Automatic identification system and method for cultivated land monitoring
By performing grayscale processing and image segmentation on the original image, combined with preset neural network algorithm and scale invariant feature transformation algorithm, the automatic identification system solves the problems of low efficiency and poor accuracy in traditional methods, achieving high efficiency and accuracy plant identification.
Patent Information
- Application Number
- CN202210118576.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-02-08
- Publication Date
- 2025-02-14
- Estimated Expiration
- 2042-02-08
AI Technical Summary
The traditional scale-invariant feature transformation model requires multiple interventions in the process of image feature vector extraction, resulting in the feature vectors not reflecting the image characteristics, which are inefficient, workloads and human errors.
An automatic identification system for farmland monitoring was designed. The original image is grayscale and image segmented through the data building module, training samples are obtained, and the corresponding between plant names and leaf feature vectors is determined using a preset neural network algorithm. relation. Then, through the target acquisition module and the image processing module, the target feature vector of the target picture is obtained, and the plant name is determined based on the feature vector and the sample database.
It improves processing efficiency, reduces artificial intervention, reduces range overlap of feature vectors, enhances the accuracy of recognition, and avoids interference from background image data.
Smart Images
Figure CN114821291B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of automatic identification, and in particular to an automatic identification system and method for farmland monitoring. Background Art
[0002] With the development of science and technology, the scale-invariant feature transformation model has been widely used in the field of recognition due to its high tolerance to changes in light, noise, and micro-perspective.
[0003] At present, the technical solution for image recognition using a scale-invariant feature transformation model is mainly to compare the feature vector in the image to be tested with the pre-stored feature vector to determine the plant species corresponding to the image.
[0004] However, the traditional scale-invariant feature transformation model requires multiple interventions in the process of extracting feature vectors from images. The proposed feature vectors cannot reflect the characteristics of the image, and the feature vector ranges of various recognition objects overlap a lot, resulting in low efficiency, heavy workload and human errors. Summary of the invention
[0005] The embodiments of the present invention provide an automatic identification system and method for farmland monitoring to solve the above problems.
[0006] In a first aspect, an embodiment of the present invention provides an automatic identification system for arable land monitoring, the system comprising: a data building module, used to use a number of leaf images marked with plant names as training samples; based on the training samples and a preset neural network algorithm, determining the correspondence between the plant name and the leaf feature vector to update the preset sample database; a target acquisition module, used to obtain a collection image uploaded by a preset collection device, and based on a maximum inter-class variance algorithm, obtain a target image corresponding to the plant leaves in the collection image; an image processing module, used to import the target image into a preset scale-invariant feature transformation algorithm to obtain a target feature vector corresponding to the target image; and then based on the target feature vector and the sample database, determining the plant name corresponding to the collection image.
[0007] Furthermore, the data construction module includes: a sample acquisition unit, a sample processing unit and a database update unit; the sample acquisition unit is used to acquire a number of original images marked with plant names; the sample processing unit is used to perform grayscale processing and image segmentation on the original images to obtain leaf images as training samples; the training samples are imported into a preset neural network algorithm to obtain leaf feature vectors corresponding to the training samples; the plant name is used as the clustering center to cluster the leaf feature vectors through a proximity algorithm to remove invalid feature vectors to obtain the correspondence between the leaf feature vectors and the plant name; the database update unit is used to update the correspondence between the leaf feature vectors and the plant name to a preset sample database.
[0008] Furthermore, the image processing module includes: a feature transformation unit; a feature transformation unit, used to import the target image into a scale-invariant feature transformation algorithm to obtain key points in the target image through a Gaussian function preset in the algorithm; determine the Gaussian weight function parameters of the histogram corresponding to the key points; determine the Taylor expansion of the histogram to perform quadratic curve fitting to determine the direction corresponding to the key points; determine the key point descriptor corresponding to the key point through scale-invariant feature transformation; determine the target feature vector corresponding to the target image based on the key points, key point directions and key point descriptors.
