Seed spectral data segmentation preprocessing method based on large language model

Abstract

The invention discloses a seed spectral data segmentation preprocessing method based on a large language model, and relates to the technical field of spectral data processing. According to the seed spectral data segmentation preprocessing method based on the large language model, by establishing a seed spectral image segmentation model, the optimization effect of a spectral data image is effectively improved, and by establishing a spectral data segmentation standard model, the optimization effect of the spectral data image is effectively improved. The difference between the spectral data and the standard model in different waveband subintervals is analyzed, the seed type corresponding to the standard sensitive waveband model can be rapidly matched, the analysis of different waveband intervals of the spectral data is matched, and the problem that data processing needs to be performed again due to the fact that identification cannot be matched when impurities exist on the surfaces of particles can be effectively solved. The data processing time can be effectively saved, and the preprocessing effect of the seed spectrum data is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of spectral data processing, and in particular to a seed spectral data segmentation preprocessing method based on a large language model. Background Art

[0002] Referring to a Chinese patent, a method for preprocessing near-infrared diffuse reflectance spectrum data of corn variety seeds (Announcement No.: CN101788475A, Announcement Date: 2010-07-28) This patent uses a method for preprocessing near-infrared diffuse reflectance spectrum data of corn variety seeds, only intercepting the first 1 / 3 of the data as valid data, performing DFT transformation on the selected valid data, and selecting the amplitude part of the transformed data, ignoring the phase angle part of the data. The present invention can effectively reduce data processing and storage, and the feature data points are more concentrated than the original data, with a greater degree of difference, which is conducive to subsequent classification and identification of corn seeds.

[0003] However, the existing seed spectral data segmentation preprocessing method based on the large language model still has other defects, for example: the acquired spectral data is not bright enough, resulting in blurring of the detail information of the spectral data image, and when the brightness is enhanced, the spectral data image will be distorted. There is a lack of a standard model for spectral data. When matching seed spectral data, the processing time is long and the matching efficiency is relatively low. When there are impurities on the surface of crop seed particles, the identified spectral data is inaccurate, and the matching requires re-processing of the data, which increases the data processing time and reduces the preprocessing effect of the seed spectral data. To this end, the present invention proposes a seed spectral data segmentation preprocessing method based on a large language model. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the deficiencies in the prior art, the present invention provides a seed spectrum data segmentation preprocessing method based on a large language model, which solves the problems mentioned in the above background technology.

[0006] (II) Technical solution

[0007] To achieve the above objectives, the present invention is implemented by the following technical scheme: a seed spectrum data segmentation preprocessing method based on a large language model, specifically comprising the following steps:

[0008] Step 1: Acquire target data of the basic object, establish a seed spectral image segmentation model, and perform segmentation comparison and weighted analysis on the target data and the seed spectral image model to analyze whether the target data is a pathological image, wherein the basic object is specifically crop seeds, and the target data is specifically spectral data;

[0009] Step 2: Based on the acquired target data, several regions of the target data are intercepted to establish a segmented model of the spectral data, analyze the differences between the spectral data of the several regions and the standard model, and match the seed lesion type corresponding to the standard model, wherein the standard model is a data model of a single or multiple seed types;

[0010] Step three: When the target data does not match the standard model, the collected target data is analyzed, the direction of the target data is calculated, and the target data is corrected, wherein the direction of the target data is specifically the direction of the regional data.

[0011] As an improved technical solution, the specific method of analyzing whether the target data is a lesion image and performing segmented analysis on the spectral image in step 1 is:

[0012] S1: Extract from the segmented features and overall features of the image. The specific formula is as follows:

[0013]

[0014] In the formula, S represents the description of seed poisoning, and B represents the image Base The similarity of Y n represents the weight of a feature segment, F(B n ) represents the similarity of a certain feature segment, n represents the segment sequence where the image is located, the image mean S is the correlation of a certain seed with virus, and the model builds a correlation model between S and the image;

[0015] S2: The specific expression for correlation weighted identification analysis of infected images is:

[0016] RTN_D θ ∝M(x,y)

[0017] Where, RTN_D θ represents the segmented weighted image similarity matrix of the target, M(x, y) represents the spectral segmentation model similarity matrix of the system training model, and the similarity between the target seed and the poisonous image in the model is obtained through the analysis of the segmented model.

