Adaptive light force matching method and system based on deep learning

Abstract

The invention belongs to the technical field of optical micromanipulation, and particularly relates to a self-adaptive light force matching method and system based on deep learning, and the method comprises the steps: inputting a training database into a neural network model for training, and obtaining a neural network agent model; initial light field parameters are set, and initial light trap rigidity is obtained; setting a light field parameter range; inputting the real-time microsphere size and the initial light trap rigidity into a neural network agent model, increasing the initial light trap rigidity by fixed step length, finding out the maximum light trap rigidity when the predicted light field parameter is greater than the light field parameter range, combining the initial light trap rigidity to form a light trap rigidity critical interval, and taking the light field parameter range as a constraint to obtain the maximum light trap rigidity. And iterating the optical trap stiffness by using a dichotomy and a neural network agent model to obtain optimal optical trap stiffness and optimal optical field parameters. The method has the advantages that the optimal light field parameters can be quickly obtained for captured objects of different sizes, and the stable capturing effect is achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of optical micro-manipulation, and in particular relates to an adaptive optical force matching method and system based on deep learning. Background Art

[0002] Optical tweezers is a technology that uses the pressure of light and the gradient of the light field to capture and manipulate tiny particles (such as cells, biomolecules, and nanoparticles). Optical tweezers have the advantages of contactless manipulation and no mechanical damage, and have become an important research tool in the fields of quantum physics, life sciences, and precision measurement. In the field of life sciences, optical tweezers are widely used in the study of the mechanical properties of macromolecules or single cells, the interaction between DNA and protein molecules, the interaction between colloidal particles, and the control of the crystallization process of crystals.

[0003] In the related art, in the application process of optical tweezers technology, it is often necessary to capture objects of different sizes. When capturing objects of different sizes, the method usually adopted is to increase the optical power to match large-sized cells.

[0004] Regarding the above-mentioned related technologies, increasing the optical power to match large-sized cells will affect the cell activity to a certain extent, and it is impossible to accurately find the optimal optical power. When using traditional optical tweezers toolkits to optimize the light field parameters, the optimization process requires complex simulation calculations based on the cell size, and the calculation cost is too high. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide an adaptive optical force matching method and system based on deep learning, which can quickly obtain the optimal light field parameters for capture objects of different sizes and achieve stable capture.

[0006] An adaptive optical force matching method based on deep learning, comprising:

[0007] According to the test microsphere size, test light field parameters and optical force calculation model, the corresponding test light trap stiffness is obtained;

[0008] Establishing a training database including a mapping relationship according to the test microsphere size, the test light trap stiffness, and the test light field parameters;

[0009] Inputting the training database into the neural network model, training the neural network model with the test microsphere size and the test light trap stiffness as input and the test light field parameters as output to obtain a neural network proxy model;

[0010] Setting initial light field parameters, and obtaining a real-time microsphere size, and calculating a current light trap stiffness according to the real-time microsphere size and the initial light field parameters as an initial light trap stiffness;

[0011] Set the light field parameter range;

[0012] Inputting the real-time microsphere size and the initial light trap stiffness into a neural network proxy model, increasing the initial light trap stiffness by a fixed step length, comparing the predicted light field parameters output by the neural network proxy model with the light field parameter range, finding the maximum light trap stiffness corresponding to when the predicted light field parameters are greater than the light field parameter range, and combining the initial light trap stiffness and the maximum light trap stiffness to form a critical interval of light trap stiffness;

[0013] In the critical interval of the light trap stiffness, the light field parameter range is constrained, the light trap stiffness is iterated using the dichotomy method and the neural network proxy model to obtain the optimal light trap stiffness corresponding to the real-time microsphere size, and then the optimal light field parameter is obtained through the neural network proxy model;

[0014] The capture object having a size equal to the size of the real-time microsphere is captured by matching the optical force with the optimal light field parameters.

[0015] Optionally, obtaining the corresponding test light trap stiffness according to the microsphere size, light field parameters and light force calculation model includes:

[0016] According to the test light field parameters, the electric field intensity is obtained;

[0017] According to the electric field strength and the gradient force formula, the gradient force of the light force on the microsphere is obtained;

[0018] Obtain the refractive index of the environment medium and the refractive index of the microsphere;

[0019] According to the refractive index of the environment medium, the refractive index of the microsphere, the size of the microsphere and the scattering force formula, the scattering force of the light force exerted on the microsphere is obtained;

[0020] Obtaining the light force exerted on the microsphere according to the gradient force and the scattering force;

[0021] According to the axial optical force of the optical force exerted on the microsphere and the stable capture formula, the stiffness of the test optical trap when the axial optical force exerted on the microsphere is the maximum is calculated.

[0022] Optionally, the gradient force formula is:

[0023]

[0024] Where c is the speed of light in vacuum, ε 0 is the dielectric constant in vacuum, is the square of the electric field gradient, F grad is the gradient force;

[0025]

[0026] Among them, n med is the refractive index of the surrounding medium, m is the relative refractive index, m = n part / n med , n part is the refractive index of the microsphere, W is the size of the microsphere;

[0027] The scattering force formula is:

[0028]

[0029] Among them, F scat is the scattering force, <s>is the time-averaged Poynting vector, σ is the scattering cross section;

[0030]

[0031] where a is related to the wavelength and n med The relevant constants;

[0032] The stable capture formula is:

[0033] F z =-k z (zz 0 );

[0034] Among them, F z is the axial optical force on the microsphere, z is when F z The position of the microsphere when it is at its maximum value, z 0 is the equilibrium point position, k z is the light trap stiffness.

