CN113252586A

CN113252586A - Hyperspectral image reconstruction method, terminal device and computer-readable storage medium

Info

Publication number: CN113252586A
Application number: CN202110467817.XA
Authority: CN
Inventors: 梁正平; 王志强; 林万鹏
Original assignee: Shenzhen University
Current assignee: Shenzhen University
Priority date: 2021-04-28
Filing date: 2021-04-28
Publication date: 2021-08-13
Anticipated expiration: 2041-04-28
Also published as: CN113252586B

Abstract

The application is suitable for the field of hyperspectral image reconstruction, and provides a hyperspectral image reconstruction method, a hyperspectral image reconstruction device, terminal equipment and a computer-readable storage medium, wherein the hyperspectral image reconstruction method comprises the following steps: establishing a target function in a decision space according to a target variable of a hyperspectral image reconstruction task; establishing an initial population in a decision space; estimating the curvature of the pareto frontier according to the initial population; generating a first reference vector for matching the pareto front surface according to the curvature; and carrying out environment selection processing on the initial population according to the curvature and the first reference vector to obtain an elite population. By the method, the matching degree of the reference vector and the real pareto frontier plane of the problem to be solved can be improved, and further, the multi-objective optimization result is perfected, so that the comprehensive performance of hyperspectral image reconstruction is improved.

Description

Hyperspectral image reconstruction method, terminal device and computer-readable storage medium

Technical Field

The application belongs to the technical field of hyperspectral image reconstruction, and particularly relates to a hyperspectral image reconstruction method and device, terminal equipment and a computer-readable storage medium.

Background

Spectral images with spectral resolution in the order of 10 λ are called Hyperspectral images (Hyperspectral images). The spectral imaging technology is based on the imaging technology, and uses a light splitting element, such as a prism, a grating, etc., to acquire image information of an object in each wavelength band. The spectral information of the object can be extracted from the image information, so that the spectral curve and the image of the target are obtained. The composition of the object can be detected by using the spectral curve, and the image information provides characteristics such as appearance of the object, and the combination of the two provides great convenience for analyzing biochemical characteristics of the target object. With the development of spectral imaging technology, the spectral resolution is higher and higher, and the system volume is smaller and smaller, and the system is developed from a dispersion type imaging spectrometer which is large in volume and needs fine scanning to a portable device which does not need scanning. According to different scanning wave bands, the system can be divided into a multispectral imaging system, a hyperspectral imaging system and a hyperspectral imaging system, wherein the wave bands are respectively 10-20 nm, 100-200 nm and 1000-2000 nm.

A Liquid Crystal Tunable Filter (LCTF) imaging system adopts a single-area array CCD liquid crystal tunable filter structure, the LCTF is a novel optical device manufactured according to the electric control birefringence effect of liquid crystal and the interference principle of polarized light, and has the advantages of narrow bandwidth, low power consumption, wide tuning range, low driving voltage, simple structure, wide view field angle, large aperture, no moving parts and the like. The LCTF changes the transmittance of light of the liquid crystal filter for each wavelength band by changing a voltage, so that one voltage corresponds to one transmittance curve, the horizontal axis represents the wavelength band, and the vertical axis represents the transmittance. Therefore, the spectral information can be compressed, and then a more complete hyperspectral image with larger information amount is reconstructed by using a compressed sensing theory and with fewer measurement times (one voltage corresponds to one measurement) (sampling times and photographing times). Therefore, a novel and effective solution idea can be provided by a multi-target evolution algorithm based on decomposition if a combination of measurement times as few as possible is required to be reconstructed to a hyperspectral image as complete as possible.

Multiobjective optimization refers to the process of finding the optimal solution for multiple objectives in a feasible domain. The difficulty of multi-objective optimization compared to single-objective optimization is that it focuses on not the optimal solution of a single objective, but on the equilibrium relationship between the optimal solutions of each of multiple objectives. The result of the multi-objective optimization is a set comprising a plurality of sets of optimal solutions, and the values of the objective functions corresponding to the optimal solutions form a pareto frontier. The evolutionary algorithm is used as a heuristic algorithm derived from a biological evolutionary thought, has the characteristics of high robustness and wide adaptability, and provides a novel and effective solution thought for solving the multi-objective optimization problem. In the prior art, a decomposition-based evolutionary algorithm has a great competitive advantage, and the core idea is to divide a target space by using a group of reference vectors, so that a multi-target optimization problem is decomposed into a plurality of single-target optimization problems which can be solved independently. The performance of the algorithm depends on the matching degree of the adopted reference vector and the real pareto frontage of the problem to be solved to a great extent, and the reference vector generated by the existing method cannot be effectively matched with the real pareto frontage, so that the optimization result is poor.

Disclosure of Invention

The embodiment of the application provides a hyperspectral image reconstruction method and device, terminal equipment and a computer-readable storage medium, which can improve the matching degree of a reference vector and a real pareto frontier of a problem to be solved, further improve a multi-objective optimization result and improve the comprehensive performance of hyperspectral image reconstruction.

In a first aspect, an embodiment of the present application provides a hyperspectral image reconstruction method, where the hyperspectral image reconstruction method includes:

establishing a target function in a decision space according to a target variable of a hyperspectral image reconstruction task;

establishing an initial population in the decision space, wherein each individual in the initial population represents a set of initial solutions of decision variables of the hyperspectral image reconstruction task;

estimating the curvature of the pareto frontier according to the initial population;

generating a first reference vector for matching the pareto front surface according to the curvature;

and carrying out environment selection processing on the initial population according to the curvature and the first reference vector to obtain an elite population, wherein individuals in the elite population represent a group of pareto optimal solutions of the decision variables of the hyperspectral image reconstruction task.

In the embodiment of the application, the curvature of the pareto front surface is estimated according to the initial population, and then a first reference vector used for matching the pareto front surface is generated according to the estimated curvature; since the estimated curvature can more accurately reflect the shape of the real pareto front surface, the first reference vector generated according to the estimated curvature can more accurately match the real pareto front surface; and finally, carrying out environment selection processing on the initial population according to the estimated curvature and the first reference vector to obtain an elite population. By the method, the matching degree of the reference vector and the real pareto frontier plane of the problem to be solved can be effectively improved, and further, the multi-objective optimization result is perfected, so that the comprehensive performance of hyperspectral image reconstruction is improved.