[0009] Furthermore, the system also includes a secondary verification module; the secondary verification module includes a standard picture unit and a MATLAB verification unit; the standard picture unit is used to store standard pictures corresponding to plant names; the MATLAB correction unit is used to determine the similarity between the collected picture and the standard picture based on MATLAB software and the plant name after determining the plant name corresponding to the collected picture; when the similarity is less than a preset similarity threshold, an error maintenance instruction is generated to the maintenance terminal.
[0010] In a second aspect, an embodiment of the present invention provides an automatic identification method for cultivated land monitoring, the method comprising: using a number of leaf images marked with plant names as training samples; based on the training samples and a preset neural network algorithm, determining the correspondence between the plant name and the leaf feature vector to update a preset sample database; obtaining a collection picture uploaded by a preset collection device, and based on a maximum inter-class variance algorithm, obtaining a target picture corresponding to the plant leaves in the collection picture; importing the target picture into a preset scale-invariant feature transformation algorithm to obtain a target feature vector corresponding to the target picture; and then determining the plant name corresponding to the collection picture based on the target feature vector and the sample database.
[0011] Furthermore, several leaf images marked with plant names are used as training samples; based on the training samples and a preset neural network algorithm, the correspondence between the plant names and the leaf feature vectors is determined to update the preset sample database, specifically including: obtaining several original images marked with plant names; performing grayscale processing and image segmentation on the original images to obtain leaf images as training samples; importing the training samples into the preset neural network algorithm to obtain leaf feature vectors corresponding to the training samples; using the plant name as the clustering center to cluster the leaf feature vectors through a proximity algorithm to remove invalid feature vectors to obtain the correspondence between the leaf feature vectors and the plant name; and updating the correspondence between the leaf feature vectors and the plant name to the preset sample database.
[0012] Furthermore, the target image is imported into a preset scale-invariant feature transformation algorithm to obtain a target feature vector corresponding to the target image, specifically including: importing the target image into the scale-invariant feature transformation algorithm to obtain key points in the target image through a Gaussian function preset in the algorithm; determining Gaussian weight function parameters for calculating a histogram corresponding to the key points; determining a Taylor expansion of the histogram to perform quadratic curve fitting to determine the direction corresponding to the key points; determining a key point descriptor corresponding to the key point through scale-invariant feature transformation; and determining a target feature vector corresponding to the target image based on the key points, the key point directions and the key point descriptors.
[0013] Furthermore, a preset verification database is used to store standard pictures corresponding to plant names; the method also includes: after determining the plant name corresponding to the collected picture, based on MATLAB software and a preset verification database, determining the similarity between the collected picture and the standard picture; when the similarity is less than a preset similarity threshold, generating an error maintenance instruction to the maintenance terminal.
[0014] It can be understood by those skilled in the art that the present invention has at least the following beneficial effects:
[0015] The original image is gray-scaled and segmented through the data building module to obtain training samples. The amount of original image data is downgraded, which is convenient for less calculation in subsequent processing, thereby improving work efficiency. Through the preset neural network algorithm, the leaf feature vector corresponding to each plant name is determined, so that the plant name corresponding to the leaf corresponding to the image can be directly determined by the verification vector in the future, which improves processing efficiency. Through the target acquisition module, the leaf is deducted from the collected image as the target image, so that the subsequent image to be detected is unique (all leaf data), avoiding interference from other background image data. Through the image processing module, the target image is input into the preset scale-invariant feature transformation algorithm, so that the image processed by the preset scale-invariant feature transformation algorithm is as small as possible, and the processed image is as typical as possible, which improves the accuracy of the preset scale-invariant feature transformation algorithm and reduces the range overlap of the feature vector.
[0016] In addition, the invention has a reliable design principle, a simple structure and a very broad application prospect. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Some embodiments of the present disclosure are described below with reference to the accompanying drawings, in which:
[0018] Figure 1 It is a schematic diagram of the internal structure of the system according to a specific implementation mode of the present invention.
