Complex system image recognition method based on projection gradient descent PGD artificial intelligence algorithm

Abstract

The invention discloses a complex system image recognition method based on a projection gradient descent PGD artificial intelligence algorithm, and the method comprises the following steps: constructing a space-time reaction diffusion model structure based on a complex system image, and the structure comprises the design of a propagation mechanism and a feedback mechanism; setting a loss function, wherein the loss function is used for quantifying the difference between the model output and the target output; initial data related to the target picture is determined and collected, including design data and model parameters. The PGD algorithm has extremely high recognition speed while ensuring high precision. Through efficient iterative optimization and parallel calculation, the PGD algorithm can complete a complex image identification task in dozens of seconds to several minutes. The rapid processing capability obviously improves the medical working efficiency.

Description

Technical Field

[0001] The present invention relates to the fields of computer vision and image processing, and particularly to the development and application of image recognition technology for complex systems based on the projection gradient descent (PGD) artificial intelligence algorithm. This technology aims to improve the accuracy and efficiency of image recognition and is widely applied in fields such as medical image analysis, autonomous driving, and security monitoring. Background Art

[0002] With the rapid development of deep learning and computer vision technologies, image recognition technology has been widely applied in various industries. However, existing image recognition methods often rely on a large amount of labeled data and complex model training, resulting in long model training time and high consumption of computing resources. In addition, traditional complex system parameter recognition methods are prone to falling into local optimal solutions when dealing with high-dimensional data, which affects the recognition effect.

[0003] Therefore, how to effectively perform system parameter recognition and improve the accuracy and efficiency of image recognition has become an urgent problem to be solved in the current technical field. Summary of the Invention

[0004] The present invention provides an application technology for biomedical image recognition based on the PGD algorithm, aiming to improve the accuracy and efficiency of image recognition by optimizing the algorithm.

[0005] The content of this application is partially used to introduce concepts in the form of an example, and these concepts will be described in detail in the subsequent specific implementation part. The content of this application is not intended to identify the key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.

[0006] Specifically, the main content of the present invention includes:

[0007] 1. Data acquisition module:

[0008] Extract biomedical image data from the image database, denoted as and perform grayscale processing to improve the recognition accuracy.

[0009] 2. Image processing module:

[0010] Establish a reaction-diffusion system corresponding to biomedical image recognition. First, divide the population into 5 categories:

[0011] Susceptibles (S): People who have not been infected but may be infected, lacking immunity and being vulnerable to pathogen infection, and are potential targets for disease transmission.

[0012] Infected (I): People who have been infected and are infectious, capable of transmitting pathogens to susceptibles, and are the direct source of disease transmission.

[0013] Recovered (R): People who have recovered from the infection and acquired immunity, no longer contagious and not susceptible to reinfection, are the end point of disease transmission.

[0014] Unknown (U): People who have not been infected and are far from the disease transmission chain, neither infected nor exposed to the pathogen, outside the scope of disease transmission.

[0015] Immune (P): People who have acquired complete immunity after recovery from illness, having developed antibodies through the process of infection and recovery, not susceptible to reinfection and no longer capable of transmitting the disease, differing from the recovered in having a more persistent and stable immune status.

[0016] Based on the above settings, after clarifying the characteristics of each population, the following disease transmission model is established at time y and space x:

[0017]

[0018] where the initial conditions are

[0019]

[0020] Considering no population movement and no spread of green behavior at the system boundary, Neumann boundary conditions are selected:

[0021]

[0022] In addition, Ω represents the scope of spatial activity; represents the boundary of Ω; n represents the unit normal vector along the region boundary; Δ represents the Laplace operator; represents the partial derivative of S with respect to t, Similarly; represents the derivative of S, Similarly.

[0023] 3. Parameter Identification Module:

[0024] Set the loss function as the gap between the actual solution and the target at time T, the gap between the actual solution and the target within the time range (0, T), and the loss caused by the change in the values of the parameters to be identified during the time period (0, T), with the specific form:

[0025]

[0026] where β1, β2, β3, α1, α2, α3, and θ are positive weight coefficients, (S, I, R) and (S T , I T , R TThey are the solution of system (1) and the target pattern respectively, and α is the parameter group to be identified. The first three items represent the gap between the actual solution and the target of the green behavior at time T. The next three items represent the gap between the solution and the target within the time range (0, T). The last item represents the loss caused by the change of the parameter value to be identified during the time period (0, T).

