An image recognition-based chicken body type and weight evaluation system and method

Abstract

The present application relates to the technical field of chicken body type and weight evaluation detection, and proposes a chicken body type and weight evaluation system and method based on image recognition, which extracts a plurality of chicken images from chicken house images, calculates the 2D and 3D features of each chicken corresponding to the chicken images by using the depth information of the chicken house images, then generates the 3D model of each chicken by using the 2D and 3D features and calculates the weight parameters, and finally determines whether to feed back the evaluation alarm signal according to the number of chicken images with abnormal size parameters and / or weight parameters in the plurality of chicken house images. Thus, the combination of image recognition and depth information makes the position of the image sensor in the chicken house no longer constrained, improves the scene adaptability, at the same time, the volume is corrected by generating the 3D model of each chicken, and the weight parameters are determined based on the conversion relationship between the model unit volume and weight, which improves the body type and weight evaluation accuracy of a large number of chickens in the chicken house.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of chicken body weight evaluation, and in particular to a chicken body weight evaluation system and method based on image recognition. BACKGROUND

[0002] Chicken body weight evaluation plays a crucial role in the poultry industry. Accurate measurement of body weight can directly reflect the growth status and nutritional level of chickens, providing a scientific basis for feeding management. Changes in body weight directly reflect the growth rate and health status of chickens, allowing breeders to adjust feed formulations and feeding environments in a timely manner to ensure that chickens receive optimal nutritional support at each growth stage. Secondly, body weight is closely related to key production indicators such as egg production rate, egg weight, and meat quality. By reasonably evaluating these indicators, breeders can optimize feeding strategies, select more suitable breeds for breeding, and thus improve overall economic efficiency. For example, larger chickens may lay more eggs, while chickens with moderate body weight may have better meat quality, all of which require precise evaluation to make judgments. In addition, body weight evaluation also helps in early warning of diseases and improvement of breeds. Sudden changes in body weight may indicate health problems, and through timely detection and treatment, disease transmission can be reduced and economic losses can be minimized.

[0003] Although there are many methods for detecting the body size or weight of animals and plants on the market, such as CN118038160A proposes a tuna weight estimation method based on a deep learning algorithm, which converts pixel size to actual distance to estimate the weight of tuna. In the context of chicken breeding, the number of chickens in a single chicken coop can reach several hundred or even thousands, but the detection objects of existing methods are basically single or small numbers of animals and plants, and they do not have the ability to detect the body size or weight of a large number of animals and plants simultaneously.

[0004] Therefore, how to improve the accuracy of body size and weight evaluation of a large number of chickens in a chicken coop and provide precise data support for chicken feeding management is a technical problem that needs to be solved. SUMMARY

[0005] To solve the above technical problems, the present application provides a chicken body weight evaluation system and method based on image recognition, which aims to solve the problems of low detection accuracy, high computational load, and high equipment cost of existing animal and plant body size or weight detection methods when detecting a large number of animals and plants.

[0006] The present application provides a chicken body weight evaluation system based on image recognition, comprising:

[0007] An image acquisition and processing module configured to acquire a plurality of chicken house images of a target chicken house, identify and segment a plurality of chicken images in each of the chicken house images, and extract depth information in the chicken house images;

[0008] A size measurement module configured to measure size parameters of each chicken in each of the chicken house images based on the segmented plurality of chicken images and the depth information, wherein the size parameters include 2D features and 3D features of each chicken;

[0009] A weight estimation module configured to generate a 3D model of each chicken based on the 2D features and 3D features of each chicken in each of the chicken house images, correct the size parameters of each chicken, and calculate weight parameters of each chicken;

[0010] A result feedback module configured to determine whether there are chicken images with abnormal size parameters and / or weight parameters exceeding a preset number in each of the chicken house images collected at each time, and if so, feed back evaluation warning information.

[0011] Optionally, the image acquisition and processing module specifically includes:

[0012] An image acquisition unit;

[0013] The image acquisition unit is configured to acquire chicken house images collected by a plurality of image sensors arranged at different positions of the target chicken house.

[0014] The chicken house images collected by the plurality of image acquisition units are configured as a plurality of images covering different ranges of the target chicken house.

[0015] Optionally, the image acquisition and processing module specifically includes:

[0016] An image segmentation unit;

[0017] The image segmentation unit is configured to use a PIDNet model trained based on chicken images to identify and segment chicken images in a plurality of chicken house images of the target chicken house, to obtain chicken image detection results of each image.

[0018] The chicken image detection results specifically include: 、 、 ;

[0019] wherein, represents chicken image detection results of the Mth chicken house image, represents a mask of the Nth chicken detected in the Mth chicken house image.

[0020] Optionally, the image acquisition and processing module specifically comprises:

[0021] a depth estimation unit;

[0022] The depth estimation unit is configured to extract the depth information of each chicken coop image using a SideRT model and store the depth information in the form of a depth map.

[0023] Optionally, the size calculation module specifically comprises:

[0024] a 2D feature extraction unit;

[0025] The 2D feature extraction unit is configured to scale the information extracted from the mask generated based on each chicken image based on the depth information, and calculate the 2D features of each chicken image.

[0026] The 2D features include a projected area S, a width W of the chicken image, and a perimeter L of the chicken image, and the expression is specifically:

[0027] ;

[0028] ;

[0029] ;

[0030] wherein, is the area of the mask, is the width of the mask, is the perimeter of the mask, D is the average depth value at the position of the mask, and k is a constant for adjusting the degree of nonlinearity, and b are proportion factors for adjusting the influence of depth on area, and c is an offset for preventing the area from being scaled too large when the depth value is too small.

