Breeding animal identification method and device, computer equipment and storage medium

Abstract

The embodiment of the invention belongs to the field of image processing, and relates to a bred animal identification method and device, computer equipment and a storage medium, and the method comprises the following steps: obtaining a multispectral bred animal image; preprocessing the multispectral bred animal image to obtain a corrected bred animal image; performing feature vector extraction on the corrected bred animal image to obtain a texture feature vector and a frequency domain feature vector; performing fusion analysis processing on the texture feature vector and the frequency domain feature vector to obtain multispectral feature representation; training a preset feature recognition model according to the multispectral feature representation to obtain a bred animal recognition model; and acquiring a real-time bred animal image, and inputting the real-time bred animal image into the bred animal identification model to obtain a bred animal identification result. According to the invention, the method can achieve the accurate and effective recognition of the bred animals, and improves the recognition accuracy and efficiency of the bred animals.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, specifically to the field of financial technology, and in particular to a method, device, computer equipment and storage medium for identifying farmed animals. Background Art

[0002] In the field of livestock insurance business, the identification of the number of livestock (such as pigs, cattle, sheep, chickens, ducks, etc.), that is, the determination of insurance or loss assessment of target objects, is one of the common tasks in property insurance business scenarios. It is usually used to check live animals, death conditions, death time, estimated weight, etc. The goal of pig body recognition is to automatically identify the individual identity and behavior of livestock. In order to achieve this goal, the existing technology uses image feature matching methods to match the features extracted from livestock images with livestock features to identify livestock targets in livestock images.

[0003] However, in the actual pig farm environment, a key technical challenge faced by farmed animal identification is how to accurately capture the multispectral image features of individuals when farmed animals are free to move. Due to the wide range of activities and fast movement speed of farmed animals, as well as frequent mutual occlusion and changes in body posture, the position, angle and distance of farmed animals in the imaging screen are constantly changing. These factors seriously affect the image quality and increase the difficulty of feature extraction and matching. At the same time, the surface of farmed animals is easily stained, and the skin characteristics of different parts vary greatly, which reduces the consistency and stability of multispectral reflectance signals. In addition, the complex and changeable lighting conditions in the pig farm environment, such as changes in weather and interference from artificial lighting, also affect the effect of multispectral imaging, making it difficult to effectively identify farmed animals. Summary of the invention

[0004] The purpose of the embodiments of the present application is to provide a method, apparatus, computer equipment and storage medium for identifying farmed animals, so as to solve the problem that farmed animals in images cannot be accurately and effectively identified.

[0005] In order to solve the above technical problems, the present application embodiment provides a method for identifying farmed animals, which adopts the following technical solutions:

[0006] Acquire multispectral images of farmed animals;

[0007] Preprocessing the multispectral farmed animal image to obtain a corrected farmed animal image;

[0008] Extracting feature vectors from the corrected farmed animal images to obtain texture feature vectors and frequency domain feature vectors;

[0009] Performing fusion analysis on the texture feature vector and the frequency domain feature vector to obtain a multi-spectral feature representation;

[0010] The preset feature recognition model is trained according to the multi-spectral feature representation to obtain a farmed animal recognition model;

[0011] Real-time farmed animal images are acquired, and the real-time farmed animal images are input into the farmed animal recognition model to obtain farmed animal recognition results.

[0012] Furthermore, the step of obtaining multispectral images of farmed animals specifically includes:

[0013] Obtain images of farmed animals and extract identification;

[0014] Extracting original farmed animal images from a database according to the farmed animal image extraction identifier;

[0015] Preset band range information is obtained, and the original farmed animal images are screened according to the preset band range information to obtain visible light band images of farmed animals, near infrared band images of farmed animals, and mid-infrared band images of farmed animals, and the visible light band images of farmed animals, near infrared band images of farmed animals, and mid-infrared band images of farmed animals are integrated into the multispectral farmed animal images.

[0016] Furthermore, the step of preprocessing the multispectral farmed animal image to obtain a corrected farmed animal image specifically includes:

[0017] Performing denoising processing on the multispectral farmed animal image to obtain a denoised farmed animal image;

[0018] Performing distortion correction on the multispectral denoised farmed animal image to obtain an initial corrected farmed animal image;

[0019] Performing a quality assessment on the initial corrected farmed animal image, and determining whether the quality assessment result meets a preset quality requirement;

[0020] If the initial corrected farmed animal image meets the preset quality requirement, the initial corrected farmed animal image is used as the corrected farmed animal image.

[0021] Furthermore, the step of extracting feature vectors from the corrected farmed animal images to obtain texture feature vectors and frequency domain feature vectors specifically includes:

[0022] Performing radiation correction processing on the corrected farmed animal image to obtain a standard farmed animal image;

[0023] Extracting a standard visible light farmed animal image from the standard farmed animal image, inputting the standard visible light farmed animal image into a pre-trained texture feature extraction model to obtain image texture features, and mapping the image texture features into the texture feature vector;

[0024] Extracting a standard near-infrared farmed animal image and a standard mid-infrared farmed animal image from the standard farmed animal image, and performing frequency domain feature extraction on the standard near-infrared farmed animal image and the standard mid-infrared farmed animal image according to a wavelet transform method to obtain image frequency domain features;

[0025] The image frequency domain features are converted into the frequency domain feature vector based on the statistical value calculation of the wavelet coefficients.