[0019] Figure 2 It is a method flow chart of a specific implementation mode of the present invention. DETAILED DESCRIPTION
[0020] It should be understood by those skilled in the art that the embodiments described below are only preferred embodiments of the present disclosure, and do not mean that the present disclosure can only be implemented through the preferred embodiments. The preferred embodiments are only used to explain the technical principles of the present disclosure, and are not used to limit the protection scope of the present disclosure. Based on the preferred embodiments provided by the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work should still fall within the protection scope of the present disclosure.
[0021] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity 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, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.
[0022] The technical solution proposed in the embodiment of the present invention is described in detail below with reference to the accompanying drawings.
[0023] Figure 1 The automatic identification system for farmland monitoring provided by the embodiment of the present invention. Figure 1 As shown, the automatic recognition system provided by the embodiment of the present invention mainly includes: a data building module 110, a target acquisition module 120 and an image processing module 130.
[0024] Among them, the data construction module 110 is mainly used to use a number of leaf images marked with plant names as training samples; based on the training samples and the preset neural network algorithm, the leaf feature vector corresponding to each plant name is determined to update the preset sample database.
[0025] It should be noted that the method for obtaining a number of leaf images with plant names marked can be obtained by manually uploading data. The preset neural network algorithm can be any feasible algorithm that can learn the correspondence between leaf images and leaf feature vectors from training samples. The sample database is used to store plant names, leaf feature vectors, and the correspondence between plant names and leaf feature vectors.
[0026] As an example, the data construction module 110 includes: a sample acquisition unit 111, a sample processing unit 112 and a database update unit 113. The sample acquisition unit 111 is used to acquire a number of original images with plant names marked; the sample processing unit 112 is used to perform grayscale processing and image segmentation on the original images to obtain leaf images as training samples; import the training samples into a preset neural network algorithm to obtain leaf feature vectors corresponding to the training samples; cluster the leaf feature vectors through a proximity algorithm to remove invalid feature vectors to obtain the corresponding relationship between the leaf feature vectors and the plant names; the database update unit 113 is used to update the corresponding relationship between the leaf feature vectors and the plant names to the preset sample database.
[0027] The target acquisition module 120 is used to acquire the collected pictures uploaded by the preset collection device, and obtain the target pictures corresponding to the plant leaves in the collected pictures based on the maximum inter-class variance algorithm.
[0028] It should be noted that the image acquisition module is any feasible device with an image acquisition function. And the installation position of the image acquisition module can be any feasible position, such as on an aircraft. The image to be tested is a picture of a plant leaf that needs to be inspected for pests and diseases. The pre-processed picture is a picture of the image to be tested after grayscale processing.
[0029] In addition, the specific execution process corresponding to the maximum inter-class variance algorithm can be performed by existing methods or devices, and this embodiment does not impose too many limitations on this.
[0030] The image processing module 130 is used to import the target image into a preset scale-invariant feature transformation algorithm to obtain a target feature vector corresponding to the target image; and then determine the plant name corresponding to the collected image based on the target feature vector and the sample database.
[0031] The image processing module 130 includes: a feature transformation unit 131; the feature transformation unit 131 is used to import the target image into the scale-invariant feature transformation algorithm to obtain the key points in the target image through the Gaussian function preset in the algorithm; determine the Gaussian weight function parameters of the histogram corresponding to the key points; determine the Taylor expansion of the histogram to perform quadratic curve fitting to determine the direction corresponding to the key points; determine the key point descriptor corresponding to the key point through the scale-invariant feature transformation; determine the target feature vector corresponding to the target image according to the key point, the key point direction and the key point descriptor. It should be noted that the specific operation process of the Gaussian function, quadratic curve fitting, etc. can be implemented by existing methods or technologies, and the embodiments of the present application are not limited to this.
[0032] In addition, this implementation can also verify the accuracy of the preset scale-invariant feature transformation algorithm through third-party tools.
[0033] As an example, the system also includes a secondary verification module 140; the secondary verification module 140 includes a standard picture unit and a MATLAB verification unit; the standard picture unit is used to store standard pictures corresponding to plant names; the MATLAB correction unit is used to determine the similarity between the collected picture and the standard picture based on the MATLAB software and the plant name after determining the plant name corresponding to the collected picture; when the similarity is less than a preset similarity threshold, an error maintenance instruction is generated to the maintenance terminal.