[0018] As an improved technical solution, RTN_D in S2 θ It represents a similarity matrix model based on a multimodal large model, obtains the overall similarity and segmental difference between the target spectrum and the model, performs weighted processing, and thus obtains the final spectral difference similarity analysis result.

[0019] As an improved technical solution, in step 2, the differences between the spectral data of the plurality of regions and the standard model are analyzed, and the specific method of matching the seed type corresponding to the standard model is as follows:

[0020] M1: First, the spectral data is divided into several regions j, and then n regions are selected from the several regions j. At this time, the standard model Hj is set according to the type of seed. In the standard model Hj, the spectral data of different seed types are matched accordingly;

[0021] M2: At this time, the spectral data of n regions are compared with the standard model Hj. If all n regions are consistent with the standard model Hj, the spectral data matches the standard model Hj and the type of seed poisoning is determined. If any region is inconsistent with the standard model Hj, the standard model Hj cannot be matched and the type of seed poisoning is to be determined.

[0022] As an improved technical solution, if any region in M2 is inconsistent with the standard model Hj, the standard model Hj cannot be matched. When the seed type is pending, the spectral data of the changed type of seed is reacquired, and the selected n regions are adjusted until all n regions are consistent with the standard model Hj.

[0023] As an improved technical solution, in step three, the collected target data is analyzed to calculate the trend of the target data.

[0024] (III) Beneficial effects

[0025] The present invention provides a seed spectrum data segmentation preprocessing method based on a large language model. Compared with the prior art, it has the following beneficial effects:

[0026] The seed spectral data segmentation preprocessing method based on the large language model can retain more image detail information, reduce the degree of spectral data image distortion, and effectively improve the spectral data image optimization effect by establishing a seed spectral image model and enhancing the seed spectral image. By establishing a standard model of spectral data and analyzing the difference between the spectral data and the standard model, the seed type corresponding to the standard model can be quickly matched. By correcting the spectral data, the problem of identification failure and need for data reprocessing when there are impurities on the particle surface can be effectively avoided. This can effectively save data processing time and improve the preprocessing effect of seed spectral data. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a flow chart of a seed spectrum data segmentation preprocessing method based on a large language model according to an embodiment of the present application. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] See also Figure 1 , the embodiment of the present invention provides three technical solutions:

[0030] Embodiment 1

[0031] A seed spectrum data segmentation preprocessing method based on a large language model specifically comprises the following steps:

[0032] Step 1: Obtain target data of the basic object, establish a seed spectral image segmentation model, and perform segmentation comparison and weighted analysis on the target data and the seed spectral image model to analyze whether the target data is a pathological image, wherein the basic object is specifically crop seeds, and the target data is specifically spectral data; the specific method of analyzing whether the target data is a pathological image and performing segmentation analysis on the spectral image in step 1 is:

[0033] S1: Extract from the segmented features and overall features of the image. The specific formula is as follows:

[0034]

[0035] In the formula, S represents the description of seed poisoning, and B represents the image Base The similarity of Y n represents the weight of a feature segment, F(B n ) represents the similarity of a certain feature segment, n represents the segment sequence where the image is located, the image mean S is the correlation of a certain seed with virus, and the model builds a correlation model between S and the image;

[0036] S2: The specific expression for correlation weighted identification analysis of infected images is:

[0037] RTN_D θ ∝M(x,y)

[0038] Where, RTN_D θ represents the segmented weighted image similarity matrix of the target, M(x, y) represents the spectral segmentation model similarity matrix of the system training model, and the similarity between the target seed and the poisonous image in the model is obtained through the analysis of the segmented model.