[0026] Furthermore, the step of fusing and analyzing the texture feature vector and the frequency domain feature vector to obtain a multi-spectral feature representation specifically includes:

[0027] Performing feature vector fusion on the texture feature vector and the frequency domain feature vector to obtain a fused feature vector;

[0028] Performing principal component analysis on the fused feature vector to obtain a difference feature vector;

[0029] The difference feature vector is learned and optimized according to a multi-kernel learning algorithm to obtain a multi-spectral feature representation.

[0030] Furthermore, the step of training a preset feature recognition model according to the multispectral feature representation to obtain a farmed animal recognition model specifically includes:

[0031] Obtaining the identity information of farmed animals;

[0032] Associating the farmed animal identity information with the multispectral feature representation to construct a feature identity mapping data set;

[0033] Performing data enhancement on the feature identity mapping dataset to obtain an enhanced feature identity mapping dataset;

[0034] The preset feature recognition model is trained and optimized according to the enhanced feature identity mapping data set to obtain the farmed animal recognition model.

[0035] Furthermore, after the step of acquiring the real-time farmed animal image, inputting the real-time farmed animal image into the farmed animal recognition model, and obtaining the farmed animal recognition result, the following steps are also included:

[0036] Extracting farmed animal identification failure information from the farmed animal identification results;

[0037] Determining a farmed animal target according to the farmed animal recognition failure information, and extracting a feature vector of the farmed animal target to obtain a target feature vector;

[0038] The target feature vector is added to a farmed animal feature library, and the farmed animal identification model is updated according to the farmed animal feature library.

[0039] In order to solve the above technical problems, the present application also provides a farmed animal identification device, which adopts the following technical solution:

[0040] An image acquisition module, used for acquiring multi-spectral images of farmed animals;

[0041] An image processing module, used for preprocessing the multispectral farmed animal image to obtain a corrected farmed animal image;

[0042] A feature extraction module is used to extract feature vectors from the corrected farmed animal image to obtain texture feature vectors and frequency domain feature vectors;

[0043] A feature processing module, used for fusing and analyzing the texture feature vector and the frequency domain feature vector to obtain a multi-spectral feature representation;

[0044] A model training module is used to train a preset feature recognition model according to the multispectral feature representation to obtain a farmed animal recognition model;

[0045] The result recognition module is used to obtain real-time farmed animal images, input the real-time farmed animal images into the farmed animal recognition model, and obtain farmed animal recognition results.

[0046] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0047] A computer device comprises a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the farmed animal identification method as described in any one of the above when executing the computer-readable instructions.

[0048] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0049] A computer-readable storage medium having computer-readable instructions stored thereon, wherein the computer-readable instructions, when executed by a processor, implement the steps of any of the above methods for identifying farmed animals.

[0050] Compared with the prior art, the embodiment of the present application has the following beneficial effects: the embodiment acquires multispectral farmed animal images; pre-processes the multispectral farmed animal images to obtain corrected farmed animal images; extracts feature vectors from the corrected farmed animal images to obtain texture feature vectors and frequency domain feature vectors; fuses and analyzes the texture feature vectors and the frequency domain feature vectors to obtain multispectral feature representation; trains a preset feature recognition model according to the multispectral feature representation to obtain a farmed animal recognition model; acquires real-time farmed animal images, and inputs the real-time farmed animal images into the farmed animal recognition model to obtain farmed animal recognition results. This enables accurate and effective farmed animal recognition, and improves the accuracy and efficiency of farmed animal recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0052] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0053] Figure 2 A flow chart of an embodiment of a method for identifying farmed animals according to the present application;

[0054] Figure 3 yes Figure 2 A flowchart of a specific implementation of step S10;

[0055] Figure 4 yes Figure 2 A flowchart of a specific implementation of step S20;

[0056] Figure 5 yes Figure 2 A flowchart of a specific implementation of step S30;

[0057] Figure 6 yes Figure 2 A flowchart of a specific implementation of step S40;

[0058] Figure 7 yes Figure 2 A flowchart of a specific implementation of step S50;

[0059] Figure 8 is a schematic structural diagram of an embodiment of a farmed animal identification device according to the present application;

[0060] Fig. 9 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0061] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0062] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor does it refer to non-related or alternative embodiments that are mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0063] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0064] like Figure 1 As shown, the system architecture 100 may include a terminal device 101, a network 102 and a server 103. The terminal device 101 may be a laptop 1011, a tablet computer 1012 or a mobile phone 1013. The network 102 is used to provide a medium for a communication link between the terminal device 101 and the server 103. The network 102 may include various connection types, such as wired, wireless communication links or optical fiber cables.

[0065] The user can use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages, etc. Various communication client applications can be installed on the terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0066] The terminal device 101 can be any electronic device with a display screen and supporting web browsing. In addition to a laptop computer 1011, a tablet computer 1012 or a mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV), a laptop computer, a desktop computer, etc.

[0067] The server 103 may be a server that provides various services, such as a background server that provides support for web pages displayed on the terminal device 101 .

[0068] It should be noted that the farmed animal identification method provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the farmed animal identification device is generally set in the server / terminal device.

[0069] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0070] Continue to refer Figure 2 , shows a flow chart of an embodiment of a method for identifying farmed animals according to the present application. The method for identifying farmed animals comprises the following steps:

[0071] Step S10, obtaining multispectral images of farmed animals;

[0072] In this embodiment, the multispectral farmed animal image refers to an image subject of farmed animals, and contains image information of multiple spectral bands, wherein the farmed animals include farmed species such as pigs, cattle, sheep, chickens, ducks, geese, etc. The multispectral farmed animal image is obtained by multispectral imaging technology. Multispectral imaging technology is a technology that uses multiple spectral bands to image objects. It can capture the reflection or emission characteristics of objects at different wavelengths, thereby providing richer information than a single spectral band. Multispectral farmed animal images can be obtained by photographing farmed animals with multispectral imaging equipment.