In one possible implementation manner of the first aspect, the generating a first reference vector for matching the pareto front surface according to the curvature includes:

calculating the number of first sampling points according to a target dimension and the number of individuals of the initial population, wherein the target dimension represents the number of target functions;

if the number of the first sampling points is greater than or equal to the target dimension, generating the first reference vector according to a first preset method, wherein the first preset method is a method for generating a vector according to outer layer data, and the outer layer data is data on an edge line of the pareto frontier calculated according to the curvature;

and if the number of the first sampling points is smaller than the target dimension, generating the first reference vector according to a second preset method, wherein the second preset method is a method for generating vectors according to outer layer data and inner layer data, and the inner layer data is data on a non-edge line of the pareto front surface calculated according to the curvature.

In a possible implementation manner of the first aspect, the generating the first reference vector according to a first preset method includes:

sampling data on the arc line segment corresponding to the curvature according to the number of the first sampling points to obtain first sampling data;

generating a second reference vector according to the first sampling data;

determining the second reference vector as the first reference vector.

In a possible implementation manner of the first aspect, the generating the first reference vector according to a second preset method includes:

calculating the number of second sampling points according to the target dimension, the number of individuals of the initial population and the number of the first sampling points;

sampling data on the arc line segment corresponding to the curvature according to the number of the second sampling points to obtain second sampling data;

generating a third reference vector according to the second sampling data;

determining the first reference vector from the second reference vector and the third reference vector.

In a possible implementation manner of the first aspect, the performing environment selection processing on the initial population according to the curvature and the first reference vector to obtain an elite population includes:

acquiring a cross variation population of the initial population, wherein the cross variation population comprises a parent population and a child population of the initial population;

selecting a candidate population from the cross variation population according to the curvature and the first reference vector;

if the number of individuals of the candidate population is greater than that of the cross variation population, selecting the elite population meeting a second preset condition from the candidate population;

and if the number of individuals of the candidate population is less than or equal to the number of individuals of the cross variation population, determining the elite population according to the candidate population and the curvature.

In a possible implementation manner of the first aspect, the selecting a candidate population from the cross variation populations according to the curvature and the first reference vector includes:

screening out a fourth reference vector from the first reference vector, wherein the fourth reference vector is related to non-dominant solution in the cross variation population;

for each of the fourth reference vectors, if the curvature is within a preset numerical range, calculating an aggregation function value between each non-dominant solution in the cross variant population and the fourth reference vector based on a first aggregation function;

if the curvature is not within a preset numerical range, calculating an aggregation function value between each non-dominant solution in the cross variant population and the fourth reference vector based on a second aggregation function;

and adding a non-dominant solution in the cross variation population corresponding to the minimum value in the aggregation function values to the candidate population.

In a possible implementation manner of the first aspect, if the number of individuals of the candidate population is less than or equal to the number of individuals of the cross variation population, determining the elite population according to the candidate population and the curvature includes:

if the number of individuals of the candidate population is smaller than that of the cross variation population, respectively calculating the minimum distance between each non-dominant solution in the cross variation population and the candidate population;

obtaining a fifth reference vector, wherein the fifth reference vector is a non-dominated solution corresponding to a maximum value in the minimum distances in the cross variation population;

screening out target individuals from a non-dominant solution of the cross variation population according to the curvature and the fifth reference vector;

adding the target individuals to the candidate population, and deleting the target individuals in the cross-variant population.

In a possible implementation manner of the first aspect, after performing environment selection processing on the initial population according to the curvature and the first reference vector to obtain an elite population, the method includes:

and continuously carrying out multi-objective optimization processing on the elite population until the preset evolution times are met.

In a second aspect, an embodiment of the present application provides a hyperspectral image reconstruction apparatus, including:

the modeling unit is used for establishing a target function in a decision space according to a target variable of the hyperspectral image reconstruction task;

a population obtaining unit, configured to establish an initial population in the decision space, where each individual in the initial population represents a set of initial solutions of a decision variable of the hyperspectral image reconstruction task;

a curvature estimation unit for estimating a curvature of the pareto frontier based on the initial population;

a vector generation unit configured to generate a first reference vector for matching the pareto front surface according to the curvature;

and the environment selection unit is used for carrying out environment selection processing on the initial population according to the curvature and the first reference vector to obtain an elite population, wherein individuals in the elite population represent a group of pareto optimal solutions of the decision variables of the hyperspectral image reconstruction task.

In a third aspect, an embodiment of the present application provides a terminal device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor, when executing the computer program, implements the hyperspectral image reconstruction method according to any of the first aspects.

In a fourth aspect, the present application provides a computer-readable storage medium, and the present application provides a computer-readable storage medium, where a computer program is stored, where the computer program is executed by a processor to implement the hyperspectral image reconstruction method according to any of the above first aspects.

In a fifth aspect, an embodiment of the present application provides a computer program product, which when run on a terminal device, causes the terminal device to execute the hyperspectral image reconstruction method according to any of the above first aspects.

It is understood that the beneficial effects of the second aspect to the fifth aspect can be referred to the related description of the first aspect, and are not described herein again.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments or the prior art descriptions will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise.

FIG. 1 is a schematic flow chart of a hyperspectral image reconstruction method provided by an embodiment of the application;

FIG. 2 is a schematic diagram of a hierarchical sampling point provided by an embodiment of the present application;

FIG. 3 is a schematic diagram of a reference vector generation process provided by an embodiment of the present application;

FIG. 4 is a schematic diagram of environment selection provided by an embodiment of the present application;

fig. 5 is a block diagram of a hyperspectral image reconstruction apparatus provided in an embodiment of the present application;

fig. 6 is a schematic structural diagram of a terminal device according to an embodiment of the present application.

Detailed Description

In the following description, for purposes of explanation and not limitation, specific details are set forth, such as particular system structures, techniques, etc. in order to provide a thorough understanding of the embodiments of the present application. It will be apparent, however, to one skilled in the art that the present application may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.

It will be understood that the terms "comprises" and/or "comprising," when used in this specification and the appended claims, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.