[0039] S2RTN_D θIt represents a similarity matrix model based on a multimodal large model, obtains the overall similarity and segmental difference between the target spectrum and the model, performs weighted processing, and thus obtains the final spectral difference similarity analysis result;

[0040] Step 2: Based on the acquired target data, several regions of the target data are intercepted, a segmented model of the spectral data is established, the differences between the spectral data of several regions and the standard model are analyzed, and the seed lesion type corresponding to the standard model is matched, wherein the standard model is a data model of a single or multiple seed types; the specific method of analyzing the differences between the spectral data of several regions and the standard model in step 2 and matching the seed type corresponding to the standard model is:

[0041] M1: First, the spectral data is divided into several regions j, and then n regions are selected from the several regions j. At this time, the standard model Hj is set according to the type of seed. In the standard model Hj, the spectral data of different seed types are matched accordingly;

[0042] M2: At this time, the spectral data of n regions are compared with the standard model Hj. If all n regions are consistent with the standard model Hj, the spectral data matches the standard model Hj and the seed poisoning type is determined. If any region is inconsistent with the standard model Hj, the standard model Hj cannot be matched and the seed poisoning type is to be determined. If any region in M2 is inconsistent with the standard model Hj, the standard model Hj cannot be matched and the seed type is to be determined. The spectral data of the changed type of seed is reacquired and the selected n regions are adjusted until all n regions are consistent with the standard model Hj;

[0043] Step three: When the target data does not match the standard model, the collected target data is analyzed, the direction of the target data is calculated, and the target data is corrected, wherein the direction of the target data is specifically the direction of the regional data; in the step three, the collected target data is analyzed and the direction of the target data is calculated.

[0044] Embodiment 2

[0045] A seed spectrum data segmentation preprocessing method based on a large language model specifically comprises the following steps:

[0046] Step 1: Obtain target data of the basic object, establish a seed spectral image segmentation model, and perform segmentation comparison and weighted analysis on the target data and the seed spectral image model to analyze whether the target data is a pathological image, wherein the basic object is specifically crop seeds, and the target data is specifically spectral data; the specific method of analyzing whether the target data is a pathological image and performing segmentation analysis on the spectral image in step 1 is:

[0047] S1: Extract from the segmented features and overall features of the image. The specific formula is as follows:

[0048]

[0049] In the formula, S represents the description of seed poisoning, and B represents the image Base The similarity of Y n represents the weight of a feature segment, F(B n ) represents the similarity of a certain feature segment, n represents the segment sequence where the image is located, the image mean S is the correlation of a certain seed with virus, and the model builds a correlation model between S and the image;

[0050] S2: The specific expression for correlation weighted identification analysis of infected images is:

[0051] RTN_D θ ∝M(x,y)

[0052] Where, RTN_D θ represents the segmented weighted image similarity matrix of the target, M(x, y) represents the spectral segmentation model similarity matrix of the system training model, and the similarity between the target seed and the poisonous image in the model is obtained through the analysis of the segmented model.

[0053] S2RTN_D θ It represents a similarity matrix model based on a multimodal large model, obtains the overall similarity and segmental difference between the target spectrum and the model, performs weighted processing, and thus obtains the final spectral difference similarity analysis result;

[0054] Step 2: Based on the acquired target data, several regions of the target data are intercepted, a segmented model of the spectral data is established, the differences between the spectral data of several regions and the standard model are analyzed, and the seed lesion type corresponding to the standard model is matched, wherein the standard model is a data model of a single or multiple seed types; the specific method of analyzing the differences between the spectral data of several regions and the standard model in step 2 and matching the seed type corresponding to the standard model is:

[0055] M1: First, the spectral data is divided into several regions j, and then n regions are selected from the several regions j. At this time, the standard model Hj is set according to the type of seed. In the standard model Hj, the spectral data of different seed types are matched accordingly;

[0056] M2: At this time, the spectral data of n regions are compared with the standard model Hj. If all n regions are consistent with the standard model Hj, the spectral data matches the standard model Hj and the seed poisoning type is determined. If any region is inconsistent with the standard model Hj, the standard model Hj cannot be matched and the seed poisoning type is to be determined. If any region in M2 is inconsistent with the standard model Hj, the standard model Hj cannot be matched and the seed type is to be determined. The spectral data of the seed of the changed type is reacquired and the selected n regions are adjusted until all n regions are consistent with the standard model Hj.