[0073] Step S20, preprocessing the multispectral farmed animal image to obtain a corrected farmed animal image;

[0074] In this embodiment, preprocessing includes denoising and distortion correction. By preprocessing the multispectral farmed animal images, the image noise of the multispectral farmed animal images is effectively removed, and the image distortion, stretching or compression caused by factors such as the lens or shooting environment is eliminated to obtain clearer and more accurate corrected farmed animal images.

[0075] Step S30, extracting feature vectors from the corrected farmed animal image to obtain texture feature vectors and frequency domain feature vectors;

[0076] In this embodiment, the texture feature vector extraction of the corrected farmed animal images is performed based on a pre-trained texture feature extraction model, and the frequency domain feature vector extraction of the corrected farmed animal images is performed based on a wavelet transform method, wherein the pre-trained texture feature extraction model can adopt a convolutional neural network model, specifically, the InceptionNet (Inception architecture) module can be used to combine convolution kernels and pooling operations of different scales, so as to effectively capture the texture features of the corrected farmed animal images.

[0077] Step S40, performing fusion analysis on the texture feature vector and the frequency domain feature vector to obtain a multi-spectral feature representation;

[0078] In this embodiment, the fusion analysis process includes principal component analysis and learning optimization based on a multi-kernel learning algorithm, wherein principal component analysis is a method of converting a set of variables that may be correlated into a set of linearly uncorrelated variables, i.e., principal components, through orthogonal transformation. The principal components are arranged from large to small according to variance, and the first principal component has the largest variance, i.e., explains the largest variability in the data. Multi-kernel learning is a machine learning method that combines multiple kernel functions to build a more powerful learning model. Kernel functions are a key component in kernel methods such as SVM, which are used to map data to a high-dimensional space so that it is linearly separable in the high-dimensional space.

[0079] Step S50, training a preset feature recognition model according to the multispectral feature representation to obtain a farmed animal recognition model;

[0080] In this embodiment, the preset feature recognition model can adopt a support vector machine SVM model. The core idea of ​​SVM is to find an optimal hyperplane in a high-dimensional space to maximize the interval between data points of different categories. In the binary classification problem of farmed animal identification, SVM attempts to find a hyperplane so that the two types of data points in the multispectral feature representation are located on both sides of the hyperplane, and the distance (i.e., the interval) from the hyperplane to the nearest point of the two types of data in the multispectral feature representation is the largest, thereby effectively distinguishing the two types of data in the multispectral feature representation to form effective farmed animal identification.

[0081] Step S60, acquiring real-time farmed animal images, inputting the real-time farmed animal images into the farmed animal recognition model, and obtaining farmed animal recognition results.

[0082] In this embodiment, the real-time farmed animal images are images obtained by photographing farmed animals in real time using a multispectral imaging device. Before the real-time farmed animal images are input into the farmed animal recognition model, the real-time farmed animal images are pre-processed such as the denoising and distortion correction described above, and then the real-time multispectral feature representation corresponding to the pre-processed real-time farmed animal images is extracted. The real-time multispectral feature representation is input into the farmed animal recognition model to identify the farmed animals in the images.

[0083] This embodiment obtains multispectral farmed animal images; preprocesses the multispectral farmed animal images to obtain corrected farmed animal images; extracts feature vectors from the corrected farmed animal images to obtain texture feature vectors and frequency domain feature vectors; performs fusion analysis on the texture feature vectors and the frequency domain feature vectors to obtain multispectral feature representation; trains a preset feature recognition model based on the multispectral feature representation to obtain a farmed animal recognition model; obtains real-time farmed animal images, and inputs the real-time farmed animal images into the farmed animal recognition model to obtain farmed animal recognition results. This enables accurate and effective farmed animal recognition, and improves the accuracy and efficiency of farmed animal recognition.

[0084] Continue to refer Figure 3 In some optional implementations of this embodiment, step S10 includes the following steps:

[0085] Step S101, obtaining farmed animal image extraction identification;

[0086] In this embodiment, the farmed animal image extraction identifier is unique identification information corresponding to the original farmed animal image, and the farmed animal image extraction identifier can be an information ID or a UUID (universally unique identifier).

[0087] Step S102, extracting original farmed animal images from a database according to the farmed animal image extraction identifier;

[0088] In this embodiment, the original farmed animal image refers to the original image data directly collected from the farm or growth environment of the animal farm through a multispectral imaging device. The farmed animal image extraction identifier is used as a query condition to perform a traversal query in the database, and the image data matched by the query is extracted to obtain the original farmed animal image.

[0089] Step S103, obtaining preset band range information, screening the original farmed animal images according to the preset band range information, obtaining visible light band farmed animal images, near infrared band farmed animal images, and mid-infrared band farmed animal images, and integrating the visible light band farmed animal images, near infrared band farmed animal images, and mid-infrared band farmed animal images into the multi-spectral farmed animal images.