[0035] Optionally, the real-time microsphere size and the initial light trap stiffness are input into a neural network proxy model, the initial light trap stiffness is increased by a fixed step length, the predicted light field parameters output by the neural network proxy model are compared with the light field parameter range, and the maximum light trap stiffness corresponding to the predicted light field parameter is found when the light field parameter is greater than the light field parameter range. The initial light trap stiffness and the maximum light trap stiffness are combined to form a critical interval of light trap stiffness, including:

[0036] Get the set step size;

[0037] Obtaining an iterative light trap stiffness according to the initial light trap stiffness and a set step size;

[0038] Inputting the iterative light trap stiffness and the real-time microsphere size into a neural network proxy model to obtain predicted light field parameters;

[0039] If the predicted light field parameter is within the light field parameter range, the iterative light trap stiffness is continuously increased by setting the step size until the predicted light field parameter calculated by iterative light trap stiffness and the real-time microsphere size exceeds the light field parameter range, and the first iterative light trap stiffness whose predicted light field parameter calculated by iterative light trap stiffness and the real-time microsphere size exceeds the light field parameter range is taken as the maximum light trap stiffness;

[0040] The initial light trap stiffness and the maximum light trap stiffness form a light trap stiffness critical interval.

[0041] Optionally, within the critical interval of the light trap stiffness, with the light field parameter range as a constraint, the light trap stiffness is iterated using a dichotomy method and a neural network proxy model to obtain an optimal light trap stiffness corresponding to the real-time microsphere size, and then the optimal light field parameters are obtained through the neural network proxy model, including:

[0042] In the critical interval of the light trap stiffness, a dichotomy method is used to obtain an intermediate value of the light trap stiffness;

[0043] The intermediate value of the light trap stiffness and the real-time microsphere size are substituted into the neural network proxy model, and the light trap stiffness is iterated with the light field parameter range as a constraint to obtain the optimal light trap stiffness corresponding to the real-time microsphere size, and then the optimal light field parameters are obtained through the neural network proxy model.

[0044] Optionally, the intermediate value of the light trap stiffness and the real-time microsphere size are substituted into a neural network proxy model, and the light trap stiffness is iterated with the light field parameter range as a constraint to obtain an optimal light trap stiffness corresponding to the real-time microsphere size, and then the optimal light field parameter is obtained through the neural network proxy model, including:

[0045] Substituting the intermediate value of the light trap stiffness and the real-time microsphere size into a neural network proxy model, iterating the light trap stiffness, and obtaining iterative light field parameters;

[0046] If the iterative light field parameter is within the light field parameter range, taking the maximum value of the light trap stiffness intermediate value and the light trap stiffness critical interval as the iterative critical interval;

[0047] If the iterative light field parameter is not within the light field parameter range, taking the minimum value of the light trap stiffness intermediate value and the light trap stiffness critical interval as the iterative critical interval;

[0048] Set the fitness function and stop threshold;

[0049] The iterative critical interval is continuously iterated using the dichotomy method until the fitness function value is less than the stop threshold and the iterative light field parameters are within the light field parameter range, and the light trap stiffness corresponding to the time when the fitness function value is less than the stop threshold and the iterative light field parameters are within the light field parameter range is taken as the optimal light trap stiffness, and then the optimal light field parameters are obtained through the neural network model.

[0050] Optionally, the fitness function is:

[0051]

[0052] Where Δf is the stop threshold, abs represents the absolute value, and They represent the light trap stiffness at the current iteration and the light trap stiffness at the previous iteration respectively.

[0053] A cell capture parameter determination system based on deep learning, comprising:

[0054] A calculation module, used to obtain the corresponding test light trap stiffness according to the test microsphere size, the test light field parameters and the optical force calculation model;

[0055] A training database establishment module, used to establish a training database containing a mapping relationship according to the test microsphere size, the test light trap stiffness and the test light field parameters;

[0056] A training module, used for inputting the training database into the neural network model, training the neural network model with the test microsphere size and the test light trap stiffness as input and the test light field parameters as output, to obtain a neural network proxy model;

[0057] An initialization module is used to set initial light field parameters and obtain real-time microsphere size, and calculate the current light trap stiffness as the initial light trap stiffness according to the real-time microsphere size and the initial light field parameters;

[0058] A setting module, used to set the light field parameter range;

[0059] A comparison module is used to input the real-time microsphere size and the initial light trap stiffness into a neural network proxy model, increase the initial light trap stiffness by a fixed step length, compare the predicted light field parameters output by the neural network proxy model with the light field parameter range, find the maximum light trap stiffness corresponding to when the predicted light field parameters are greater than the light field parameter range, and form a critical interval of light trap stiffness with the initial light trap stiffness and the maximum light trap stiffness;

[0060] An iterative module is used to iterate the light trap stiffness within the critical interval of the light trap stiffness and with the light field parameter range as a constraint, using a dichotomy method and a neural network proxy model to obtain an optimal light trap stiffness corresponding to the real-time microsphere size, and then obtain an optimal light field parameter through the neural network proxy model;

[0061] The capture module is used to capture the capture object having the size of the real-time microsphere by matching the optical force with the optimal light field parameters.