As used in this specification and the appended claims, the term "if" may be interpreted contextually as "when.. or" upon "or" in response to a determination "or" in response to a detection ".

Furthermore, in the description of the present application and the appended claims, the terms "first," "second," "third," and the like are used for distinguishing between descriptions and not necessarily for describing or implying relative importance.

Reference throughout this specification to "one embodiment" or "some embodiments," or the like, means that a particular feature, structure, or characteristic described in connection with the embodiment is included in one or more embodiments of the present application. Thus, appearances of the phrases "in one embodiment," "in some embodiments," "in other embodiments," or the like, in various places throughout this specification are not necessarily all referring to the same embodiment, but rather "one or more but not all embodiments" unless specifically stated otherwise.

In recent decades, with social progress and technological development, hyperspectral remote sensing plays an increasingly important role in various fields. The hyperspectral imaging technology is an important part of the technology, and compared with a general remote sensing technology, the hyperspectral data has the characteristics of large information amount and high dimensionality, and due to the characteristics, the hyperspectral remote sensing technology can more accurately acquire ground object information. However, these characteristics also bring some problems in data processing, and how to process data more effectively is a leading edge of current hyperspectral technology research.

The hyperspectral remote sensing technology is created from twenty years ago, and has been developed for a long time, but a plurality of problems still need to be solved. The most prominent aspect is that the hyperspectral data, because it contains too many bands, makes it more difficult to process the data. Moreover, different remote sensing platforms due to different environmental factors cause unsatisfactory quality of hyperspectral data, and the quality of the data is reduced, so that the spectrum matching effect is influenced. In order to solve the problem that the difficulty in processing data is increased due to too many wave bands, a multi-objective optimization algorithm is introduced to solve the problem, the algorithm has the characteristics of high robustness and wide adaptability, a good solution effect can be still ensured for the discontinuous and undifferentiable complex multi-objective optimization problem, and the method is very suitable for processing the hyperspectral image reconstruction problem.

The multi-objective optimization algorithm is explained as follows:

generally, a multi-objective optimization problem (MOP) can be expressed in the following mathematical form:

wherein x is (x)₁,x₂,…,x_n) Is an n-dimensional decision vector in the decision space, which is the M objective functions to be solved. The objective of the multi-objective optimization is to find out N groups of optimal solutions from the feasible solutions according to a plurality of objective functions and preset constraint conditions, wherein the numerical values of the objective functions corresponding to the N groups of optimal solutions form a pareto frontier.

When M is greater than 3, it is also called a hyper-objective optimization problem (MaOPs). The evolutionary algorithm based on decomposition has a great competitive advantage for solving the MaOPs, and is widely applied to solving various MaOPs. The core of the decomposition evolution algorithm is to divide a target space by using a reference vector, decompose a single super-multi-target optimization problem into a plurality of single-target optimization problems and then solve the problems. A key factor affecting the performance of such algorithms is the degree of matching of the employed reference vector with the real pareto front surface (PF) of the problem to be solved.

Common reference vector generation methods in a decomposition-based hyper-multi-objective evolution algorithm (MaOEA/D) include three types of NBI, K-layer, and texture Uniform Design (MUD). Among them, NBI is the most widely used reference vector generation method, and is very suitable for the processing of the linear PF optimization problem. The K-layer method has a good effect on the optimization problem of the concave PF. The advantage of the MUD approach is that any number of reference vectors are supported, but the diversity of the generated reference vectors is not as diverse as NBI and K-layer. In general, the three types of reference vector generation methods are respectively suitable for processing some specific regular PF optimization problems, and for irregular PF optimization problems such as disconnection, degradation, preference and the like, due to the fact that the generated reference vectors are not uniformly distributed on the PF, a good effect is difficult to obtain.

In order to support the processing of optimization problems of PF with different shapes, the existing MaOEA/D algorithm generally introduces a strategy of self-adaptive adjustment of a reference vector in the population evolution process. According to different heuristic information sources used in the adjusting process, the method can be divided into a random reference vector, a fitting self-adaptive reference vector, a local population guided self-adaptive reference vector, a local archive guided self-adaptive reference vector, a self-adaptive reference vector based on an adjacent reference vector, a self-adaptive reference vector based on preference and the like. Although the above reference vector adjustment strategy can achieve a good effect in solving various irregular PF optimization problems, it is easy to cause performance deterioration in processing regular PF optimization problems.

In order to simultaneously process PF optimization problems of different shapes and better balance diversity and convergence of a population, the embodiment of the application provides a hyperspectral image reconstruction method.

According to the hyperspectral image reconstruction method, firstly, an objective function in a decision space is established according to an objective variable of a hyperspectral image reconstruction task, and the decision variable is determined.

For example, a voltage value may be determined as a decision variable. The corresponding multi-objective optimization problem is then: the most suitable at least 2 voltage values are selected from the plurality of voltage values. At least 2 of the number of voltages, sparsity, euclidean distance, spectral angle, and matrix coherence may be taken as target variables. Specifically, the method comprises the following steps:

1. voltage number: one voltage represents a photographing sampling measurement, the transmittance of each spectrum can be different under each specific voltage, and the smaller the voltage number is, the shorter the total measurement time is, and more information can be compressed.

2. Sparsity: the sparsity of the sparse coefficient of the original complete spectrum signal is smaller, and the image reconstruction time is shorter.

3. Euclidean distance: the Euclidean distance between the spectrum signal in the reconstructed image and the spectrum signal in the original image is smaller, which indicates that the reconstruction effect is better and the error is smaller.

4. Spectral angle: the smaller the included angle between the spectrum signal in the reconstructed image and the spectrum signal in the original image is, the better the reconstruction effect is, and the smaller the error is.

5. Matrix coherence: and measuring the coherence of the matrix and sparse bases, wherein the smaller the coherence is, the better the reconstruction effect is, and more information can be compressed.

An algorithm framework of the hyperspectral image reconstruction method provided by the embodiment of the application is described below, and is shown as follows.