[0027] The process of parameter identification is as follows:

[0028]

[0029] The algorithm uses the PGD (Projected Gradient Descent) algorithm to identify parameters of image features. Initialize the model parameters and set the initial value as θ (0) ; In each iteration, calculate the gradient of the loss function and update the model parameters. The specific update formula is:

[0030]

[0031] where δ is the learning rate, is the gradient of the loss function with respect to the parameter. The selection of the learning rate δ is based on the adaptive learning rate adjustment algorithm. The specific algorithm process is as follows.

[0032]

[0033] 4. Result evaluation module: Evaluate the recognition effect by introducing a validation set, and comprehensively evaluate the recognition results using multiple evaluation metrics (such as relative error, etc.).

[0034] 5. Beneficial effects:

[0035] (1) High recognition accuracy

[0036] The PGD (Projected Gradient Descent) algorithm shows extremely high recognition accuracy in biomedical image recognition. By optimizing the objective function and the gradient descent strategy, PGD can accurately capture the subtle features in the image, such as lesion areas or cell structures. The test results show that its recognition results are highly consistent with the theoretical expectations and the error is extremely small. This high accuracy is particularly important in medical diagnosis, which can effectively reduce misdiagnosis and missed diagnosis and provide a reliable basis for clinical decision-making.

[0037] (2) Fast recognition speed

[0038] While ensuring high precision, the PGD algorithm has an extremely fast recognition speed. Through efficient iterative optimization and parallel computing, PGD can complete the recognition task of complex images within dozens of seconds to a few minutes. This fast processing ability significantly improves the efficiency of medical work. Especially in real-time diagnosis and large-scale screening scenarios, it can quickly provide results, shorten the waiting time of patients, and contribute to the optimal allocation of medical resources. Brief Description of the Drawings

[0039] Figure 1 It is a structural diagram of an image recognition application system for red blood cell diffusion provided by the present invention;

[0040] Figure 2 It is a structural diagram of an image recognition application system for carbon emission distribution picture provided by the present invention. Detailed Embodiments

[0041] To make the objectives, technical solutions and advantages of the present invention clearer, the following will further describe the embodiments of the present invention in detail with reference to the accompanying Figure 1-2 drawings.

[0042] 1. Data acquisition module:

[0043] Extract biomedical image data from the image database, denoted as and perform grayscale processing to improve the recognition accuracy.

[0044] 2. Picture processing module:

[0045] Establish a reaction-diffusion system corresponding to biomedical image recognition. First, divide the population into 5 categories:

[0046] Susceptibles (S): People who have not been infected but may be infected, lacking immunity and being vulnerable to pathogen infection, are potential targets for disease transmission.

[0047] Infected (I): People who have been infected and are infectious, capable of transmitting pathogens to susceptibles, are the direct sources of disease transmission.

[0048] Recovered (R): People who have recovered from the infection and acquired immunity, no longer infectious and will not be infected again, are the end points of disease transmission.

[0049] Unknowns (U): People who have not been infected and are far from the disease transmission chain, neither infected nor exposed to pathogens, are outside the scope of disease transmission.

[0050] Immunes (P): People who have obtained complete immunity after recovery from the disease, who have produced antibodies through the infection and recovery process, will not be infected again and no longer have the ability to transmit the disease. The difference from recovered people is that their immune status is more persistent and stable.

[0051] Based on the above settings, after clarifying the characteristics of each population, the possible situations of the spread of green disease behavior among the population are further analyzed.

[0052] When U delves deeply into green behavior, they will naturally grow into S, I, and R, with their numbers being A1, A2, and 1 - A1 - A2 respectively.

[0053] When the green behavior spreads between the S and I populations, with a contact rate of β, βSI in S 2 will flow to I. At the same time, if the effect of the green behavior does not reach the ideal state, there is a probability of μ1 for I to transform into S.

[0054] When the green behavior spreads between the I and R populations, due to reasons such as the unsatisfactory effect of the green behavior or unreasonable policies, I will transform into S, and the process can be expressed by μ2I. At the same time, under the influence of factors such as population communication, national policies, and social public opinion, R will transform into P and no longer participate in the spread process, with a probability of α.

[0055] Under the influence of external factors such as national policies and government publicity, S will become P with a probability of γ.

[0056] There is a self - diffusion phenomenon among individuals with the same attitude towards green behavior. d1, d2, and d3 are the self - diffusion coefficients of populations S, I, and R respectively. These coefficients mean that the more firm the attitude, the stronger the influence on those with less firm beliefs. d1ΔS represents the change quantity of S in the x - space at time t, and d2ΔI and d3ΔR are similar.

