancement method based on image processing provided by the embodiment of the present invention includes:

[0091] S1. Obtain an infrared temperature image, apply multi-scale image processing technology to extract data at different frequencies and scales in the infrared temperature image, normalize the data respectively, and construct a neural network model; wherein the multi-scale image processing technology includes:

[0092] S11, applying wavelet transform to perform multi-scale decomposition on the image to obtain wavelet coefficients at different scales;

[0093] S12, enhancing the wavelet coefficients at each scale to improve the information expression of the image at different scales;

[0094] S13, reconstructing the enhanced wavelet coefficients into an enhanced multi-scale infrared temperature image through inverse wavelet transform;

[0095] S2. Perform image enhancement at multiple scales, input the preprocessed infrared temperature images at multiple scales into the neural network model, design a Q-learning algorithm, learn and update the Q value of the infrared temperature image under different enhancement strategies;

[0096] S3. According to the Q value learned by the Q-learning algorithm, select and execute the best enhancement strategy to train the neural network model, and optimize the loss function of the neural network model; where the loss function L is expressed as

[0097]

[0098] Where y represents the target Q value, which is calculated based on the reward (r) and the maximum Q value of the next state; Q(s,a) is the model's prediction of the Q value for the current state s and action a;

[0099] S4. The multiple multi-scale infrared temperature images output by the neural network model are post-processed by fusing the enhancement results at multiple scales using image fusion technology to optimize the visual effect of the image; wherein the image fusion technology fuses the enhancement results at different scales according to the weights, which is expressed as follows:

[0100]

[0101] Where F represents the fused image, c represents the number of multi-scale infrared temperature images involved in the fusion, and w k represents the weight of the kth multi-scale infrared temperature image enhancement, E i represents the kth multi-scale infrared temperature enhanced image;

[0102] Optionally, the post-processing includes contrast enhancement, color correction, etc., to further improve the visual effect and enhancement effect of the image.

[0103] S5. Output the enhanced infrared temperature image.

[0104] Optionally, display the enhanced image directly on the screen for visual assessment.

[0105] Furthermore, the Q-learning algorithm includes:

[0106] A. Initialize the Q value table and assign an initial Q value to each scale infrared temperature image state-action pair, as shown below:

[0107] Q(s,a)=E[R _t+1 +γR _t+2 +γ^2R _t+3 +…|S_t=s,A_t=a], where s represents the current state, a represents the action taken, R_t represents the immediate reward obtained at time t, γ is the discount factor, S_t represents the current state, and the action A_t is executed;

[0108] B. Select and execute an enhancement strategy based on the state of the current scale infrared temperature image;

[0109] C. Observe the result after executing the enhancement strategy, that is, the state of the next pixel or pixel block and the corresponding reward value;

[0110] D. Fusion evaluation results of the image T otal As a reward, update the Q value table, update the formula V(s) = max[R(s,a)+γ*V(s')], where V(s) represents the value function of state s, R(s,a) represents the reward obtained by taking action a in state s, γ is a discount factor used to weigh the importance of current rewards and future rewards, and V(s') represents the value function of the subsequent state s';

[0111] Among them, the image fusion evaluation indicators include entropy, mean, standard deviation, and visibility;

[0112] The image fusion evaluation is calculated as follows:

[0113] The entropy Q is calculated as follows:

[0114]

[0115] in, The grayscale value of the current multi-size infrared temperature image, Indicates the ratio of the number of pixels of the multi-size infrared temperature image to the total number of pixels;

[0116] The mean MEAN is calculated as follows:

[0117]

[0118] Wherein, f(i, j) represents the region coordinates of the multi-scale infrared temperature image, and m×n represents the size of the multi-scale infrared temperature image;

[0119] The standard deviation STD is calculated as follows:

[0120]

[0121] The Visibility VI is calculated as follows:

[0122]

[0123] Among them, μ represents the visual constant;

[0124] The image fusion evaluation index is expressed as follows:

[0125]

[0126] E. Repeat steps B to D until the Q value table converges or the preset number of iterations is reached.