Algorithm 1: algorithm framework of hyperspectral image reconstruction method

1-1 initial population Q (number of individuals N)

1-2 while not reaching the preset evolution times do

1-3, estimating the curvature p of the PF according to the current population Q

1-4 generating a first reference vector W from a curvature p

1-5 Generation of progeny O Using Cross-variations

1-6 Environment selection of QU O based on p and W

1-7:end while

1-8 output Elite population P (number of individuals N)

The algorithm 1 described above is described below. Referring to fig. 1, which is a schematic flowchart of a hyperspectral image reconstruction method provided in an embodiment of the present application, by way of example and not limitation, the method may include the following steps.

And S101, acquiring an initial population.

Wherein each individual in the initial population represents a set of initial solutions for decision variables of the hyperspectral image reconstruction task. In the model shown in the above formula (1), each set of initial solutions includes n solutions, where n is the number of target variables of the problem to be solved.

The hyperspectral image reconstruction provided by the embodiment of the application is based on an evolutionary algorithm, namely, an elite population after evolution is obtained by evolving an initial population. In the first evolution, the steps include S102-S106. The details are as follows.

And S102, estimating the curvature of the pareto frontier according to the initial population.

This corresponds to steps 1-3 in algorithm 1 above. In one embodiment, the curvature estimation method may include:

I. normalization

Since the scaling of the PF affects the estimation of the curvature p, the generated first reference vector does not match the shape of the real PF well. For this reason, the initial population is normalized before estimating the curvature of the PF. In the embodiment of the present application, the minimum value of the non-dominant layer of the initial population Q on each dimension component is taken as the corresponding component of the ideal point z. The maximum value of the corner solution on each dimension component is taken as the corresponding component of the extreme difference point znad. The corner solution is an individual closest to each direction axis (coordinate axis in the coordinate system where the pareto front surface is located), and the calculation formula of the ith dimension component value is as follows:

wherein, x is a target vector corresponding to the corner solution, e is a unit direction vector of each axis, and dist is the Euclidean distance from x to the axial unit direction vector e.

It should be noted that the above dimension refers to the number of objective functions in the hyperspectral image reconstruction problem. For example: when 2 objective functions are provided, 2 dimensions are provided, and the formed pareto frontier is a curve. When 3 objective functions are provided, 3 dimensions are provided, and the formed pareto frontier is a curved surface.

After the ideal point z and the range point znad are calculated, the formula for normalizing each dimension of x is as follows:

II. Estimating curvature

Because of the poor quality of the dominant individuals in the population, estimates of curvature are easily misled. To this end, the current 2REA method is used herein to select only the non-dominant individuals in the starting population and base their L_pThe distance is used to adaptively estimate the curvature p. Wherein L is_pThe calculation formula of the distance is as follows:

the curvature estimation method comprises the following specific processes: firstly, limiting a p value in a proper value area, sampling based on a certain interval, and then respectively calculating the L of all non-dominant individuals under each value in the value area_pThe standard deviation corresponding to the distance. Since the closer the selected p-value corresponds to a surface to the approximate PF formed by non-dominated individuals, the smaller the corresponding standard deviation. Therefore, the p-value curved surface with the minimum standard deviation is closest to the approximate PF formed by the non-dominant individuals, so that the initial population can be fitted by the curved surface corresponding to the p value, and the p value is used as the estimated value of the curvature of the PF corresponding to the initial population.

And S103, generating a first reference vector for matching the pareto front surface according to the curvature.

Step S103 corresponds to steps 1-4 in algorithm 1 above. For a specific method for generating the first reference vector, reference may be made to algorithm 2 described in the following embodiments, which is not described herein again.

And S104, carrying out environment selection processing on the initial population according to the curvature and the first reference vector to obtain an elite population.

Each individual in the elite population represents a set of pareto optimal solutions for the decision variables of the hyperspectral image reconstruction task.

Step S104 corresponds to steps 1-5 and 1-6 in algorithm 1 above. The specific environment selection processing method can be referred to as algorithm 3 described in the following embodiments, and is not described herein again.

And S105, if the current evolution frequency reaches the preset evolution frequency, determining the current obtained elite population as the final elite population, and stopping the evolution.

And S106, if the current evolution times do not reach the preset evolution times, continuously performing multi-objective optimization processing on the currently obtained elite population.

Here, continuing to perform the multi-objective optimization processing on the elite population currently obtained means that the above-described steps S102 to S106 are performed on the elite population, specifically:

re-estimating the curvature of the pareto frontier according to the elite population; regenerating a first reference vector for matching the pareto frontier from the re-estimated curvature; performing environment selection processing on the elite population according to the re-estimated curvature and the re-generated first reference vector to obtain a new elite population; if the current evolution times reach the preset evolution times, determining the new elite population obtained currently as the final elite population, and stopping the evolution; and if the current evolution times do not reach the preset evolution times, continuously performing multi-objective optimization processing on the currently obtained new elite population.

In the embodiment of the application, the curvature of the pareto frontier is estimated according to the current population, a reference vector used for matching the pareto frontier is generated according to the estimated curvature, and the generated reference vector is gradually matched with the shape of the real pareto frontier by continuously estimating the curvature again and adjusting the reference vector in the evolution process. By the method, the matching degree of the reference vector and the real pareto front surface of the problem to be solved can be effectively improved, and further a multi-objective optimization result is perfected; meanwhile, the self-adaptability of the method can be improved, and the generated reference vector can be self-adaptively matched with pareto frontplanes in various shapes.

In one embodiment, the algorithm framework for generating the first reference vector based on curvature in S103 is as follows.

Algorithm 2. reference vector generation.