[0062] A terminal device includes a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor loads and executes the computer program, an adaptive optical force matching method based on deep learning is adopted.

[0063] A computer-readable storage medium stores a computer program. When the computer program is loaded and executed by a processor, an adaptive optical force matching method based on deep learning is adopted.

[0064] The beneficial effects of the present invention are:

[0065] By taking the light field parameter range as a constraint, iterating the light trap stiffness, substituting it into the trained neural network proxy model, and reversely designing the light field parameters, first, according to the initial light field parameters and the real-time microsphere size, iterating continuously in a fixed step length manner to obtain the critical interval of the light trap stiffness. Then, within the critical interval of the light trap stiffness, using the dichotomy method and the neural network proxy model, with the light field parameters as constraints, the light trap stiffness is iterated again continuously to find the light trap stiffness when the fitness function value is less than the stop threshold and the corresponding iterative light field parameters are within the light field parameter range as the optimal light trap stiffness, and then the optimal light field parameters are obtained by the optimal light trap stiffness and the real-time microsphere size. Compared with the traditional optical tweezers toolkit calculation method, the present application can quickly obtain the optimal light field parameters for captured objects of different sizes, thereby realizing adaptive optical force matching for microspheres of different sizes and achieving stable capture. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 A flowchart of an adaptive optical force matching method based on deep learning;

[0067] Figure 2 is a schematic diagram of the microspheres of the present invention being stably captured;

[0068] Figure 3 For K z =2.91×10 -2 pN·μm -1 mW -1 The corresponding relationship between the size of silica microspheres and the light power when

[0069] Figure 4 Schematic diagram of the axial optical trap stiffness of microspheres of different sizes at P = 100mW calculated using the traditional optical tweezers toolkit and predicted by the neural network model;

[0070] Figure 5 The prediction results of the neural network model of this application on the axial optical trap stiffness of microspheres with different optical powers and different sizes;

[0071] Figure 6 Schematic diagram of the calculation results using a traditional optical tweezers toolkit. DETAILED DESCRIPTION

[0072] An adaptive optical force matching method based on deep learning, such as Figure 1 As shown, including:

[0073] S1. Obtain the corresponding test light trap stiffness according to the test microsphere size, test light field parameters and light force calculation model.

[0074] Specifically, the conventional optical tweezers toolkit simulation calculation method is used to find the test light trap stiffness corresponding to the maximum axial optical force under the test microsphere size and test light field parameters, and use it as the test light trap stiffness.

[0075] The microsphere size is the diameter of the microsphere.

[0076] Through simulation calculations of a traditional optical tweezers toolkit, several databases containing test microsphere sizes, test light field parameters, and corresponding test light trap stiffness values ​​are calculated.

[0077] S11. Obtaining the electric field gradient according to the test light field parameters.

[0078] Specifically, the light field parameters and the electric field intensity are positively correlated. When the light field parameters increase, the electric field intensity increases. The electric field gradient is how the electric field intensity changes in space. Therefore, when the light field parameters change, the electric field gradient will be affected.

[0079] S12. According to the electric field gradient and the gradient force formula, the gradient force of the light force acting on the microsphere is obtained.

[0080] The gradient force formula is:

[0081]

[0082] Where c is the speed of light in vacuum, ε 0 is the dielectric constant in vacuum, is the square of the electric field gradient, F grad is the gradient force.

[0083]

[0084] Among them, n med is the refractive index of the surrounding medium, m is the relative refractive index, m = n part / n med , n part is the refractive index of the microsphere, and W is the size of the microsphere.

[0085] The term related to the refractive index is expressed as:

[0086]

[0087] Among them, the magnitude of the light scattering force is related to Δn 2 Proportional to W 6 Proportional, Δn=n part -n med is the refractive index difference between the microsphere and the surrounding medium.

[0088] S13. Obtain the refractive index of the ambient medium and the refractive index of the microsphere.

[0089] S14. According to the refractive index of the environment medium, the refractive index of the microsphere, the size of the microsphere and the scattering force formula, the scattering force of the light force exerted on the microsphere is obtained.

[0090] The scattering force formula is:

[0091]

[0092] Among them, F scat is the scattering force, <s>is the time-averaged Poynting vector and σ is the scattering cross section.

[0093]

[0094] Where a is related to the wavelength and n med The relevant constants.

[0095] S15. Obtain the light force exerted on the microsphere according to the gradient force and the scattering force.

[0096] Specifically, the optical force F exerted on the microsphere is divided into the gradient force F grad and the scattering force F scat , F=F grad +F scat Among them, the gradient force is mainly used to capture the microspheres stably, while the scattering force is mainly used to push the microspheres away from the stable capture point. Therefore, in order to achieve stable capture of microspheres, it is necessary to enhance the gradient force and reduce the scattering force.