[0034] In addition, the embodiment of the present invention also provides an automatic identification method for farmland monitoring. Figure 1 The automatic identification method provided by the embodiment of the present invention mainly includes the following steps:
[0035] Step 210: Use a number of leaf images with plant names marked as training samples; based on the training samples and a preset neural network algorithm, determine the correspondence between the plant name and the leaf feature vector to update the preset sample database.
[0036] Among them, a number of leaf images marked with plant names are used as training samples; based on the training samples and the preset neural network algorithm, the correspondence between the plant name and the leaf feature vector is determined to update the preset sample database, which can be specifically as follows: a number of original images marked with plant names are obtained; the original images are gray-processed and segmented to obtain leaf images as training samples; the training samples are imported into the preset neural network algorithm to obtain the leaf feature vectors corresponding to the training samples; the plant name is used as the clustering center to cluster the leaf feature vectors through the proximity algorithm, and the invalid feature vectors are removed to obtain the correspondence between the leaf feature vector and the plant name; the correspondence between the leaf feature vector and the plant name is updated to the preset sample database.
[0037] Step 220: Obtain a collection image uploaded by a preset collection device, and obtain a target image corresponding to the plant leaves in the collection image based on a maximum inter-class variance algorithm.
[0038] Step 230: import the target image into a preset scale-invariant feature transformation algorithm to obtain a target feature vector corresponding to the target image; and then determine the plant name corresponding to the collected image based on the target feature vector and the sample database.
[0039] Among them, the target image is imported into a preset scale-invariant feature transformation algorithm to obtain a target feature vector corresponding to the target image, which can be specifically as follows: the target image is imported into the scale-invariant feature transformation algorithm to obtain key points in the target image through a Gaussian function preset in the algorithm; the Gaussian weight function parameters of the histogram corresponding to the key points are determined; the Taylor expansion of the histogram is determined to perform quadratic curve fitting to determine the direction corresponding to the key point; the key point descriptor corresponding to the key point is determined through the scale-invariant feature transformation; and the target feature vector corresponding to the target image is determined according to the key point, the key point direction and the key point descriptor.
[0040] In addition, this embodiment can also verify the accuracy of the preset scale-invariant feature transformation algorithm through a third-party tool.
[0041] As an example, there is a preset verification database for storing standard pictures corresponding to plant names; after determining the plant name corresponding to the collected picture, the server determines the similarity between the collected picture and the standard picture based on MATLAB software and the preset verification database; when the similarity is less than the preset similarity threshold, an error maintenance instruction is generated to the maintenance terminal.
[0042] So far, the technical solutions of the present disclosure have been described in combination with the above multiple embodiments, but it is easy for those skilled in the art to understand that the protection scope of the present disclosure is not limited to these specific embodiments. Without departing from the technical principles of the present disclosure, those skilled in the art can split and combine the technical solutions in the above-mentioned various embodiments, and can also make equivalent changes or replacements to the relevant technical features. Any changes, equivalent replacements, improvements, etc. made within the technical concept and / or technical principle of the present disclosure will fall within the protection scope of the present disclosure.
Claims
1. An automatic identification system for cultivated land monitoring, characterized in that: The system comprises: A data building module is used to use a number of leaf images marked with plant names as training samples; based on the training samples and a preset neural network algorithm, determine the correspondence between the plant name and the leaf feature vector to update the preset sample database; The target acquisition module is used to acquire the collected pictures uploaded by the preset collection device, and obtain the target pictures corresponding to the plant leaves in the collected pictures based on the maximum inter-class variance algorithm; An image processing module is used to import the target image into a preset scale-invariant feature transformation algorithm to obtain a target feature vector corresponding to the target image; and then determine the plant name corresponding to the collected image based on the target feature vector and the sample database; The data building module includes: a sample acquisition unit, a sample processing unit and a database updating unit; The sample acquisition unit is used to acquire a number of original images with plant names marked; The sample processing unit is used to perform grayscale processing and image segmentation on the original image to obtain a leaf image as a training sample; import the training sample into a preset neural network algorithm to obtain a leaf feature vector corresponding to the training sample; use the plant name as the clustering center to perform clustering processing on the leaf feature vector through a proximity algorithm, remove invalid feature vectors, and obtain a corresponding relationship between the leaf feature vector and the plant name; The database updating unit is used to update the corresponding relationship between the leaf feature vector and the plant name to a preset sample database.