[0057] Combined with Figure 1 and the previous assumptions, the following disease - spreading model is obtained:

[0058]

[0059] where the initial conditions are

[0060]

[0061] Considering that there is no population movement and the spread of green disease behavior at the system boundary, the Neumann boundary condition is selected:

[0062]

[0063] In addition, Ω represents the scope of spatial activities; represents the boundary of Ω; n represents the unit normal vector along the regional boundary; Δ represents the Laplace operator; represents the partial derivative of S with respect to t; represents the derivative of S.

[0064] 3. Parameter Identification Module:

[0065] Set the loss function as the gap between the actual solution and the target at time T, the gap between the actual solution and the target within the time range (0, T), and the loss caused by the change in the parameter values to be identified during the time period (0, T). The specific form is:

[0066]

[0067] where β1, β2, β3, α1, α2, α3, and θ are positive weight coefficients, and (S, I, R) and (S T , I T , R T ) are the solution of system (6) and the target pattern respectively. The first three terms represent the gap between the actual solution and the target of the green behavior at time T, and α is the parameter group to be identified. The next three terms represent the gap between the solution and the target within the time range (0, T). The last term represents the loss caused by the change in the parameter values to be identified during the time period (0, T).

[0068] The process of parameter identification is as follows:

[0069]

[0070] The algorithm uses the PGD (Projected Gradient Descent) algorithm to identify the parameters of the image features. Initialize the model parameters, and set the initial value as θ (0) ; In each iteration, calculate the gradient of the loss function and update the model parameters. The specific update formula is:

[0071]

[0072] where δ is the learning rate, is the gradient of the loss function with respect to the parameters. The selection of the learning rate δ is based on the adaptive learning rate adjustment algorithm. The specific algorithm process is as follows.

[0073]

[0074] 4. Result Evaluation Module:

[0075] Evaluate the recognition effect by introducing a validation set or a test set, and comprehensively evaluate the recognition results using multiple evaluation metrics (such as accuracy, relative error, etc.).

[0076] Figure 1Shows the distribution of normal red blood cells and elliptocytes under the microscope. Through grayscale processing technology, the contrast between red blood cells and the background in the image has been significantly enhanced, making the morphological characteristics of red blood cells more clearly visible. The recognition result on the far right shows that the algorithm can effectively distinguish normal red blood cells and elliptocytes, with good recognition effect. This technology is of great significance for medical diagnosis, especially in the detection of blood diseases such as elliptocytosis. Through automated image recognition, doctors can diagnose diseases more quickly and accurately, reducing human errors. In addition, the application of this technology can also be extended to the recognition of other cell morphologies, providing strong tool support for medical research and clinical diagnosis.

[0077] Figure 2 Shows the carbon emission distribution in a certain area. Through image recognition of some areas, the recognition result on the far right clearly reflects the spatial distribution of carbon emissions. This technology can help environmental scientists and policymakers identify hotspots of carbon emissions, thus formulating more targeted emission reduction strategies.

[0078] 5. Beneficial effects:

[0079] (1) High recognition accuracy

[0080] The PGD (Projected Gradient Descent) algorithm shows extremely high recognition accuracy in biomedical image recognition. By optimizing the objective function and gradient descent strategy, PGD can accurately capture subtle features in the image, such as lesion areas or cell structures. The test results show that its recognition results are highly consistent with the theoretical expectations, with extremely small errors. This high accuracy is particularly important in medical diagnosis, which can effectively reduce misdiagnosis and missed diagnosis, providing a reliable basis for clinical decision-making.

[0081] (2) Fast recognition speed

[0082] While ensuring high accuracy, the PGD algorithm has an extremely fast recognition speed. Through efficient iterative optimization and parallel computing, PGD can complete the recognition task of complex images within dozens of seconds to a few minutes. This fast processing ability significantly improves the efficiency of medical work. Especially in real-time diagnosis and large-scale screening scenarios, it can quickly provide results, shorten the waiting time of patients, and contribute to the optimal allocation of medical resources.

Claims

1. A method for image recognition of complex systems based on the projection gradient descent (PGD) artificial intelligence algorithm, characterized in that, It includes the following steps: Construct a spatio-temporal reaction-diffusion model structure based on the complex system image, which includes the design of the propagation mechanism and the feedback mechanism; Set the loss function, which is used to quantify the difference between the model output and the target output; Determine and collect the initial data related to the target picture, including design data and model parameters.