[0097] S16. According to the axial optical force of the optical force acting on the microsphere and the stable capture formula, the test optical trap stiffness when the axial optical force acting on the microsphere is the maximum is calculated.

[0098] The stable capture formula is:

[0099] F z =-k z (zz 0 ).

[0100] Among them, F z is the axial optical force on the microsphere, z is when F z The position of the microsphere when it is at its maximum value, z 0 is the equilibrium point position, k z is the light trap stiffness.

[0101] Specifically, as long as there is a value of z where Fz has a zero point, it is considered that stable capture exists. By calculating the maximum axial optical force, the test optical trap stiffness is calculated. The schematic diagram of the microsphere 1 being stably captured is shown in FIG. Figure 2 shown.

[0102] Although the traditional simulation method can obtain the mapping relationship between some microsphere sizes, light trap stiffness and light field parameters, the traditional simulation method is computationally complex and the data that can be calculated is limited and cannot be adapted to all microsphere sizes.

[0103] S2. Establish a training database containing a mapping relationship according to the test microsphere size, the test light trap stiffness, and the test light field parameters.

[0104] Specifically, a training database is established with existing data, and a mapping relationship is established between the test microsphere size, the test light trap stiffness, and the test light field parameters. The training database is used to proxy the neural network model so that the neural network model learns to output the corresponding light field parameters when the input is the microsphere size and the light trap stiffness.

[0105] S3. Input the training database into the neural network model, use the test microsphere size and the test light trap stiffness as input, and the test light field parameters as output to train the neural network model, and obtain the neural network proxy model.

[0106] Specifically, after the neural network model is trained through the training database, the trained neural network model can obtain the corresponding light trap stiffness and light field parameters under any microsphere size.

[0107] S4. Setting initial light field parameters and obtaining real-time microsphere size, and calculating the current light trap stiffness according to the real-time microsphere size and the initial light field parameters as the initial light trap stiffness.

[0108] Specifically, the initial light field parameters are set by the user during the experiment, and the current light trap stiffness under the real-time microsphere size and the initial light field parameters is calculated as the initial light trap stiffness. The initial light trap stiffness can be obtained through the mapping relationship in the training database, or can be obtained through traditional simulation calculations, or can be calculated through traditional optical tweezers toolkits.

[0109] S5. Set the light field parameter range.

[0110] Specifically, when the capture object is an active cell, a light field parameter that is too large will affect the activity of the cell, and a light field parameter that is too small may make it impossible to capture. Therefore, a light field parameter range is set within which the cell can be captured without affecting the activity of the cell.

[0111] S6. Input the real-time microsphere size and the initial light trap stiffness into the neural network proxy model, increase the initial light trap stiffness by a fixed step size, compare the predicted light field parameters output by the neural network proxy model with the light field parameter range, find the maximum light trap stiffness corresponding to when the predicted light field parameters are greater than the light field parameter range, and combine the initial light trap stiffness and the maximum light trap stiffness to form the critical interval of the light trap stiffness.

[0112] The real-time microsphere size and initial light trap stiffness are input into the neural network proxy model. The initial light trap stiffness is increased by a fixed step length. The predicted light field parameters output by the neural network proxy model are compared with the light field parameter range. The maximum light trap stiffness corresponding to the predicted light field parameter is found when it is greater than the light field parameter range. The initial light trap stiffness and the maximum light trap stiffness form the critical interval of the light trap stiffness, including:

[0113] S61. Obtain a set step length.

[0114] Specifically, the specific value of the step size can be set by oneself, and the step size is used to control the magnitude of each change in the light trap stiffness when iterating the light trap stiffness.

[0115] S62. Obtain an iterative light trap stiffness according to the initial light trap stiffness and the set step size.

[0116] Specifically, iterative light trap stiffness = initial light trap stiffness + set step size.

[0117] S63. Input the iterative light trap stiffness and the real-time microsphere size into the neural network proxy model to obtain the predicted light field parameters.

[0118] S64. If the predicted light field parameters are within the light field parameter range, continue to increase the iterative light trap stiffness by setting the step size until the predicted light field parameters calculated by iterative light trap stiffness and real-time microsphere size exceed the light field parameter range. The first iterative light trap stiffness for which the predicted light field parameters calculated by iterative light trap stiffness and real-time microsphere size exceed the light field parameter range is taken as the maximum light trap stiffness.

[0119] Specifically, when iterating the light trap stiffness, the real-time microsphere size and the iterative light trap stiffness are input into the neural network proxy model. The output predicted light field parameters are not necessarily within the square parameter range. The larger the light trap stiffness, the more stable the capture. Therefore, the light trap stiffness is continuously increased, and the first iterative light trap stiffness whose predicted light field parameters output by iterative light trap stiffness and real-time microsphere size exceed the light field parameter range is found as the maximum light trap stiffness.

[0120] S65, combining the initial light trap stiffness and the maximum light trap stiffness into a light trap stiffness critical interval.

[0121] Specifically, after determining the critical interval of light trap stiffness, the optimal light trap stiffness is found within the critical interval of light trap stiffness to further obtain the optimal light field parameter. In this embodiment, the light field parameter is the optical power.