[0127] Furthermore, the process of reconstructing the enhanced wavelet coefficients into an enhanced multi-scale infrared temperature image through inverse wavelet transform specifically includes:

[0128] a. Calculate the regional energy required for decomposition of the multi-scale infrared temperature image, expressed as follows:

[0129]

[0130] in, represents the regional energy centered at m*n, σ=1,2,3 represent the horizontal, vertical and diagonal directions respectively, i represents the i-th layer, Represents the corresponding wavelet coefficient matrix L i The weight coefficient, L i represents the wavelet coefficient matrix corresponding to the i-th layer, (m', n') represents the corresponding matrix, and K and J represent the range of variation corresponding to m' and n' respectively;

[0131] b. Calculate the matching degree Match between the multi-scale infrared temperature image and the infrared temperature image i (m, n), expressed as follows:

[0132]

[0133] Among them, L i,s represents the wavelet coefficient matrix of the multi-scale infrared temperature image corresponding to the i-th layer, L i,f Represents the wavelet coefficient matrix of the infrared temperature image corresponding to the i-th layer, Represents the regional energy of the multi-scale infrared temperature image centered at m*n, Indicates the energy of the area centered at m*n in the infrared temperature image;

[0134] Optionally, m*n includes 3×3, 5×5 and 7×7.

[0135] The Q value table is updated according to the result of image fusion evaluation, which is expressed as follows:

[0136] The state S represents the characteristics of the infrared temperature image or the multi-size infrared temperature image, and the action A represents the adjustment of the weight coefficient in the wavelet coefficient. The relevant actions are performed according to the characteristics of the infrared temperature image or the multi-size infrared temperature image, and the Q value table is updated, which is expressed as follows:

[0137] Q[s, a]=(1-β)*Q[s, a]+β*(R+γ*max(Q[s′,a']))

[0138] Among them, β represents the learning rate, which controls the influence of the new image fusion result on the old Q value; γ represents the discount factor, which determines the importance of future rewards; s' represents the new state after executing action a, and max(Q[s',a']) represents the Q value of the action with the largest expected return among all possible actions a' taken in the new state.

[0139] Furthermore, the learning rate β also follows the greedy strategy, which is expressed as follows:

[0140]

[0141] Among them, P represents probability, ∈ represents the selection of RandomAction random action with probability ∈, and argmaxQ(a) represents the selection of the action that maximizes the Q value within a certain range with probability 1-∈.

[0142] c. Determine the fusion operator to calculate the wavelet coefficients, and then reconstruct the wavelet coefficients into an enhanced multi-scale infrared temperature image through inverse wavelet transform.

[0143] The determination fusion operator calculates the wavelet coefficients, which are expressed as follows:

[0144] Determine the matching threshold T, if Match i (m, n)<T,

[0145] Optionally, the range of T includes 0.5 to 1, which is expressed as:

[0146]

[0147] If Match i (m, n) ≥ T, which is expressed as follows:

[0148]

[0149] Then there is,

[0150]

[0151] in, represents the wavelet coefficients of the fused image, The wavelet coefficients representing the multi-scale infrared temperature image, Represents the wavelet coefficients of the infrared temperature image, Represents the maximum weight coefficient of the corresponding wavelet coefficient matrix, Represents the minimum weight coefficient of the corresponding wavelet coefficient matrix.

[0152] like Figure 2 As shown, the embodiment of the present invention also provides an infrared temperature image enhancement system based on image processing, characterized in that the system includes:

[0153] The image acquisition unit 501 is used to acquire an infrared temperature image, extract data at different frequencies and scales in the infrared temperature image using a multi-scale image processing technique, normalize the data respectively, and construct a neural network model; wherein the multi-scale image processing technique includes:

[0154] S11, applying wavelet transform to perform multi-scale decomposition on the image to obtain wavelet coefficients at different scales;

[0155] S12, enhancing the wavelet coefficients at each scale to improve the information expression of the image at different scales;

[0156] S13, reconstructing the enhanced wavelet coefficients into an enhanced multi-scale infrared temperature image through inverse wavelet transform;

[0157] The image enhancement unit 502 is used to perform image enhancement at multiple scales, input the preprocessed infrared temperature images at multiple scales into the neural network model, design a Q-learning algorithm, learn and update the Q value of the infrared temperature image under different enhancement strategies;

[0158] The model updating unit 503 is used to select and execute the best enhancement strategy to train the neural network model according to the Q value learned by the Q-learning algorithm, and optimize the loss function of the neural network model; wherein the loss function L is expressed as

[0159]

[0160] Where y represents the target Q value, which is calculated based on the reward (r) and the maximum Q value of the next state; Q(s,a) is the model's prediction of the Q value for the current state s and action a;

[0161] The image fusion unit 504 is used to use the image fusion technology to fuse the enhancement results at multiple scales of the multiple multi-scale infrared temperature images output by the neural network model and then perform post-processing to optimize the visual effect of the image; wherein the image fusion technology fuses the enhancement results at different scales according to the weights, which is expressed as follows:

[0162]

[0163] Where F represents the fused image, c represents the number of multi-scale infrared temperature images involved in the fusion, and w k represents the weight of the kth multi-scale infrared temperature image enhancement, E i represents the kth multi-scale infrared temperature enhanced image;

[0164] The image output unit 505 is used to output the enhanced infrared temperature image.