[0057] Embodiment 3

[0058] A seed spectrum data segmentation preprocessing method based on a large language model specifically comprises the following steps:

[0059] Step 1: Obtain target data of the basic object, establish a seed spectral image model, compare the target data with the seed spectral image model, analyze whether the target data is a seed spectral image, and enhance the seed spectral image, wherein the basic object is specifically crop seeds and the target data is specifically spectral data;

[0060] Step 2: Based on the acquired target data, several regions of the target data are intercepted to establish a standard model of the spectral data, analyze the differences between the spectral data of several regions and the standard model, and match the seed type corresponding to the standard model, wherein the standard model is a data model of multiple seed types;

[0061] Step 3: When the target data does not match the standard model, the collected target data is analyzed, the direction of the target data is calculated, and the target data is corrected, wherein the direction of the target data is specifically the direction of the regional data.

[0062] Meanwhile, the contents not described in detail in this specification belong to the prior art known to those skilled in the art.

[0063] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0064] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0065] The embodiments of the present application have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A seed spectrum data segmentation preprocessing method based on a large language model, characterized in that: The specific steps include: Step 1: Acquire target data of the basic object, establish a seed spectral image segmentation model, and perform segmentation comparison and weighted analysis on the target data and the seed spectral image model to analyze whether the target data is a pathological image, wherein the basic object is specifically crop seeds and the target data is specifically spectral data; Step 2: Based on the acquired target data, several regions of the target data are intercepted to establish a segmented model of the spectral data, analyze the differences between the spectral data of the several regions and the standard model, and match the seed lesion type corresponding to the standard model, wherein the standard model is a data model of a single or multiple seed types; Step three: When the target data does not match the standard model, the collected target data is analyzed, the direction of the target data is calculated, and the target data is corrected, wherein the direction of the target data is specifically the direction of the regional data.

2. The seed spectrum data segmentation preprocessing method based on a large language model according to claim 1, characterized in that: The specific method of analyzing whether the target data is a lesion image and performing segmented analysis on the spectral image in step 1 is: S1: Extract from the segmented features and overall features of the image. The specific formula is as follows: In the formula, S represents the description of seed poisoning, and B represents the image Base The similarity of Y n represents the weight of a feature segment, F(B n ) represents the similarity of a certain feature segment, n represents the segment sequence where the image is located, the image mean S is the correlation of a certain seed with virus, and the model builds a correlation model between S and the image; S2: The specific expression for correlation weighted identification analysis of infected images is: RTN_D θ ∝M(x,y) Where, RTN_D θ represents the segmented weighted image similarity matrix of the target, M(x, y) represents the spectral segmentation model similarity matrix of the system training model, and the similarity between the target seed and the poisonous image in the model is obtained through the analysis of the segmented model.

3. The seed spectrum data segmentation preprocessing method based on a large language model according to claim 1, characterized in that: The S2 RTN_D θ It represents a similarity matrix model based on a multimodal large model, obtains the overall similarity and segmental difference between the target spectrum and the model, performs weighted processing, and thus obtains the final spectral difference similarity analysis result.

4. The seed spectrum data segmentation preprocessing method based on a large language model according to claim 1, characterized in that: In step 2, the specific method of analyzing the differences between the spectral data of the plurality of regions and the standard model and matching the seed type corresponding to the standard model is as follows: M1: First, the spectral data is divided into several regions j, and then n regions are selected from the several regions j. At this time, the standard model Hj is set according to the type of seed. In the standard model Hj, the spectral data of different seed types are matched accordingly; M2: At this time, the spectral data of n regions are compared with the standard model Hj. If all n regions are consistent with the standard model Hj, the spectral data matches the standard model Hj and the type of seed poisoning is determined. If any region is inconsistent with the standard model Hj, the standard model Hj cannot be matched and the type of seed poisoning is to be determined.

5. The seed spectrum data segmentation preprocessing method based on a large language model according to claim 4, characterized in that: If any region in M2 is inconsistent with the standard model Hj, the standard model Hj cannot be matched. When the seed type is to be determined, the spectral data of the changed type of seed is reacquired and the selected n regions are adjusted until all n regions are consistent with the standard model Hj.

6. The seed spectrum data segmentation preprocessing method based on a large language model according to claim 1, characterized in that: In the step three, the collected target data is analyzed and the trend of the target data is calculated.

Images