[0122] S7. Within the critical interval of the light trap stiffness, with the light field parameter range as a constraint, the light trap stiffness is iterated using the dichotomy method and the neural network proxy model to obtain the optimal light trap stiffness corresponding to the real-time microsphere size, and then the optimal light field parameters are obtained through the neural network proxy model.

[0123] In the critical interval of the light trap stiffness, the light field parameter range is constrained, the light trap stiffness is iterated using the dichotomy method and the neural network proxy model to obtain the optimal light trap stiffness corresponding to the real-time microsphere size, and then the optimal light field parameters obtained by the neural network proxy model include:

[0124] S71. In the critical interval of the light trap stiffness, use the dichotomy method to obtain the intermediate value of the light trap stiffness.

[0125] Specifically, the middle value of the light trap stiffness is obtained by using the dichotomy method: the middle value of the light trap stiffness is obtained by taking the average value of the maximum value and the minimum value of the critical interval of the light trap stiffness.

[0126] S72, substituting the intermediate value of the light trap stiffness and the real-time microsphere size into the neural network proxy model, iterating the light trap stiffness with the light field parameter range as a constraint, obtaining the optimal light trap stiffness corresponding to the real-time microsphere size, and then obtaining the optimal light field parameters through the neural network proxy model.

[0127] Specifically, the intermediate value of the light trap stiffness and the real-time microsphere size are substituted into the neural network proxy model to iterate the light trap stiffness, and the direction of the iteration is determined according to whether the calculated light field parameters are within the light field parameter range.

[0128] Substitute the intermediate value of the light trap stiffness and the real-time microsphere size into the neural network proxy model to iterate the light trap stiffness to obtain the optimal light trap stiffness corresponding to the real-time microsphere size. Then, the optimal light field parameters obtained by the neural network proxy model include:

[0129] S721. Substitute the intermediate value of the light trap stiffness and the real-time microsphere size into the neural network proxy model to iterate the light trap stiffness to obtain iterative light field parameters.

[0130] S722: If the iterative light field parameter is within the light field parameter range, the maximum value of the light trap stiffness intermediate value and the light trap stiffness critical interval is used as the iterative critical interval.

[0131] S723: If the iterative light field parameter is not within the light field parameter range, the minimum value of the light trap stiffness intermediate value and the light trap stiffness critical interval is used as the iterative critical interval.

[0132] Specifically, during the iteration process, if the intermediate value of the light trap stiffness and the real-time microsphere size are input into the neural network proxy model, and the output iterative light field parameter is not within the light field parameter range, it means that the current light trap stiffness is already the maximum value. Therefore, a new iterative critical interval is formed according to the minimum value of the light trap stiffness critical interval and the intermediate value of the light trap stiffness. If it is within the light field parameter range, the intermediate value of the light trap stiffness and the maximum value of the light trap stiffness critical interval form an iterative critical interval. According to this rule, iterations are continuously performed to form a new iterative critical interval.

[0133] S724: Set the fitness function and the stop threshold.

[0134] Specifically, the fitness function is expressed as:

[0135]

[0136] Where Δf is the stop threshold, abs represents the absolute value, and They represent the light trap stiffness at the current iteration and the light trap stiffness at the previous iteration respectively.

[0137] Stop Threshold is the threshold used to stop the light trap stiffness iteration.

[0138] S725. Continue to iterate the iteration critical interval using the binary search method until the fitness function value is less than the stopping threshold and the iterative light field parameters are within the light field parameter range, and take the light trap stiffness corresponding to the time when the fitness function value is less than the stopping threshold and the iterative light field parameters are within the light field parameter range as the optimal light trap stiffness, and then obtain the optimal light field parameters through the neural network model.

[0139] Specifically, in the process of iterating the light trap stiffness, if the absolute value of the difference between the light trap stiffness of two adjacent iterations is less than the stop threshold, the iteration is stopped. The light trap stiffness corresponding to the fitness function value being less than the stop threshold and the iterative light field parameter being within the light field parameter range is taken as the optimal light trap stiffness, and then the real-time microsphere size and the optimal light trap stiffness are input into the neural network proxy model to obtain the optimal light field parameter.

[0140] Figure 3 A schematic diagram showing the comparison of light field parameters obtained by the present application and the traditional optical tweezers toolkit simulation method, showing the K z =2.91×10 -2 pN·μm -1 mW -1 The correspondence between the size of silica microspheres and the light power when .

[0141] Specifically, at this time, the optimal light field parameters corresponding to the real-time microsphere size have been obtained, and the light field parameters are changed to the optimal light field parameters to adaptively match the light force to capture the capture object.