2-1 calculation of H, H 'Using equations (5), (6), respectively'

2-2, uniformly dividing the arc line segment with the curvature of p into H parts by the formula (7), and calculating corresponding t₀,t₁,…,t_H}

2-3 according to { t₀,t₁,…,t_HConstructing a second reference vector W by the formulas (8) and (9)₁

2-4:if H≥M then

W＝W₁

2-5:else

2-6, evenly dividing the arc line segment with the curvature of p into H' parts by a formula (7), and calculating corresponding { t₀,t₁,…,t_H}

2-7 according to { t₀,t₁,…,t_HConstructing a third reference vector W by the formulas (8) and (9)₂

2-8 using equation (10) to convert W₂Inwardly scaled 1/2

2-9:W＝W₁∪W₂

2-10:end if

2-11 outputting a first reference vector W

The core idea of the algorithm 2 is to calculate the division number H of each dimension using equation (5) according to the target dimension M and the number of individuals N of the initial population. If H is larger than or equal to M, generating a first reference vector by using a first preset method (such as a single-layer point sampling method); if H < M, a second predetermined method (e.g., a dual-layer sampling method) is used to generate the first reference vector. Referring to fig. 2, a schematic diagram of a hierarchical sampling point provided in the embodiment of the present application is shown. Fig. 2 (a) shows a single sampling point, in which a connecting line between a solid point and a coordinate origin represents a second reference vector generated according to the first predetermined method. Fig. 2 (b) shows a double-layer sampling point, in which a connecting line between the solid point of the outer layer and the origin of coordinates represents a second reference vector, and a connection between the solid point of the inner layer and the origin of coordinates represents a third reference vector generated according to a second preset method. The reason for adopting the layering strategy is that for high-dimensional problems, the number of sampling points in an internal area is sparse and the diversity is difficult to guarantee due to the fact that only single-layer sampling points are used.

Algorithm 2 is described below. S103 specifically comprises the following steps:

s301, calculating the number H of first sampling points according to the target dimension M and the number N of the initial population.

Wherein the target dimensions represent the number of target functions.

Calculating the number H of first sampling points by equation (5):

s302, if the number H of the first sampling points is greater than or equal to the target dimension M, a first reference vector is generated according to a first preset method.

In the first preset method (i.e., the single layer sampling method), skin data (i.e., data on an edge line of the pareto front calculated from the curvature, as shown by a solid point in fig. 2 (a)) is used. Optionally, generating the first reference vector according to a first preset method includes:

sampling data on an arc line segment corresponding to the curvature according to the number of first sampling points to obtain first sampling data; generating a second reference vector according to the first sampling data; the second reference vector is determined as the first reference vector.

Specifically, an arc segment with curvature p is uniformly divided into H parts, and first sampling data { t } is obtained₀,t₁,…,t_HAs component values in the dimensions of the second reference vector. Wherein, t₀＝0，t _H1, the rest t_k(k ∈ {1,2, …, H-1}) satisfies the following formula:

a second reference vector is then generated from the first sampled data using equation (8):

wherein u is_iRepresents the ith second reference vector,

represents u_iComponent values in the j-th dimension and satisfying the following formula:

the generated second reference vector is determined as the first reference vector.

And S303, if the number H of the first sampling points is less than the target dimension M, generating a first reference vector according to a second preset method.

In the second preset method (i.e., the double layer sampling method), the outer layer data (i.e., data on the edge line of the pareto front calculated from the curvature, such as the solid point of the outer layer shown in (b) of fig. 2) and the inner layer data (i.e., data on the non-edge line of the pareto front calculated from the curvature, such as the solid point of the inner layer shown in (b) of fig. 2) are used. Optionally, generating the first reference vector according to a second preset method includes:

calculating the number of second sampling points according to the target dimension, the number of individuals of the initial population and the number of the first sampling points; sampling data on the arc line segment corresponding to the curvature according to the number of second sampling points to obtain second sampling data; generating a third reference vector according to the second sampling data; a first reference vector is determined from the second reference vector and the third reference vector.

For the high-dimensional problem of double-layer sampling points, further uniformly dividing each arc line segment of the inner layer into H' parts according to a formula (7), and obtaining { t₀,t₁,…,t_H’Constructing a reference vector W by the formulas (8) and (9)₂. Because the outer layer sampling points of the high-dimensional problem are dense in edges and sparse in the interior, in order to improve the overall diversity of the reference vector, the W can be combined₂The inner scale 1/2 is used to supplement the number of inner layer reference vectors. Scaling formulaThe following were used:

finally, the second reference vector W is added₁And a scaled third reference vector W₂And combining to obtain the first reference vector of the high-dimensional problem.

Exemplarily, refer to fig. 3, which is a schematic diagram of a reference vector generation process provided in an embodiment of the present application. Fig. 3 visually illustrates a process of processing a three-dimensional problem by using the reference vector generation method provided in the embodiment of the present application, where the initial population size N is set to 28, and the estimated curvature p is set to 2. From equation (5), H is known to be 6. Due to H>And M, generating all reference vectors by using single-layer sampling points. FIG. 3 (a) is a diagram of a uniformly sliced arc segment, where { t is obtained from equation (7)₀,t₁,t₂,t₃,t₄,t₅,t₆Corresponding to {0,0.2588,0.5000,0.7071,08660,0.9659,1}, respectively. Fig. 3 (b) shows a process of constructing reference vector coordinates (0,0,1), (0,0.2588,0.9659), …, (0.9659,0.2588,0), (1,0,0) according to formula (9) where u is¹、u²、u³The components of the reference vector in the first, second and third dimensions, respectively. Fig. 3 (c) shows the finally generated reference vectors, which are uniformly distributed on the curved surface corresponding to the estimated curvature.

In the embodiment of the application, the actual PF with different curvatures is used as the f with different curvatures based on the estimated curvature p generated in a self-adaptive mode₁ ^p+f₂ ^p+...+f_M ^pThe method comprises the steps of fitting a curved surface as 1, and generating uniformly distributed reference vectors with good diversity based on the curved surface, so that the method can better support the processing of various PF optimization problems.

In one embodiment, an algorithm framework for performing the environment selection process on the initial population according to the curvature and the first reference vector in S104 is as follows.

Algorithm 3. Environment selection

3-1 finding corner solutions from the cross-mutation population R and putting them into S

3-2 normalization of cross variation population R

3-3, eliminating the reference vector without non-dominant disassociation in the first reference vector W to obtain W₃

3-4 according to curvature p and W₃Selecting candidate population S 'from cross variation population R'

3-5:S＝S∪S’，R＝R\S

3-6:if|S|>N then

3-7, eliminating | S | -N solutions with maximum local density in S

3-8:P＝S

3-9:end if

3-10:if|S|<N then

3-11, determining an elite population P according to the curvature P, the current candidate population S and the current cross variation population R

3-12:end if

3-13 outputting Elite population P

The core of the algorithm 3 is to select an elite population by adaptively selecting an aggregation function based on the estimated curvature and the generated reference vector, and to better support various irregular PF optimization problems by dynamically adjusting the reference vector.