[0090] In this embodiment, the preset band range information is preset information for performing image band screening on the original farmed animal images, and the preset band range information includes visible light band information for screening visible light band images of 400-700 nanometers; near infrared band information for screening near infrared band images of 700-1100 nanometers; and mid-infrared band information for screening mid-infrared band images of 1100-3000 nanometers. In this embodiment, the original farmed animal images can be screened by image processing software, and the visible light band farmed animal images are generated by extracting spectral information within the range of 400-700nm from the original farmed animal images, and the near infrared band farmed animal images are generated by extracting spectral information within the range of 700-1100nm from the original farmed animal images, and the mid-infrared band farmed animal images are generated by extracting spectral information within the range of 1100-3000nm from the original farmed animal images.

[0091] This embodiment obtains farmed animal image extraction identification; extracts original farmed animal images from a database according to the farmed animal image extraction identification; obtains preset band range information, and screens the original farmed animal images according to the preset band range information to obtain visible light band farmed animal images, near infrared band farmed animal images, and mid-infrared band farmed animal images, and integrates the visible light band farmed animal images, near infrared band farmed animal images, and mid-infrared band farmed animal images into the multispectral farmed animal images. This effectively realizes the acquisition of multispectral farmed animal images including farmed animal images of different bands, so as to provide reliable data support for farmed animal identification.

[0092] Continue to refer Figure 4 In some optional implementations of this embodiment, step S20 includes the following steps:

[0093] Step S201, performing denoising processing on the multispectral farmed animal image to obtain a denoised farmed animal image;

[0094] In this embodiment, Gaussian filtering can be used to denoise the multispectral farmed animal images. By selecting the Gaussian kernel size and determining the standard deviation, the selected Gaussian kernel (including size and standard deviation) is used to filter the farmed animal images in each light band of the multispectral farmed animal images, thereby obtaining denoised farmed animal images.

[0095] Step S202, performing distortion correction on the multispectral denoised farmed animal image to obtain an initial corrected farmed animal image;

[0096] In this embodiment, the distortion correction of the multispectral denoised farmed animal image includes the following steps: camera calibration: camera calibration is the process of establishing camera imaging model parameters, which include internal parameters of the camera (such as focal length, optical center, distortion coefficient, etc.) and external parameters (such as the position and posture of the camera in the world coordinate system); distortion model establishment: the distortion model of the camera is established according to the result of camera calibration, and the distortion model describes the nonlinear relationship between the pixel position in the image and its actual world coordinate; reading distortion parameters: reading internal parameters and distortion coefficients from the camera calibration results; calculating correction mapping: according to the distortion model and distortion parameters, calculating the mapping relationship of each pixel from the distorted image position to the corrected image position; resampling: using an interpolation algorithm (such as bilinear interpolation, bicubic interpolation, etc.) to resample the distorted image to generate a corrected image. In the resampling process, according to the correction mapping relationship, the pixel values ​​in the distorted image are mapped to the corresponding positions of the corrected image, thereby completing the distortion correction of the multispectral denoised farmed animal image.

[0097] Step S203, performing quality assessment on the initial corrected farmed animal image, and determining whether the quality assessment result meets the preset quality requirements;

[0098] In this embodiment, the preset quality requirement refers to a preset index that satisfies the image quality, and the preset indexes corresponding to the preset quality requirement include: clarity, contrast, noise level, color fidelity, structural similarity, and peak signal-to-noise ratio.

[0099] Step S204, if the initial corrected farmed animal image meets the preset quality requirement, the initial corrected farmed animal image is used as the corrected farmed animal image;

[0100] In this embodiment, the image indicators corresponding to the initial corrected farmed animal image are compared with the preset indicators corresponding to the above-mentioned preset quality requirements to determine whether the initial corrected farmed animal image meets the preset image requirements. When the image indicators corresponding to the initial corrected farmed animal image are greater than or equal to the preset indicators corresponding to the above-mentioned preset quality requirements, it is determined that the initial corrected farmed animal image meets the preset quality requirements.

[0101] Step S205, if the initial corrected farmed animal image does not meet the preset quality requirement, the multispectral farmed animal image is re-denoised and distortion corrected until the re-processed initial corrected farmed animal image meets the preset quality requirement.

[0102] In this embodiment, the above-mentioned re-denoising and distortion correction of the multispectral farmed animal images is an iterative process. By iteratively denoising and distortion correction of the multispectral farmed animal images, it is effectively ensured that the output initial corrected farmed animal images meet the preset quality requirements.

[0103] This embodiment performs denoising on the multispectral farmed animal image to obtain a denoised farmed animal image; performs distortion correction on the multispectral denoised farmed animal image to obtain an initial corrected farmed animal image; performs quality assessment on the initial corrected farmed animal image, and determines whether the quality assessment result meets the preset quality requirements; if the initial corrected farmed animal image meets the preset quality requirements, the initial corrected farmed animal image is used as the corrected farmed animal image; if the initial corrected farmed animal image does not meet the preset quality requirements, the multispectral farmed animal image is denoised and distortion corrected again until the reprocessed initial corrected farmed animal image meets the preset quality requirements. In this way, the multispectral farmed animal image is effectively processed into a corrected farmed animal image whose quality meets the preset quality requirements, so as to facilitate the subsequent feature vector extraction process.