[0142] Compared with the traditional particle swarm algorithm, this application has higher stability and faster convergence speed. Taking the radius of the microsphere to be captured as 3196nm, when the light field parameter range is 100mW to 300mW, the optimal light power obtained by the adaptive optical force matching algorithm of this application is 299.8mW, and the maximum light trap stiffness is K z =2.48×10 -2 pN·μm -1 mW -1 , through the particle swarm optimization algorithm, three light field optimizations were performed, and three sets of optimization parameters were obtained: the optimal light power was 295.0mW, the maximum light trap stiffness was K z =2.40×10 -2 pN·μm -1 mW -1 ; The optimal optical power is 299.8mW, and the maximum light trap stiffness is K z =5.49×10 -2 pN·

[0143] μm -1 mW -1 ; The optimal optical power is 299.7mW, and the maximum light trap stiffness is K z =2.48×10 -2 pN·μm -1 mW -1 ; If the particle swarm algorithm is used for optimization, K in these three sets of data z =5.49×10 -2 pN·μm -1 mW -1 This group is the optimal solution. But it is different from the existing Kz in the database. best and the corresponding P best After comparison, it was found that this value was an outlier. mW -2 It is considered that the optimal solution is found, where ΔK z =K z -Kz best , ΔP=abs(P z -P best ), and finally the optimal power of 299.7mW was selected as the optimal light field parameter, corresponding to the maximum light trap stiffness K z =2.48×10 -2 pN·μm -1 mW -1 Therefore, the particle swarm algorithm has a local optimal solution, which leads to inaccurate optimal light field parameters.

[0144] The present application can also train a neural network proxy model with microsphere size and light field parameters as input and light trap stiffness as output by equally dividing the space spanned by microsphere size and light field parameters into selected points as data groups, and using the data groups as training sets to feed the neural network. The function of this neural network proxy model is to use the calculation of the neural network to replace the traditional method of simulation calculation to obtain the result. After the training is completed, the neural network proxy model can quickly calculate the corresponding light trap stiffness based on any given set of microsphere size and light field parameters, thereby greatly shortening the simulation time, and the error between the neural network calculation result and the traditional calculation of the light trap stiffness is very small.

[0145] The present application can quickly predict the light trap stiffness of the microspheres when they are captured in real time based on the light field parameters and the size of the captured microspheres, which can reduce the real-time calculation time cost by one order of magnitude. If the light trap stiffness prediction is to be performed for a large number of microspheres of different sizes under a wide range of light field parameters, the simulation time cost can be reduced by three orders of magnitude, greatly improving the calculation efficiency.

[0146] Traditional optical tweezers kit: The time required is about 50 seconds per particle;

[0147] Neural network: The time required is about 2 seconds per particle.

[0148] When simulating the optical trap stiffness of a large number of microspheres (about 550), the traditional optical tweezers toolkit takes about 25 hours, while the neural network takes about 40 seconds.

[0149] Figure 4 The prediction results of the axial optical trap stiffness of microspheres of different sizes when P = 100 mW, with a mean square error of 1.99×10 -7 .

[0150] Figure 5 The predicted results of the axial optical trap stiffness of microspheres of different sizes with different optical powers are shown in Figure 1. The mean square error is 1.01×10 -6 The mean square error is used to measure the degree of difference between the predicted value of the neural network model and the true value. The smaller the mean square error, the better the model capability.

[0151] Figure 6 This is a schematic diagram of the calculation results using the traditional optical tweezers toolkit. Figure 5 and Figure 6 , it can be seen that the results predicted by the neural network proxy model are highly similar to those calculated using the traditional optical tweezers toolkit, and the neural network proxy model takes less time.

[0152] S8. Capture the capture object having the size of a real-time microsphere by matching the optical force with the optimal light field parameters.

[0153] A cell capture parameter determination system based on deep learning, comprising:

[0154] A calculation module, used to obtain the corresponding test light trap stiffness according to the test microsphere size, the test light field parameters and the optical force calculation model;

[0155] A training database establishment module is used to establish a training database containing a mapping relationship according to the test microsphere size, the test light trap stiffness and the test light field parameters;

[0156] A training module is used to input the training database into the neural network model, and train the neural network model with the test microsphere size and the test light trap stiffness as input and the test light field parameters as output to obtain a neural network proxy model;

[0157] An initialization module is used to set initial light field parameters and obtain real-time microsphere size. According to the real-time microsphere size and initial light field parameters, the current light trap stiffness is calculated as the initial light trap stiffness.

[0158] A setting module, used to set the light field parameter range;

[0159] A comparison module is used to input the real-time microsphere size and the initial light trap stiffness into the neural network proxy model, increase the initial light trap stiffness by a fixed step length, compare the predicted light field parameters output by the neural network proxy model with the light field parameter range, find the maximum light trap stiffness corresponding to when the predicted light field parameters are greater than the light field parameter range, and combine the initial light trap stiffness and the maximum light trap stiffness to form a critical interval of light trap stiffness;

[0160] An iteration module is used to iterate the light trap stiffness within the critical interval of the light trap stiffness by using the dichotomy method and the neural network proxy model to obtain the optimal light trap stiffness corresponding to the real-time microsphere size, and then obtain the optimal light field parameters through the neural network proxy model;

[0161] The capture module is used to capture the capture object with the size of the real-time microsphere by matching the optical force through the optimal light field parameters.

[0162] An embodiment of the present application also discloses a terminal device, including a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor loads and executes the computer program, an adaptive optical force matching method based on deep learning is adopted.

[0163] The terminal device may be a computer device such as a desktop computer, a laptop computer or a cloud server, and the terminal device includes but is not limited to a processor and a memory. For example, the terminal device may also include input and output devices, a network access device and a bus.