2. The automatic identification system for cultivated land monitoring according to claim 1 is characterized in that: The image processing module includes: a feature transformation unit; The feature transformation unit is used to import the target image into the scale-invariant feature transformation algorithm to obtain the key points in the target image through the Gaussian function preset in the algorithm; determine the Gaussian weight function parameters for calculating the histogram corresponding to the key points; determine the Taylor expansion of the histogram to perform quadratic curve fitting to determine the direction corresponding to the key points; determine the key point descriptor corresponding to the key point through the scale-invariant feature transformation; and determine the target feature vector corresponding to the target image based on the key points, the key point directions and the key point descriptors.
3. The automatic identification system for farmland monitoring according to claim 1 is characterized in that: The system also includes a secondary verification module; The secondary verification module includes a standard picture unit and a MATLAB verification unit; The standard picture unit is used to store standard pictures corresponding to plant names; The MATLAB verification unit is used to determine the similarity between the collected picture and the standard picture based on the MATLAB software and the plant name after determining the plant name corresponding to the collected picture; When the similarity is less than a preset similarity threshold, an error maintenance instruction is generated to the maintenance terminal.
4. An automatic identification method for cultivated land monitoring, characterized in that: Methods include: Use several leaf images with plant names labeled as training samples; Based on the training samples and the preset neural network algorithm, the corresponding relationship between the plant name and the leaf feature vector is determined to update the preset sample database; Obtain the collected images uploaded by the preset collection device, and obtain the target images corresponding to the plant leaves in the collected images based on the maximum inter-class variance algorithm; Importing the target image into a preset scale-invariant feature transformation algorithm to obtain a target feature vector corresponding to the target image; and then determining the plant name corresponding to the collected image based on the target feature vector and the sample database; Use several leaf images with plant names labeled as training samples; Based on the training samples and the preset neural network algorithm, the corresponding relationship between the plant name and the leaf feature vector is determined to update the preset sample database, specifically including: Obtain several original images with plant names marked; The original image is subjected to grayscale processing and image segmentation to obtain a leaf image as a training sample; the training sample is imported into a preset neural network algorithm to obtain a leaf feature vector corresponding to the training sample; the plant name is used as a clustering center to cluster the leaf feature vectors through a proximity algorithm, and invalid feature vectors are removed to obtain a corresponding relationship between the leaf feature vector and the plant name; The correspondence between the leaf feature vector and the plant name is updated to the preset sample database.
5. The automatic identification method for cultivated land monitoring according to claim 4, characterized in that: Importing the target image into a preset scale-invariant feature transformation algorithm to obtain a target feature vector corresponding to the target image specifically includes: Import the target image into the scale-invariant feature transformation algorithm to obtain the key points in the target image through the Gaussian function preset in the algorithm; Determine and calculate the Gaussian weight function parameters of the histogram corresponding to the key point; Determine the Taylor expansion of the histogram to perform quadratic curve fitting to determine the direction corresponding to the key point; determine the key point descriptor corresponding to the key point through scale-invariant feature transformation; According to the key points, key point directions and key point descriptors, the target feature vector corresponding to the target image is determined.
6. The automatic identification method for cultivated land monitoring according to claim 4, characterized in that: The preset verification database is used to store standard pictures corresponding to plant names; The method further comprises: After determining the plant name corresponding to the collected image, the similarity between the collected image and the standard image is determined based on MATLAB software and the preset verification database; when the similarity is less than the preset similarity threshold, an error maintenance instruction is generated to the maintenance terminal.