[0104] Continue to refer Figure 5 In some optional implementations of this embodiment, step S30 includes the following steps:

[0105] Step S301, performing radiation correction processing on the corrected farmed animal image to obtain a standard farmed animal image;

[0106] In this embodiment, radiation correction mainly solves the radiation error problem in the image, such as uneven brightness, low contrast, etc., and radiation correction includes the following steps: Atmospheric correction: To eliminate the influence of the atmosphere on the image radiation, the atmospheric transmission model or atmospheric correction algorithm can be used to perform atmospheric radiation correction on the corrected farmed animal image according to atmospheric conditions (such as aerosol concentration, atmospheric transmittance, etc.) and sensor parameters (such as spectral response function, gain, etc.); Sensor correction: Correct the radiation error of the sensor itself, including the nonlinear response, dark current, dark noise, etc. of the correction sensor of the multi-spectral imaging device, and the corrected farmed animal image can be corrected by using sensor correction parameters or sensor correction algorithms; Brightness normalization: Adjust the brightness of the corrected farmed animal image to a uniform range to improve the comparability and visual effect of the corrected farmed animal image. This can be achieved by adjusting the brightness and contrast of the corrected farmed animal image.

[0107] Step S302, extracting a standard visible light farmed animal image from the standard farmed animal image, inputting the standard visible light farmed animal image into a pre-trained texture feature extraction model to obtain image texture features, and mapping the image texture features into the texture feature vector;

[0108] In this embodiment, by checking the band information (visible light band, near infrared band, mid-infrared band) corresponding to the standard farmed animal image, and selecting the standard farmed animal image in the visible light band for extraction, a standard visible light farmed animal image is obtained. The pre-trained texture feature extraction model adopts a convolutional neural network model, and can be extracted from the model database through the feature model extraction identifier to achieve the image texture feature acquisition of the visible light band farmed animal image through the texture feature extraction model. The texture feature vector includes grayscale co-occurrence matrix features, local texture information, and grayscale run length matrix. The mapping of image texture features includes feature combination: combining the features extracted from the grayscale co-occurrence matrix, local texture information, and grayscale run length matrix to form a comprehensive texture feature set; feature selection: according to needs, you can choose to retain all features or perform feature selection to remove redundant or irrelevant features and improve the effectiveness and calculation efficiency of the feature vector; feature normalization: normalize the feature vector so that the value of each feature is on the same scale; form a texture feature vector: after the above steps, the feature vector obtained is the texture feature vector of the image.

[0109] Step S303, extracting a standard near-infrared farmed animal image and a standard mid-infrared farmed animal image from the standard farmed animal image, and performing frequency domain feature extraction on the standard near-infrared farmed animal image and the standard mid-infrared farmed animal image according to a wavelet transform method to obtain image frequency domain features;

[0110] In this embodiment, by checking the band information (visible light band, near infrared band, mid-infrared band) corresponding to the standard farmed animal images, and selecting the standard farmed animal images in the near infrared band and the mid-infrared band for extraction, a standard near-infrared farmed animal image and a standard mid-infrared farmed animal image are obtained. The standard near-infrared farmed animal image and the standard mid-infrared farmed animal image are multi-scale decomposed by the wavelet transform method, and the frequency domain features of the image are extracted at different scales to obtain the image frequency domain features. The steps of multiple-degree decomposition include selecting a wavelet basis: first, it is necessary to select a suitable wavelet basis (such as Haar wavelet, Daubechies wavelet, Symlets wavelet, etc.). In this embodiment, Haar wavelet is used; performing wavelet transform: performing two-dimensional wavelet transform on the standard near-infrared farmed animal image and the standard mid-infrared farmed animal image respectively. The two-dimensional wavelet transform can be achieved by first performing a one-dimensional wavelet transform on each row of the image and then performing a one-dimensional wavelet transform on each column after the transform, or it can be completed by a special two-dimensional wavelet transform algorithm; multi-scale decomposition: through repeated wavelet transforms, the image can be decomposed into sub-band images at multiple scales. These sub-band images respectively represent the information of the image in different frequency ranges, including low-frequency (approximate) components and high-frequency (detail) components. The sub-band image is the image frequency domain feature.

[0111] Step S304: converting the image frequency domain features into the frequency domain feature vector based on the statistical value calculation of the wavelet coefficients.

[0112] In this embodiment, by calculating the statistical values ​​of the frequency domain features of the image, such as the mean, variance, energy (the sum of the squares of the coefficients), entropy (a statistic based on the distribution of the coefficients), etc., the statistical values ​​of the frequency domain features of each image at different scales are combined to form a frequency domain feature descriptor, and then the frequency domain feature descriptor is normalized to obtain a frequency domain feature vector.

[0113] This embodiment obtains a standard farmed animal image by performing radiation correction processing on the corrected farmed animal image; extracts a standard visible light farmed animal image from the standard farmed animal image, inputs the standard visible light farmed animal image into a pre-trained texture feature extraction model to obtain image texture features, and maps the image texture features into the texture feature vector; extracts a standard near-infrared farmed animal image and a standard mid-infrared farmed animal image from the standard farmed animal image, performs frequency domain feature extraction on the standard near-infrared farmed animal image and the standard mid-infrared farmed animal image according to the wavelet transform method to obtain image frequency domain features; and converts the image frequency domain features into the frequency domain feature vector based on the statistical value calculation of the wavelet coefficient. In this way, the texture feature vector representing the texture of the farmed animal and the frequency domain feature vector representing the frequency domain difference of the image are effectively obtained, so as to facilitate the subsequent fusion analysis and processing.

[0114] Continue to refer Figure 6 In some optional implementations of this embodiment, step S40 includes the following steps:

[0115] Step S401, performing feature vector fusion on the texture feature vector and the frequency domain feature vector to obtain a fused feature vector;

[0116] In this embodiment, the texture feature vector and the frequency domain feature vector may be connected end to end to achieve feature vector fusion, thereby forming a new, longer fused feature vector.