[0164] Among them, the processor can adopt a central processing unit (CPU). Of course, according to actual usage, other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. can also be adopted. The general-purpose processor can adopt a microprocessor or any conventional processor, etc., and this application does not impose any restrictions on this.

[0165] Among them, the memory can be an internal storage unit of the terminal device, such as a hard disk or memory of the terminal device, or it can be an external storage device of the terminal device, such as a plug-in hard disk, a smart memory card (SMC), a secure digital card (SD) or a flash memory card (FC) equipped on the terminal device, etc., and the memory can also be a combination of an internal storage unit and an external storage device of the terminal device. The memory is used to store computer programs and other programs and data required by the terminal device. The memory can also be used to temporarily store data that has been output or is to be output, and this application does not impose any restrictions on this.

[0166] Among them, through this terminal device, an adaptive optical force matching method based on deep learning in the above embodiment is stored in the memory of the terminal device, and is loaded and executed on the processor of the terminal device for easy use.

[0167] An embodiment of the present application further discloses a computer-readable storage medium, and the computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, an adaptive optical force matching method based on deep learning in the above embodiment is adopted.

[0168] Among them, the computer program can be stored in a computer-readable medium, the computer program includes computer program code, the computer program code can be in the form of source code, object code, executable file or certain middleware, etc. The computer-readable medium includes any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the computer-readable medium includes but is not limited to the above-mentioned components.

[0169] Among them, through this computer-readable storage medium, an adaptive optical force matching method based on deep learning in the above embodiment is stored in a computer-readable storage medium, and is loaded and executed on a processor to facilitate the storage and application of the above method.

[0170] A person skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of protection of the present application is limited to these examples. In line with the concept of the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.

[0171] One or more embodiments of the present application are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the present application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of the present application should be included in the protection scope of the present application.< / s> < / s>

Claims

1. An adaptive optical force matching method based on deep learning, characterized in that: include: According to the test microsphere size, test light field parameters and optical force calculation model, the corresponding test light trap stiffness is obtained; Establishing a training database including a mapping relationship according to the test microsphere size, the test light trap stiffness, and the test light field parameters; Inputting the training database into the neural network model, training the neural network model with the test microsphere size and the test light trap stiffness as input and the test light field parameters as output to obtain a neural network proxy model; Setting initial light field parameters, and obtaining a real-time microsphere size, and calculating a current light trap stiffness according to the real-time microsphere size and the initial light field parameters as an initial light trap stiffness; Set the light field parameter range; Inputting the real-time microsphere size and the initial light trap stiffness into a neural network proxy model, increasing the initial light trap stiffness by a fixed step length, comparing the predicted light field parameters output by the neural network proxy model with the light field parameter range, finding the maximum light trap stiffness corresponding to when the predicted light field parameters are greater than the light field parameter range, and combining the initial light trap stiffness and the maximum light trap stiffness to form a critical interval of light trap stiffness; In the critical interval of the light trap stiffness, the light field parameter range is constrained, the light trap stiffness is iterated using the dichotomy method and the neural network proxy model to obtain the optimal light trap stiffness corresponding to the real-time microsphere size, and then the optimal light field parameter is obtained through the neural network proxy model; The capture object having a size equal to the size of the real-time microsphere is captured by matching the optical force with the optimal light field parameters.

2. The deep learning-based adaptive optical force matching method according to claim 1, characterized in that: The obtaining of the corresponding test light trap stiffness according to the microsphere size, light field parameters and light force calculation model comprises: According to the test light field parameters, the electric field intensity is obtained; According to the electric field strength and the gradient force formula, the gradient force of the light force on the microsphere is obtained; Obtain the refractive index of the environment medium and the refractive index of the microsphere; According to the refractive index of the environment medium, the refractive index of the microsphere, the size of the microsphere and the scattering force formula, the scattering force of the light force exerted on the microsphere is obtained; Obtaining the light force exerted on the microsphere according to the gradient force and the scattering force; According to the axial optical force of the optical force exerted on the microsphere and the stable capture formula, the stiffness of the test optical trap when the axial optical force exerted on the microsphere is the maximum is calculated.

3. The deep learning-based adaptive optical force matching method according to claim 2, characterized in that: The gradient force formula is: Where c is the speed of light in vacuum, ε0 is the dielectric constant in vacuum, is the square of the electric field gradient, F grad is the gradient force; Among them, n med is the refractive index of the surrounding medium, m is the relative refractive index, m = n part / n med , n part is the refractive index of the microsphere, W is the size of the microsphere; The scattering force formula is: Among them, F scat is the scattering force, <s> is the time-averaged Poynting vector, σ is the scattering cross section;< / s> <s> where a is related to the wavelength and n med The relevant constants; The stable capture formula is: F z =-k z (z-z0); Among them, F z is the axial optical force on the microsphere, z is when F z is the position of the microsphere at the maximum value, z0 is the equilibrium point, k z is the light trap stiffness.