Algorithm 3 is described below. S104 specifically comprises the following steps:

s401, acquiring a cross variation population of the initial population.

The cross variation population comprises a parent population and a child population of the initial population.

The method of determining the parents and the offspring of the initial population may employ existing genetic algorithms. Alternatively, a binary tournament method may be used to select a parent population of the initial population, and a simulated binary crossing method and a polynomial mutation method may be used to generate a child population of the initial population. And then taking the union of the parent population and the child population as the cross variation population of the initial population.

S402, selecting a candidate population S from the cross variation population R according to the curvature p and the first reference vector W.

This step corresponds to steps 3-1 to 3-5 in algorithm 3. Specifically, the method comprises the following steps:

finding out a corner solution from the variant cross population R and putting the corner solution into S; normalizing the variant cross population R; then, a reference vector without non-dominant disassociation in the first reference vector W is provided, and a fourth reference vector W after screening is obtained₃(ii) a According to the curvatures p and W₃Selecting a candidate population S' from the R; and finally updating S (S ═ S ^ S ') according to S', and updating R (R ═ R \ S) according to the updated S.

Since the corner solutions have large coverage space and are far apart from each other, in order to ensure the diversity of the whole population, the corner solutions in the population are first used as default elite solutions. Wherein, according to the curvatures p and W₃The algorithm framework for selecting the candidate population S' from R is shown below.

Algorithm 4. curvatures p and W₃Selecting candidate population S 'from R'

1:S’＝[]

2:for w∈Wdo

3:if p≥1then

4, calculating PBI aggregation function values of each non-dominant solution and reference vector w in R

5:else

6, calculating TCH aggregation function values of each non-dominated solution and the reference vector w in the R

7:end if

Finding the solution s with the optimal aggregation function value

9:S’＝S’∪s

10:end for

11 output candidate population S'

As shown in the above algorithm framework, for each of the fourth reference vectors w: if the curvature p is within a preset numerical range (for example, p is more than or equal to 1), calculating an aggregation function value between each non-dominated solution in the cross variation population R and a fourth reference vector w based on a first aggregation function (for example, a boundary penalty aggregation function PBI); if the curvature p is not within a preset range of values (e.g., p <1), calculating an aggregation function value between each non-dominant solution in the cross-variant population R and the fourth reference vector w based on a second aggregation function (e.g., chebyshev aggregation function TCH); and adding the non-dominant solution in the cross variation population R corresponding to the minimum value in the aggregation function values to the candidate population S'.

In algorithm 4, for each valid reference vector (i.e. the fourth reference vector w), a candidate solution selected as the environment is selected from the non-dominant solutions by the optimal aggregation function value. For the optimization problem that the PF shape is non-convex, the PBI can provide better diversity while ensuring the convergence of the population, and for the optimization problem that the PF shape is convex, the TCH can better balance the convergence and diversity of the population. Therefore, the aggregation function is self-adaptively selected through the estimated curvature, so that the better elite solution can be selected according to the effective reference vector when various PF shapes are fitted to the curved surface, the preference of only using a single aggregation function is avoided, and the convergence and diversity of the population are effectively balanced.

And S403, if the number of the individuals of the candidate population S is greater than the number N of the individuals of the initial population, selecting an elite population P meeting a second preset condition from the candidate population S.

This step corresponds to steps 3-7 and 3-8 in algorithm 3.

Optionally, the first N individuals with the highest local density may be selected from the candidate population S, and the N individuals may be grouped into the elite population P. And N is the number of individuals of the initial population.

S404, if the number of the candidate population S is less than or equal to the number N of the initial population, determining the elite population P according to the candidate population S and the curvature P.

This step corresponds to steps 3-11 in algorithm 3.

Optionally, an algorithm framework for the elite population P is determined in S404 based on the candidate population S and the curvature P, as shown below.

Algorithm 5. determining the final elite population according to the curvature and the current candidate population

5-1:while|S|<N do

5-2, calculating the minimum distance between each non-dominant solution in the R and all solutions in the S

5-3, finding out the corresponding solution l of the maximum value in the minimum distance in R

5-4, selecting a candidate solution u from the non-dominant solution set of R according to the curvature p and the reference vector l

5-5:S＝S∪u,R＝R\u

5-6, finding out the solution c with the worst convergence index in R

5-7:R＝R\c

5-8:end while

5-9:P＝S

5-10, outputting elite group P

As shown in the above algorithm framework, S404 may include the steps of:

if the number of individuals of the candidate population S is smaller than the number of individuals N of the initial population, respectively calculating the minimum distance between each non-dominated solution in the cross variation population R and the candidate population S; acquiring a fifth reference vector l, wherein the fifth reference vector is a non-dominated solution corresponding to the maximum value in the minimum distance in the cross variation population R; screening a target individual u from a non-dominant solution of the cross variation population R according to the curvature p and the fifth reference vector l; and adding the target individual u into the candidate population S, deleting the target individual u in the cross variation population R, and deleting the solution c with the worst convergence in the R. And sequentially circulating until the number of individuals in the candidate population S is equal to N, and determining the candidate population S at the moment as the elite population P.

Wherein, the target individual u can be determined by the method in the algorithm 4 in the step 5-4.

To ensure that the definition of the minimum distance between solutions under different PF shapes is more accurate, the distance between solutions is also adaptively calculated based on the estimated curvature. If PF is fitted into a plane in the curvature estimation process, namely p is 1, the Euclidean distance between two mapping points on the hyperplane is taken as the distance between the solutions; if the PF is fitted into a concave curved surface, namely p is greater than 1, taking an included angle formed between the two solutions and the origin as the distance between the solutions; if the PF is fitted to a convex surface, i.e. p <1, the angle between the two solutions and the range point znad is taken as the distance between the solutions. For any one of the non-dominant solutions in R, the distance between the non-dominant solution and each of the solutions in S is calculated based on the above distance calculation method, and the minimum value of the distances is determined as the minimum distance between the non-dominant solution and S.