[0117] Step S402, performing principal component analysis on the fused feature vector to obtain a difference feature vector;

[0118] In this embodiment, PCA is a commonly used data dimensionality reduction technology. The original data is converted into a new coordinate system through linear transformation, so that the converted data has the largest variance on the first principal component, and the variance on subsequent principal components decreases successively. By retaining the first few principal components (i.e., the features with the largest differences), redundancy and noise in the data can be removed while retaining most of the useful information. By performing principal component analysis on the fused feature vector, the fused feature vector is converted to a new coordinate system, and then the first few features with the largest differences in the coordinate system are retained, thereby obtaining a difference feature vector.

[0119] Step S403: performing learning optimization on the difference feature vector according to a multi-kernel learning algorithm to obtain a multi-spectral feature representation.

[0120] In this embodiment, the multi-core learning algorithm is an algorithm that combines multiple kernel functions for learning. In multi-core learning, each kernel function can be regarded as a specific representation of the feature space, and the multi-core learning algorithm finds the feature representation that best suits the current data by optimizing the combined weights of these kernel functions. The optimization goal of multi-core learning is usually to find an optimal combination of kernel functions (i.e., the optimal weight distribution) so that the learning performance (such as classification accuracy, regression error, etc.) on a given data set is optimal. The optimal solution for the feature representation is obtained by defining the optimization target and solving the optimization problem corresponding to the optimization target. The optimization target can be expressed as minimizing a certain loss function while satisfying certain constraints (such as non-negative weights, weight normalization, etc.). Common optimization algorithms include sequential minimum optimization (SMO), gradient descent, convex optimization, etc. In this embodiment, a gradient descent algorithm can be used to learn and optimize the difference feature vector by using a multi-core learning algorithm to find the most suitable multi-spectral feature representation of the difference feature vector.

[0121] This embodiment obtains a fused feature vector by performing feature vector fusion on the texture feature vector and the frequency domain feature vector; performs principal component analysis on the fused feature vector to obtain a difference feature vector; and performs learning optimization on the difference feature vector according to a multi-core learning algorithm, thereby effectively obtaining a multi-spectral feature representation with compact features to facilitate subsequent model training processing.

[0122] Continue to refer Figure 7 In some optional implementations of this embodiment, step S50 includes the following steps:

[0123] Step S501, obtaining the identity information of farmed animals;

[0124] In this embodiment, the farmed animal identity information is unique identity information assigned to each pig, which can be achieved through methods such as RFID tags, QR code ear tags, etc. Among them, this embodiment uses RFID tags, which can quickly and contactlessly read and record the identity information of farmed animals, thereby facilitating the acquisition of farmed animal identity information.

[0125] Step S502, associating the farmed animal identity information with the multispectral feature representation to construct a feature identity mapping data set;

[0126] In this embodiment, a database or data structure is created to store the identity information of farmed animals and the corresponding multispectral feature representations and the corresponding multispectral images of farmed animals, and a mapping relationship between the identity information of farmed animals and the multispectral feature representations is established in the database, so that the corresponding identity information of farmed animals can be quickly retrieved by querying the multispectral feature representation, or the corresponding multispectral feature representation can be retrieved through the identity information of farmed animals, so as to construct a feature identity mapping data set.

[0127] Step S503, performing data enhancement on the feature identity mapping data set to obtain an enhanced feature identity mapping data set;

[0128] In this embodiment, data enhancement technology is used to perform data expansion on the feature identity mapping data set to enhance the diversity and robustness of the feature identity mapping data set, thereby obtaining an enhanced feature identity mapping data set. In this embodiment, data enhancement can be achieved by performing geometric transformation (rotation, scaling, translation, cropping, flipping, etc.), pixel transformation (adding noise, color transformation, blurring, etc.), and advanced transformation (style transfer, image fusion, adversarial network (GAN) generation, etc.) on the multispectral image corresponding to the feature identity mapping data set.

[0129] Step S504: training and optimizing the preset feature recognition model according to the enhanced feature identity mapping data set to obtain the farmed animal recognition model.

[0130] In this embodiment, the preset feature recognition model can adopt a support vector machine SVM model, and the support vector machine SVM model is trained and optimized by enhancing the feature identity mapping data set to obtain a farmed animal recognition model that can effectively identify farmed animals in images.

[0131] This embodiment obtains farmed animal identity information; associates the farmed animal identity information with the multispectral feature representation to construct a feature identity mapping data set; performs data enhancement on the feature identity mapping data set to obtain an enhanced feature identity mapping data set; trains and optimizes the preset feature recognition model based on the enhanced feature identity mapping data set, thereby effectively obtaining a farmed animal recognition model that can accurately identify farmed animals, so as to facilitate subsequent accurate farmed animal identification.

[0132] In some optional implementations of this embodiment, after step S60, the following steps are further included:

[0133] Extracting farmed animal identification failure information from the farmed animal identification results;

[0134] In this embodiment, the farmed animal identification failure information includes the time and place of failure, the specific farmed animal identified (if it is identified but incorrectly), and the detailed reasons for the identification failure (such as image blur, occlusion, angle problem, etc.). The farmed animal identification failure information can be extracted by performing result information identification on the farmed animal identification results and searching for the identification information keywords corresponding to the farmed animal identification results.

[0135] Determining a farmed animal target according to the farmed animal recognition failure information, and extracting a feature vector of the farmed animal target to obtain a target feature vector;

[0136] In this embodiment, the corresponding farmed animal target can be determined by the ID of the corresponding farmed animal in the farmed animal identification failure information. After the farmed animal target is identified, the target feature vector of the farmed animal target in the image can be extracted by a deep learning algorithm (such as a convolutional neural network CNN). The target feature vector should be able to accurately describe the unique characteristics of the farmed animal, such as body shape, color, texture, etc.