4. The deep learning-based adaptive optical force matching method according to claim 1, characterized in that: The real-time microsphere size and the initial light trap stiffness are input into the neural network proxy model, the initial light trap stiffness is increased by a fixed step length, the predicted light field parameters output by the neural network proxy model are compared with the light field parameter range, and the maximum light trap stiffness corresponding to the predicted light field parameter is found when the light field parameter is greater than the light field parameter range. The initial light trap stiffness and the maximum light trap stiffness are combined to form a light trap stiffness critical interval, including: Get the set step size; Obtaining an iterative light trap stiffness according to the initial light trap stiffness and a set step size; Inputting the iterative light trap stiffness and the real-time microsphere size into a neural network proxy model to obtain predicted light field parameters; If the predicted light field parameter is within the light field parameter range, the iterative light trap stiffness is continuously increased by setting the step size until the predicted light field parameter calculated by iterative light trap stiffness and the real-time microsphere size exceeds the light field parameter range, and the first iterative light trap stiffness whose predicted light field parameter calculated by iterative light trap stiffness and the real-time microsphere size exceeds the light field parameter range is taken as the maximum light trap stiffness; The initial light trap stiffness and the maximum light trap stiffness form a light trap stiffness critical interval.

5. The deep learning-based adaptive optical force matching method according to claim 1, characterized in that: In the critical interval of the light trap stiffness, the light field parameter range is constrained, the light trap stiffness is iterated using the dichotomy method and the neural network proxy model to obtain the optimal light trap stiffness corresponding to the real-time microsphere size, and then the optimal light field parameters are obtained through the neural network proxy model, including: In the critical interval of the light trap stiffness, a dichotomy method is used to obtain an intermediate value of the light trap stiffness; The intermediate value of the light trap stiffness and the real-time microsphere size are substituted into the neural network proxy model, and the light trap stiffness is iterated with the light field parameter range as a constraint to obtain the optimal light trap stiffness corresponding to the real-time microsphere size, and then the optimal light field parameters are obtained through the neural network proxy model.

6. The deep learning-based adaptive optical force matching method according to claim 5, characterized in that: Substituting the intermediate value of the light trap stiffness and the real-time microsphere size into the neural network proxy model, iterating the light trap stiffness with the light field parameter range as a constraint, obtaining the optimal light trap stiffness corresponding to the real-time microsphere size, and then obtaining the optimal light field parameters through the neural network proxy model includes: Substituting the intermediate value of the light trap stiffness and the real-time microsphere size into a neural network proxy model, iterating the light trap stiffness, and obtaining iterative light field parameters; If the iterative light field parameter is within the light field parameter range, taking the maximum value of the light trap stiffness intermediate value and the light trap stiffness critical interval as the iterative critical interval; If the iterative light field parameter is not within the light field parameter range, taking the minimum value of the light trap stiffness intermediate value and the light trap stiffness critical interval as the iterative critical interval; Set the fitness function and stop threshold; The iterative critical interval is continuously iterated using the dichotomy method until the fitness function value is less than the stop threshold and the iterative light field parameters are within the light field parameter range, and the light trap stiffness corresponding to the time when the fitness function value is less than the stop threshold and the iterative light field parameters are within the light field parameter range is taken as the optimal light trap stiffness, and then the optimal light field parameters are obtained through the neural network model.

7. The deep learning-based adaptive optical force matching method according to claim 6, characterized in that: The fitness function is: Where Δf is the stop threshold, abs represents the absolute value, and They represent the light trap stiffness at the current iteration and the light trap stiffness at the previous iteration respectively.

8. A cell capture parameter determination system based on deep learning, characterized in that: include: A calculation module, used to obtain the corresponding test light trap stiffness according to the test microsphere size, the test light field parameters and the optical force calculation model; A training database establishment module, used to establish a training database containing a mapping relationship according to the test microsphere size, the test light trap stiffness and the test light field parameters; A training module, used for inputting the training database into the neural network model, training the neural network model with the test microsphere size and the test light trap stiffness as input and the test light field parameters as output, to obtain a neural network proxy model; An initialization module is used to set initial light field parameters and obtain real-time microsphere size, and calculate the current light trap stiffness as the initial light trap stiffness according to the real-time microsphere size and the initial light field parameters; A setting module, used to set the light field parameter range; A comparison module is used to input the real-time microsphere size and the initial light trap stiffness into a neural network proxy model, increase the initial light trap stiffness by a fixed step length, compare the predicted light field parameters output by the neural network proxy model with the light field parameter range, find the maximum light trap stiffness corresponding to when the predicted light field parameters are greater than the light field parameter range, and combine the initial light trap stiffness and the maximum light trap stiffness to form a critical interval of light trap stiffness; An iterative module is used to iterate the light trap stiffness within the critical interval of the light trap stiffness and with the light field parameter range as a constraint, using a dichotomy method and a neural network proxy model to obtain an optimal light trap stiffness corresponding to the real-time microsphere size, and then obtain an optimal light field parameter through the neural network proxy model; The capture module is used to capture the capture object having the size of the real-time microsphere by matching the optical force with the optimal light field parameters.

9. A terminal device, comprising a memory and a processor, characterized in that: The memory stores a computer program that can be run on the processor, and when the processor loads and executes the computer program, the method according to any one of claims 1 to 7 is adopted.

10. A computer-readable storage medium having a computer program stored therein, characterized in that: When the computer program is loaded and executed by a processor, the method according to any one of claims 1 to 7 is adopted. < / s>

Images