2. The method according to claim 1, wherein The construction of the reaction-diffusion model structure includes: The interactions between different species, boundary conditions and initial conditions; positive feedback and negative feedback. The specific content of model establishment is as follows: First, divide the population into 5 categories: Susceptibles S: The population that has not been infected but may be infected, lacks immunity, is vulnerable to pathogen infection, and is a potential target for disease transmission; Infected I: The population that has been infected and is infectious, can transmit the pathogen to susceptibles, and is the direct source of disease transmission; Recovered R: The population that has recovered from the infection and acquired immunity, is no longer infectious, and will not be infected again, and is the end point of disease transmission; Unknown U: The population that has not been infected and is far from the disease transmission chain. They have neither been infected nor exposed to the pathogen and are outside the scope of disease transmission; Immune P: The population that has obtained complete immunity after recovery from the disease. They have produced antibodies through the infection and recovery process, will not be infected again, and no longer have the ability to transmit the disease. The difference from the recovered is that their immune state is more persistent and stable; Based on the above settings, after clarifying the characteristics of each population, further analyze the possible situations of the spread of green behavior of the disease in the population; (1) When U deeply studies green behavior, they will naturally grow into S, I, and R, and their numbers are A1, A2, and 1 - A1 - A2 respectively; (2) When the green behavior spreads between the S and I populations, with a contact rate of β, the contact transmission of βSI among the S population will flow to I. 2 At the same time, if the effect of the green behavior does not reach the ideal state, there is a probability of μ1 for I to transform into S. (3) When green behavior spreads between the two populations of I and R, due to reasons such as the unsatisfactory effect of green behavior or unreasonable policies, I will transform into S, and the process can be expressed by μ2I; at the same time, under the influence of factors such as population communication, national policies, and social opinions, R will transform into P and no longer participate in the transmission process, and the probability is α; (4) Under the influence of external factors such as national policies and government publicity, S will become P with a probability of γ; (5) There is a self-diffusion phenomenon among individuals with the same attitude towards green behavior; d1, d2, and d3 are the self-diffusion coefficients of the populations S, I, and R respectively; these coefficients mean that the more firm the attitude, the stronger the influence on people with less firm beliefs; d1ΔS represents the change quantity of S in the x space at time t, and the meanings of d2ΔI and d3ΔR are named similarly; Combined with the previous assumptions, the following disease transmission model is obtained in the time t and space x: where the initial condition is Considering that there is no population movement and the spread of green behavior of the disease at the system boundary, the Neumann boundary condition is selected: In addition, Ω represents the scope of space activities; represents the boundary of Ω; n represents the unit normal vector along the regional boundary; Δ represents the Laplace operator; represents the partial derivative of S with respect to t, similarly; represents the derivative of S, similarly.

3. The method according to claim 1, characterized in that, The loss function l includes: The gap between the actual solution and the target at time T; the gap between the actual solution and the target within the time range (0, T); the loss caused by the change of the parameter value to be identified in the time period (0, T), and its specific form in the space Ω is: where β1, β2, β3, α1, α2, α3, and θ are positive weight coefficients, (S(x,T), I(x,T), R(x,T)) and (S T (x), I T (x), R T (x)) are the solution of system (1) and the target pattern respectively, α is the parameter group to be identified; the first three terms represent the gap between the actual solution and the target of the green behavior at time T; the next three terms represent the gap between the solution and the target within the time range (0, T); the last term represents the loss caused by the change of the parameter value to be identified during the time period (0, T).

4. The method according to claim 1, wherein Determine and collect the initial data related to the target image, including: grayscale processing the target image to improve recognition accuracy; extracting design data from the image database, where the design data includes the geometric features, color distribution, and texture information of the image; all parameters should be within a preset range to ensure the interpretability and stability of the model, and can be obtained through literature research, experimental measurement, or expert evaluation methods.

5. The method according to claim 1, wherein The implementation method for identifying the model parameters includes the following steps: Initialize the model parameters and set the initial value to θ (0) ; In each iteration, calculate the gradient of the loss function and update the model parameters. The specific update formula is as follows: where δ is the learning rate, is the gradient of the loss function with respect to the parameter, θ (k) is the value of the parameter after the k-th iteration.

6. The method according to claim 5, characterized in that The selection basis for the learning rate δ is the adaptive learning rate adjustment algorithm.

Images