[0137] The target feature vector is added to a farmed animal feature library, and the farmed animal identification model is updated according to the farmed animal feature library.

[0138] In this embodiment, the farmed animal feature library is a feature database corresponding to the farmed animal identification model, which is updated by adding the target feature vector to the farmed animal feature library. The farmed animal identification model can be retrained according to the updated farmed animal feature library and new training data (i.e., failed farmed animal images and their feature vectors), thereby realizing the update of the farmed animal identification model.

[0139] This embodiment extracts farmed animal identification failure information from the farmed animal identification result; determines the farmed animal target according to the farmed animal identification failure information, extracts feature vectors from the farmed animal target, and obtains target feature vectors; adds the target feature vectors to the farmed animal feature library, and updates the farmed animal identification model according to the farmed animal feature library. This effectively realizes real-time model update to ensure the recognition accuracy of the farmed animal identification model.

[0140] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0141] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0142] Further references Figure 8 , as a response to the above Figure 1 In order to realize the method shown in the figure, the present application provides an embodiment of a farmed animal identification device, which is similar to Figure 1 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0143] like Figure 8As shown, the farmed animal identification device 700 described in this embodiment includes: an image acquisition module 701, an image processing module 702, a feature extraction module 703, a feature processing module 704, a model training module 705, and a result identification module 706. Among them:

[0144] An image acquisition module 701 is used to acquire multi-spectral images of farmed animals;

[0145] An image processing module 702 is used to pre-process the multi-spectral farmed animal image to obtain a corrected farmed animal image;

[0146] A feature extraction module 703 is used to extract feature vectors from the corrected farmed animal image to obtain texture feature vectors and frequency domain feature vectors;

[0147] A feature processing module 704 is used to perform fusion analysis on the texture feature vector and the frequency domain feature vector to obtain a multi-spectral feature representation;

[0148] The model training module 705 is used to train the preset feature recognition model according to the multi-spectral feature representation to obtain a farmed animal recognition model;

[0149] The result recognition module 706 is used to obtain real-time farmed animal images, input the real-time farmed animal images into the farmed animal recognition model, and obtain farmed animal recognition results.

[0150] By adopting the above-mentioned farmed animal identification device, this embodiment can obtain multispectral farmed animal images; preprocess the multispectral farmed animal images to obtain corrected farmed animal images; extract feature vectors from the corrected farmed animal images to obtain texture feature vectors and frequency domain feature vectors; perform fusion analysis on the texture feature vectors and the frequency domain feature vectors to obtain multispectral feature representation; train a preset feature recognition model according to the multispectral feature representation to obtain a farmed animal recognition model; obtain real-time farmed animal images, input the real-time farmed animal images into the farmed animal recognition model, and obtain farmed animal recognition results. This achieves accurate and effective farmed animal recognition and improves the accuracy and efficiency of farmed animal recognition.

[0151] To solve the above technical problems, the present application also provides a computer device. Fig. 9 , Fig. 9 This is a basic structural block diagram of the computer device in this embodiment.

[0152] The computer device 8 includes a memory 81, a processor 82, and a network interface 83 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 8 with components 81-83, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.

[0153] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0154] The memory 81 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 81 can be an internal storage unit of the computer device 8, such as a hard disk or memory of the computer device 8. In other embodiments, the memory 81 can also be an external storage device of the computer device 8, such as a plug-in hard disk equipped on the computer device 8, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. Of course, the memory 81 can also include both the internal storage unit of the computer device 8 and its external storage device. In this embodiment, the memory 81 is generally used to store the operating system and various application software installed on the computer device 8, such as computer-readable instructions of the farmed animal identification method, etc. In addition, the memory 81 can also be used to temporarily store various types of data that have been output or are to be output.

[0155] The processor 82 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 82 is generally used to control the overall operation of the computer device 8. In this embodiment, the processor 82 is used to run computer-readable instructions or process data stored in the memory 81, such as computer-readable instructions for running the farmed animal identification method.

[0156] The network interface 83 may include a wireless network interface or a wired network interface. The network interface 83 is generally used to establish a communication connection between the computer device 8 and other electronic devices.

[0157] By using the above-mentioned computer equipment, this embodiment can obtain multispectral farmed animal images; preprocess the multispectral farmed animal images to obtain corrected farmed animal images; extract feature vectors from the corrected farmed animal images to obtain texture feature vectors and frequency domain feature vectors; perform fusion analysis on the texture feature vectors and the frequency domain feature vectors to obtain multispectral feature representation; train a preset feature recognition model according to the multispectral feature representation to obtain a farmed animal recognition model; obtain real-time farmed animal images, input the real-time farmed animal images into the farmed animal recognition model, and obtain farmed animal recognition results. This enables accurate and effective farmed animal recognition, and improves the accuracy and efficiency of farmed animal recognition.

[0158] The present application also provides another embodiment, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the farmed animal identification method as described above.

[0159] By adopting the above-mentioned computer-readable storage medium, this embodiment can obtain multispectral farmed animal images; preprocess the multispectral farmed animal images to obtain corrected farmed animal images; extract feature vectors from the corrected farmed animal images to obtain texture feature vectors and frequency domain feature vectors; perform fusion analysis on the texture feature vectors and the frequency domain feature vectors to obtain multispectral feature representation; train a preset feature recognition model according to the multispectral feature representation to obtain a farmed animal recognition model; obtain real-time farmed animal images, input the real-time farmed animal images into the farmed animal recognition model, and obtain farmed animal recognition results. This achieves accurate and effective farmed animal recognition, and improves the accuracy and efficiency of farmed animal recognition.