[0165] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0166] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a division of logical functions. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the device or unit can be electrical, mechanical or other forms.

[0167] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0168] Although embodiments of the present invention have been shown and described, those skilled in the art will appreciate that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. The infrared temperature image enhancement method based on image processing is characterized in that: The method comprises: S1. Obtain an infrared temperature image, apply multi-scale image processing technology to extract data at different frequencies and scales in the infrared temperature image, normalize the data respectively, and construct a neural network model; wherein the multi-scale image processing technology includes: S11, applying wavelet transform to perform multi-scale decomposition on the image to obtain wavelet coefficients at different scales; S12, enhancing the wavelet coefficients at each scale to improve the information expression of the image at different scales; S13, reconstructing the enhanced wavelet coefficients into an enhanced multi-scale infrared temperature image through inverse wavelet transform; S2. Perform image enhancement at multiple scales, input the preprocessed infrared temperature images at multiple scales into the neural network model, design a Q-learning algorithm, learn and update the Q value of the infrared temperature image under different enhancement strategies; S3. According to the Q value learned by the Q-learning algorithm, select and execute the best enhancement strategy to train the neural network model, and optimize the loss function of the neural network model; where the loss function L is expressed as Where y represents the target Q value, which is calculated based on the reward (r) and the maximum Q value of the next state; Q(s,a) is the model's prediction of the Q value for the current state s and action a; S4. The multiple multi-scale infrared temperature images output by the neural network model are post-processed by fusing the enhancement results at multiple scales using image fusion technology to optimize the visual effect of the image; wherein the image fusion technology fuses the enhancement results at different scales according to the weights, which is expressed as follows: Where F represents the fused image, c represents the number of multi-scale infrared temperature images involved in the fusion, and w k represents the weight of the kth multi-scale infrared temperature image enhancement, E i represents the kth multi-scale infrared temperature enhanced image; S5. Output the enhanced infrared temperature image.

2. The infrared temperature image enhancement method based on image processing according to claim 1 is characterized in that: in, The Q-learning algorithm includes: A. Initialize the Q value table and assign an initial Q value to each scale infrared temperature image state-action pair, as shown below: Q(s,a)=E[R_ t+1 +γR_ t+2 +γ^2R_ t+3 +…|S_t=s,A_t=a], where s represents the current state, a represents the action taken, R_t represents the immediate reward obtained at time t, γ is the discount factor, S_t represents the current state, and the action A_t is executed; B. Select and execute an enhancement strategy based on the state of the current scale infrared temperature image; C. Observe the result after executing the enhancement strategy, that is, the state of the next pixel or pixel block and the corresponding reward value; D. Fusion evaluation results of the image T otal As a reward, update the Q value table, update the formula V(s) = max[R(s,a)+γ*V(s')], where V(s) represents the value function of state s, R(s,a) represents the reward obtained by taking action a in state s, γ is a discount factor used to weigh the importance of current rewards and future rewards, and V(s') represents the value function of the subsequent state s'; E. Repeat steps B to D until the Q value table converges or the preset number of iterations is reached.

3. The infrared temperature image enhancement method based on image processing according to claim 1 is characterized in that: The method of reconstructing the enhanced wavelet coefficients into an enhanced multi-scale infrared temperature image by inverse wavelet transform specifically includes: a. Calculate the regional energy required for decomposition of the multi-scale infrared temperature image, expressed as follows: in, represents the regional energy centered at m*n, σ=1,2,3 represent the horizontal, vertical and diagonal directions respectively, i represents the i-th layer, Represents the corresponding wavelet coefficient matrix L i The weight coefficient, L i represents the wavelet coefficient matrix corresponding to the i-th layer, (m′, n′) represents the corresponding matrix, and K and J represent the range of variation corresponding to m′ and n′ respectively; b. Calculate the matching degree Match between the multi-scale infrared temperature image and the infrared temperature image i (m, n), expressed as follows: Among them, L i,s represents the wavelet coefficient matrix of the multi-scale infrared temperature image corresponding to the i-th layer, L i,f Represents the wavelet coefficient matrix of the infrared temperature image corresponding to the i-th layer, Represents the regional energy of the multi-scale infrared temperature image centered at m*n, Indicates the energy of the area centered at m*n in the infrared temperature image; c. Determine the fusion operator to calculate the wavelet coefficients, and then reconstruct the wavelet coefficients into an enhanced multi-scale infrared temperature image through inverse wavelet transform.