The solution l corresponding to the maximum value of the minimum distance is the solution with the best diversity in the non-dominant solutions of the current population R, and therefore the solution l is most suitable to be used as a new reference vector to select an elite solution. In addition, in order to further improve the convergence of the population, at the end of each round of selection, the solution with the worst convergence in the current remaining population is deleted. The component sum method has the advantages of simplicity, high efficiency, strong search performance and the like, so that the component sum can be used as a convergence index of each solution in the population.

Illustratively, referring to fig. 4, a schematic diagram of environment selection provided by the embodiments of the present application is shown. Fig. 4 is an example of environment selection in a two-dimensional space using algorithm 3 of an embodiment of the present application. The population size N was set to 7 and the arc represents the fitted PF curve with a curvature p of 0.5, v₁-v₇For reference vectors generated based on Algorithm 2, x₁₁-x₁₄Is a candidate solution of the population, where x₁-x₇Is a first layer non-dominant solution, x₈-x₁₁Is the second layer non-dominant solution. As can be seen in FIG. 4, x₁And x₇For corner resolution, add directly to elite population. v. of₂And v₆No non-dominant solutions are associated with them, and they are culled from the reference vector. Due to p<Selecting TCH as aggregation function to obtain reference vector v₁、v₃、v₄、v₅And v₇X corresponding one by one and having optimal aggregation function value₁、x₃、x₄、x₆And x₇As a newly obtained elite solution. At this time, the size of the elite population is 5, and the elite solution needs to be continuously selected. In the remaining population, x₂And x₅Is a non-dominant solution. As can be seen from the figure, compare x₂，x₅The minimum distance from each solution in the current elite population is larger, so x is₅As a new reference vector. It is further found that phase ratio x₂Candidate solution x₅The TCH value of the new reference vector is more optimal, x is₅An elite population was added. Then, the selected elite solution x is removed from the population₅And the solution candidate x with the worst convergence₂. Continuing to process the remaining population, x may be found₉It is an elite solution. At this time, the number of elite solutions is equal to that of the initial populationVolume number, environment selection ends.

Compared with the prior art, the hyperspectral image reconstruction method provided by the embodiment of the application has the advantages that the generated reference vector has good diversity and can be accurately matched with pareto frontplanes in various shapes; meanwhile, in the environment selection process, the diversity and the convergence of the PF can be well balanced by self-adaptively selecting the aggregation function. In addition, the time complexity of the method is low (max { O (MN)²) O (DN). Therefore, compared with other algorithms, the hyperspectral image reconstruction method has better performance.

It should be understood that, the sequence numbers of the steps in the foregoing embodiments do not imply an execution sequence, and the execution sequence of each process should be determined by its function and inherent logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

Corresponding to the hyperspectral image reconstruction method described in the above embodiment, fig. 5 is a structural block diagram of the hyperspectral image reconstruction device provided in the embodiment of the application, and for convenience of description, only the parts related to the embodiment of the application are shown.

Referring to fig. 5, the apparatus includes:

and the modeling unit 51 is used for establishing an objective function in the decision space according to the objective variable of the hyperspectral image reconstruction task.

A population obtaining unit 52, configured to obtain an initial population, where each individual in the initial population represents a set of initial solutions of a decision variable of the hyperspectral image reconstruction task.

A curvature estimation unit 53 for estimating a curvature of the pareto front surface from the initial population.

A vector generating unit 54 for generating a first reference vector for matching the pareto front surface in dependence on the curvature.

An environment selecting unit 55, configured to perform environment selection processing on the initial population according to the curvature and the first reference vector, to obtain an elite population, where individuals in the elite population represent a group of pareto optimal solutions of the decision variables of the hyperspectral image reconstruction task.

Optionally, the vector generating unit 54 is further configured to:

calculating the number of first sampling points according to a target dimension and the number of individuals of the initial population, wherein the target dimension represents the number of target functions; if the number of the first sampling points is greater than or equal to the target dimension, generating the first reference vector according to a first preset method, wherein the first preset method is a method for generating a vector according to outer layer data, and the outer layer data is data on an edge line of the pareto frontier calculated according to the curvature; and if the number of the first sampling points is smaller than the target dimension, generating the first reference vector according to a second preset method, wherein the second preset method is a method for generating vectors according to outer layer data and inner layer data, and the inner layer data is data on a non-edge line of the pareto front surface calculated according to the curvature.

Optionally, the vector generating unit 54 is further configured to:

sampling data on the arc line segment corresponding to the curvature according to the number of the first sampling points to obtain first sampling data; generating a second reference vector according to the first sampling data; determining the second reference vector as the first reference vector.

Optionally, the vector generating unit 54 is further configured to:

calculating the number of second sampling points according to the target dimension, the number of individuals of the initial population and the number of the first sampling points; sampling data on the arc line segment corresponding to the curvature according to the number of the second sampling points to obtain second sampling data; generating a third reference vector according to the second sampling data; determining the first reference vector from the second reference vector and the third reference vector.

Optionally, the environment selecting unit 55 is further configured to:

acquiring a cross variation population of the initial population, wherein the cross variation population comprises a parent population and a child population of the initial population; selecting a candidate population from the cross variation population according to the curvature and the first reference vector; if the number of individuals of the candidate population is greater than that of the cross variation population, selecting the elite population meeting a second preset condition from the candidate population; and if the number of individuals of the candidate population is less than or equal to the number of individuals of the cross variation population, determining the elite population according to the candidate population and the curvature.

Optionally, the environment selecting unit 55 is further configured to:

screening out a fourth reference vector from the first reference vector, wherein the fourth reference vector is related to non-dominant solution in the cross variation population; for each of the fourth reference vectors, if the curvature is within a preset numerical range, calculating an aggregation function value between each non-dominant solution in the cross variant population and the fourth reference vector based on a first aggregation function; if the curvature is not within a preset numerical range, calculating an aggregation function value between each non-dominant solution in the cross variant population and the fourth reference vector based on a second aggregation function; and adding a non-dominant solution in the cross variation population corresponding to the minimum value in the aggregation function values to the candidate population.

Optionally, the environment selecting unit 55 is further configured to:

if the number of individuals of the candidate population is smaller than that of the cross variation population, respectively calculating the minimum distance between each non-dominant solution in the cross variation population and the candidate population; obtaining a fifth reference vector, wherein the fifth reference vector is a non-dominated solution corresponding to a maximum value in the minimum distances in the cross variation population; screening out target individuals from a non-dominant solution of the cross variation population according to the curvature and the fifth reference vector; adding the target individuals to the candidate population, and deleting the target individuals in the cross-variant population.