[0160] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0161] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

[0162] The non-Company software tools or components appearing in the embodiments of this application are merely examples and do not represent actual use.

Claims

1. A method for identifying farmed animals, characterized in that: The steps include: Acquire multispectral images of farmed animals; Preprocessing the multispectral farmed animal image to obtain a corrected farmed animal image; Extracting feature vectors from the corrected farmed animal images to obtain texture feature vectors and frequency domain feature vectors; Performing fusion analysis on the texture feature vector and the frequency domain feature vector to obtain a multi-spectral feature representation; The preset feature recognition model is trained according to the multi-spectral feature representation to obtain a farmed animal recognition model; Real-time farmed animal images are acquired, and the real-time farmed animal images are input into the farmed animal recognition model to obtain farmed animal recognition results.

2. The method for identifying farmed animals according to claim 1, characterized in that: The step of obtaining multispectral images of farmed animals specifically includes: Obtain images of farmed animals and extract identification; Extracting original farmed animal images from a database according to the farmed animal image extraction identifier; Preset band range information is obtained, and the original farmed animal images are screened according to the preset band range information to obtain visible light band images of farmed animals, near infrared band images of farmed animals, and mid-infrared band images of farmed animals, and the visible light band images of farmed animals, near infrared band images of farmed animals, and mid-infrared band images of farmed animals are integrated into the multispectral farmed animal images.

3. The method for identifying farmed animals according to claim 1, characterized in that: The step of preprocessing the multispectral farmed animal image to obtain a corrected farmed animal image specifically includes: Performing denoising processing on the multispectral farmed animal image to obtain a denoised farmed animal image; Performing distortion correction on the multispectral denoised farmed animal image to obtain an initial corrected farmed animal image; Performing a quality assessment on the initial corrected farmed animal image, and determining whether the quality assessment result meets a preset quality requirement; If the initial corrected farmed animal image meets the preset quality requirement, the initial corrected farmed animal image is used as the corrected farmed animal image.

4. The method for identifying farmed animals according to claim 1, characterized in that: The step of extracting feature vectors from the corrected farmed animal images to obtain texture feature vectors and frequency domain feature vectors specifically includes: Performing radiation correction processing on the corrected farmed animal image to obtain a standard farmed animal image; Extracting a standard visible light farmed animal image from the standard farmed animal image, inputting the standard visible light farmed animal image into a pre-trained texture feature extraction model to obtain image texture features, and mapping the image texture features into the texture feature vector; Extracting a standard near-infrared farmed animal image and a standard mid-infrared farmed animal image from the standard farmed animal image, and performing frequency domain feature extraction on the standard near-infrared farmed animal image and the standard mid-infrared farmed animal image according to a wavelet transform method to obtain image frequency domain features; The image frequency domain features are converted into the frequency domain feature vector based on the statistical value calculation of the wavelet coefficients.

5. The method for identifying farmed animals according to claim 1, characterized in that: The step of fusing and analyzing the texture feature vector and the frequency domain feature vector to obtain a multi-spectral feature representation specifically includes: Performing feature vector fusion on the texture feature vector and the frequency domain feature vector to obtain a fused feature vector; Performing principal component analysis on the fused feature vector to obtain a difference feature vector; The difference feature vector is learned and optimized according to a multi-kernel learning algorithm to obtain a multi-spectral feature representation.

6. The method for identifying farmed animals according to claim 1, characterized in that: The step of training the preset feature recognition model according to the multispectral feature representation to obtain the farmed animal recognition model specifically includes: Obtaining the identity information of farmed animals; Associating the farmed animal identity information with the multispectral feature representation to construct a feature identity mapping data set; Performing data enhancement on the feature identity mapping dataset to obtain an enhanced feature identity mapping dataset; The preset feature recognition model is trained and optimized according to the enhanced feature identity mapping data set to obtain the farmed animal recognition model.

7. The method for identifying farmed animals according to claim 1, characterized in that: After the step of acquiring the real-time farmed animal image, inputting the real-time farmed animal image into the farmed animal recognition model, and obtaining the farmed animal recognition result, the following steps are also included: Extracting farmed animal identification failure information from the farmed animal identification results; Determining a farmed animal target according to the farmed animal recognition failure information, and extracting a feature vector of the farmed animal target to obtain a target feature vector; The target feature vector is added to a farmed animal feature library, and the farmed animal identification model is updated according to the farmed animal feature library.

8. A farmed animal identification device, characterized in that: include: An image acquisition module, used for acquiring multi-spectral images of farmed animals; An image processing module, used for preprocessing the multispectral farmed animal image to obtain a corrected farmed animal image; A feature extraction module is used to extract feature vectors from the corrected farmed animal image to obtain texture feature vectors and frequency domain feature vectors; A feature processing module, used for fusing and analyzing the texture feature vector and the frequency domain feature vector to obtain a multi-spectral feature representation; A model training module is used to train a preset feature recognition model according to the multispectral feature representation to obtain a farmed animal recognition model; The result recognition module is used to obtain real-time farmed animal images, input the real-time farmed animal images into the farmed animal recognition model, and obtain farmed animal recognition results.

9. A computer device, characterized in that: It comprises a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the farmed animal identification method as claimed in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the farmed animal identification method according to any one of claims 1 to 7.

Images