4. The infrared temperature image enhancement method based on image processing according to claim 3 is characterized in that: The determination fusion operator calculates the wavelet coefficients, which are expressed as follows: Determine the matching threshold T, if Match i (m, n)<T, then it is expressed as If Match i (m, n) ≥ T, which is expressed as follows: Then there is, in, represents the wavelet coefficients of the fused image, The wavelet coefficients representing the multi-scale infrared temperature image, Represents the wavelet coefficients of the infrared temperature image, Represents the maximum weight coefficient of the corresponding wavelet coefficient matrix, Represents the minimum weight coefficient of the corresponding wavelet coefficient matrix.

5. The infrared temperature image enhancement method based on image processing according to claim 4 is characterized in that: The range of T includes 0.5-1.

6. The infrared temperature image enhancement method based on image processing according to claim 3 is characterized in that: m*n includes 3×3, 5×5 and 7×7.

7. The infrared temperature image enhancement method based on image processing according to claim 2 is characterized in that: The Q value table is updated according to the result of image fusion evaluation, which is expressed as follows: The state S represents the characteristics of the infrared temperature image or the multi-size infrared temperature image, and the action A represents the adjustment of the weight coefficient in the wavelet coefficient. The relevant actions are performed according to the characteristics of the infrared temperature image or the multi-size infrared temperature image, and the Q value table is updated, which is expressed as follows: Q[s,a]=(1-β)*Q[s,a]+β*(R+γ*max(Q[s',a'])) Among them, β represents the learning rate, which controls the influence of the new image fusion result on the old Q value; γ represents the discount factor, which determines the importance of future rewards; s′ represents the new state after executing action a, and max(Q[s',a']) represents the Q value of the action with the largest expected return among all possible actions a' taken in the new state.

8. The infrared temperature image enhancement method based on image processing according to claim 7 is characterized in that: The learning rate β also follows the greedy strategy, which is expressed as follows: Among them, P represents probability, RandomAction selects random actions with probability ∈, and argmaxQ(a) represents selecting the action that maximizes the Q value within a certain range with probability 1-∈.

9. The infrared temperature image enhancement method based on image processing according to claim 8 is characterized in that: Image fusion evaluation indicators include entropy, mean, standard deviation, and visibility; The image fusion evaluation is calculated as follows: The entropy Ent is calculated as follows: in, The grayscale value of the current multi-size infrared temperature image, Indicates the ratio of the number of pixels of the multi-size infrared temperature image to the total number of pixels; The mean MEAN is calculated as follows: Wherein, f(i, j) represents the region coordinates of the multi-scale infrared temperature image, and m×n represents the size of the multi-scale infrared temperature image; The standard deviation STD is calculated as follows: The Visibility VI is calculated as follows: Among them, μ represents the visual constant; The image fusion evaluation index is expressed as follows:

10. The infrared temperature image enhancement system based on image processing is characterized in that: The system comprises: The image acquisition unit is used to acquire an infrared temperature image, apply a multi-scale image processing technique to extract data at different frequencies and scales in the infrared temperature image, normalize the data respectively, and construct a neural network model; wherein the multi-scale image processing technique includes: S11, applying wavelet transform to perform multi-scale decomposition on the image to obtain wavelet coefficients at different scales; S12, enhancing the wavelet coefficients at each scale to improve the information expression of the image at different scales; S13, reconstructing the enhanced wavelet coefficients into an enhanced multi-scale infrared temperature image through inverse wavelet transform; The image enhancement unit is used to perform image enhancement at multiple scales. The preprocessed infrared temperature images at multiple scales are input into the neural network model, and a Q-learning algorithm is designed to learn and update the Q value of the infrared temperature image under different enhancement strategies. The model updating unit is used to select and execute the best enhancement strategy to train the neural network model according to the Q value learned by the Q-learning algorithm, and optimize the loss function of the neural network model; the loss function L is expressed as Where y represents the target Q value, which is calculated based on the reward (r) and the maximum Q value of the next state; Q(s,a) is the model's prediction of the Q value for the current state s and action a; The image fusion unit is used to use the image fusion technology to fuse the enhancement results at multiple scales of the multiple multi-scale infrared temperature images output by the neural network model and then perform post-processing to optimize the visual effect of the image; wherein the image fusion technology fuses the enhancement results at different scales according to the weights, which is expressed as follows: Where F represents the fused image, c represents the number of multi-scale infrared temperature images involved in the fusion, and w k represents the weight of the kth multi-scale infrared temperature image enhancement, E i represents the kth multi-scale infrared temperature enhanced image; The image output unit is used to output the enhanced infrared temperature image.

Images