Optionally, the apparatus 5 further comprises:

and the frequency judging unit 56 is used for performing environment selection processing on the initial population according to the curvature and the first reference vector to obtain an elite population, and then continuing performing multi-objective optimization processing on the elite population until a preset evolution frequency is met.

It should be noted that, for the information interaction, execution process, and other contents between the above-mentioned devices/units, the specific functions and technical effects thereof are based on the same concept as those of the embodiment of the method of the present application, and specific reference may be made to the part of the embodiment of the method, which is not described herein again.

In addition, the hyperspectral image reconstruction apparatus shown in fig. 5 may be a software unit, a hardware unit, or a combination of software and hardware unit that is built in the existing terminal device, may be integrated into the terminal device as an independent pendant, or may exist as an independent terminal device.

It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-mentioned division of the functional units and modules is illustrated, and in practical applications, the above-mentioned function distribution may be performed by different functional units and modules according to needs, that is, the internal structure of the apparatus is divided into different functional units or modules to perform all or part of the above-mentioned functions. Each functional unit and module in the embodiments may be integrated in one processing unit, or each unit may exist alone physically, or two or more units are integrated in one unit, and the integrated unit may be implemented in a form of hardware, or in a form of software functional unit. In addition, specific names of the functional units and modules are only for convenience of distinguishing from each other, and are not used for limiting the protection scope of the present application. The specific working processes of the units and modules in the system may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.

Fig. 6 is a schematic structural diagram of a terminal device according to an embodiment of the present application. As shown in fig. 6, the terminal device 6 of this embodiment includes: at least one processor 60 (only one shown in fig. 6), a memory 61, and a computer program 62 stored in the memory 61 and executable on the at least one processor 60, the processor 60 implementing the steps in any of the various method embodiments described above when executing the computer program 62.

The terminal device can be a desktop computer, a notebook, a palm computer, a cloud server and other computing devices. The terminal device may include, but is not limited to, a processor, a memory. Those skilled in the art will appreciate that fig. 6 is only an example of the terminal device 6, and does not constitute a limitation to the terminal device 6, and may include more or less components than those shown, or combine some components, or different components, such as an input/output device, a network access device, and the like.

The Processor 60 may be a Central Processing Unit (CPU), and the Processor 60 may be other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic device, discrete hardware component, etc. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.

The memory 61 may in some embodiments be an internal storage unit of the terminal device 6, such as a hard disk or a memory of the terminal device 6. The memory 61 may also be an external storage device of the terminal device 6 in other embodiments, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), and the like, which are equipped on the terminal device 6. Further, the memory 61 may also include both an internal storage unit and an external storage device of the terminal device 6. The memory 61 is used for storing an operating system, an application program, a BootLoader (BootLoader), data, and other programs, such as program codes of the computer program. The memory 61 may also be used to temporarily store data that has been output or is to be output.

The embodiments of the present application further provide a computer-readable storage medium, where a computer program is stored, and when the computer program is executed by a processor, the computer program implements the steps in the above-mentioned method embodiments.

The embodiments of the present application provide a computer program product, which when running on a terminal device, enables the terminal device to implement the steps in the above method embodiments when executed.

The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, all or part of the processes in the methods of the embodiments described above can be implemented by a computer program, which can be stored in a computer-readable storage medium and can implement the steps of the embodiments of the methods described above when the computer program is executed by a processor. Wherein the computer program comprises computer program code, which may be in the form of source code, object code, an executable file or some intermediate form, etc. The computer readable medium may include at least: any entity or device capable of carrying computer program code to an apparatus/terminal device, recording medium, computer Memory, Read-Only Memory (ROM), Random-Access Memory (RAM), electrical carrier wave signals, telecommunications signals, and software distribution medium. Such as a usb-disk, a removable hard disk, a magnetic or optical disk, etc. In certain jurisdictions, computer-readable media may not be an electrical carrier signal or a telecommunications signal in accordance with legislative and patent practice.

In the above embodiments, the descriptions of the respective embodiments have respective emphasis, and reference may be made to the related descriptions of other embodiments for parts that are not described or illustrated in a certain embodiment.

Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.

In the embodiments provided in the present application, it should be understood that the disclosed apparatus/terminal device and method may be implemented in other ways. For example, the above-described embodiments of the apparatus/terminal device are merely illustrative, and for example, the division of the modules or units is only one logical division, and there may be other divisions when actually implemented, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.

The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.

The above-mentioned embodiments are only used for illustrating the technical solutions of the present application, and not for limiting the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; such modifications and substitutions do not substantially depart from the spirit and scope of the embodiments of the present application and are intended to be included within the scope of the present application.

Claims

1. A hyperspectral image reconstruction method is characterized by comprising the following steps:

2. The hyperspectral image reconstruction method of claim 1, wherein the generating a first reference vector for matching the pareto frontplane according to the curvature comprises:

3. The hyperspectral image reconstruction method of claim 2, wherein the generating the first reference vector according to a first preset method comprises:

generating a second reference vector according to the first sampling data;

determining the second reference vector as the first reference vector.

4. The hyperspectral image reconstruction method of claim 3, wherein the generating the first reference vector according to a second preset method comprises:

generating a third reference vector according to the second sampling data;

5. The hyperspectral image reconstruction method according to any of claims 1 to 4, wherein the performing environment selection processing on the initial population according to the curvature and the first reference vector to obtain an elite population comprises:

6. The hyperspectral image reconstruction method of claim 5, wherein the selecting a candidate population from the cross variation population according to the curvature and the first reference vector comprises:

7. The hyperspectral image reconstruction method of claim 5, wherein determining the elite population according to the candidate population and the curvature if the number of individuals of the candidate population is less than or equal to the number of individuals of the cross variation population comprises:

8. The hyperspectral image reconstruction method according to claim 1, wherein after performing environment selection processing on the initial population according to the curvature and the first reference vector to obtain an elite population, the method comprises:

9. A terminal device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the method according to any of claims 1 to 8 when executing the computer program.

10. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the method according to any one of claims 1 to 8.