[0031] Optionally, the size calculation module specifically comprises:

[0032] a 3D feature extraction unit;

[0033] The 3D feature extraction unit is configured to estimate the 3D features of each chicken image according to the depth information of each chicken coop image and the projected area S.

[0034] The 3D features include a volume V, and the expression is specifically:

[0035] The 3D features include a volume V, and the expression is specifically:

[0036] ;

[0037] wherein V represents the estimated volume, represents the average depth value of the mask peripheral line of each chicken image, represents the depth value of the i-th pixel corresponding to the mask position of each chicken image.

[0038] Optionally, the body weight estimation module specifically comprises:

[0039] 3D model generation unit, the 3D model generation unit is configured to generate a feature data group based on the 2D feature and the 3D feature of each chicken in each chicken coop image, input the feature data group into a neural radiance field model to obtain a feature latent representation, and input the feature latent representation into a CLIP-Mesh model to obtain a 3D model of each chicken;

[0040] Volume correction unit, the volume correction unit is configured to correct the 3D feature of each chicken to a volume corresponding to the 3D model of each chicken according to the 3D model of each chicken.

[0041] Optionally, the body weight estimation module specifically comprises:

[0042] Body weight calculation unit;

[0043] The body weight calculation unit is configured to calculate the body weight parameter of each chicken according to the 3D model of each chicken and a preset ratio of weight to unit volume.

[0044] Optionally, the result feedback module specifically comprises:

[0045] Abnormality judgment unit, the abnormality judgment unit is configured to judge whether the size parameter and / or the body weight parameter of each chicken belong to an abnormal value, and if so, determine the chicken as an abnormal target;

[0046] Evaluation alarm unit, the evaluation alarm unit is configured to judge whether the number of chickens determined as abnormal targets in each chicken coop image collected at each time exceeds a preset number, and if so, feedback an evaluation alarm signal.

[0047] The application further provides a chicken body shape and weight evaluation method based on image recognition, comprising:

[0048] S1: acquiring a plurality of chicken coop images of a target chicken coop, identifying and segmenting a plurality of chicken images in each chicken coop image to obtain depth information in the chicken coop image;

[0049] S2: based on the plurality of chicken images obtained by segmentation and the depth information, calculating the size parameter of each chicken in each chicken coop image; wherein the size parameter comprises the 2D feature and the 3D feature of each chicken;

[0050] S3: generating a 3D model of each chicken based on the 2D features and the 3D features of each chicken in each chicken house image, correcting the size parameters of each chicken and calculating the weight parameters of each chicken;

[0051] S4: judging whether there are chicken images with abnormal size parameters and / or weight parameters exceeding the preset number in each chicken house image collected at each time, and if so, feeding back evaluation alarm information.

[0052] The beneficial effects of the present application are that a chicken body type and weight evaluation system and method based on image recognition are proposed, a plurality of chicken images in a chicken house image are extracted, the 2D features and the 3D features of the corresponding chicken of each chicken image are calculated using the depth information of the chicken house image, then the 3D model of each chicken is generated using the 2D features and the 3D features and the weight parameters are calculated, and finally, whether to feed back an evaluation alarm signal is judged according to the number of chicken images with abnormal size parameters and / or weight parameters in a plurality of chicken house images. Thus, the image recognition and depth information are combined, so that the position of the image sensor in the chicken house is no longer constrained, the scene adaptability is improved, at the same time, the volume is corrected by generating the 3D model of each chicken, and the weight parameters are determined based on the conversion relationship between the model unit volume and the weight, thereby improving the body type and weight evaluation accuracy of a large number of chickens in the chicken house. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 The structure schematic diagram of the chicken body type and weight evaluation system based on image recognition provided by the present embodiment.

[0054] Figure 2 The example schematic diagram of the image sensor layout position in the chicken house in the present embodiment;

[0055] Figure 3 The monitoring range schematic diagram of each image sensor in the chicken house in the present embodiment;

[0056] Figure 4 The flow schematic diagram of the chicken body type and weight evaluation method based on image recognition provided by the present embodiment.

[0057] REFERENCE NUMERALS:

[0058] 10-image acquisition and processing module; 20-size calculation module; 30-weight estimation module; 40-result feedback module. DETAILED DESCRIPTION

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] Example 1:

[0061] like Figure 1 As shown, a chicken body size and weight assessment system based on image recognition includes: an image acquisition and processing module 10, configured to acquire several chicken coop images of a target chicken coop, identify and segment several chicken images in each chicken coop image, and extract depth information from the chicken coop images; a size calculation module 20, configured to calculate the size parameters of each chicken in each chicken coop image based on the segmented chicken images and the depth information; wherein the size parameters include 2D features and 3D features of each chicken; a weight estimation module 30, configured to generate a 3D model of each chicken based on the 2D features and 3D features of each chicken in each chicken coop image, correct the size parameters of each chicken, and calculate the weight parameters of each chicken; and a result feedback module 40, configured to determine whether there are more than a preset number of chicken images with abnormal size parameters and / or weight parameters in each chicken coop image acquired at each time moment, and if so, to provide an assessment alarm message.

[0062] It should be noted that existing methods for detecting the body size or weight of animals and plants suffer from low accuracy, high computational load, and high equipment cost when detecting large numbers of animals and plants. To address these issues, this embodiment extracts several chicken images from a chicken coop image. Using the depth information of the chicken coop images, it calculates the 2D and 3D features of the chicken in each image. Then, it generates a 3D model of each chicken using the 2D and 3D features and calculates its weight parameters. Finally, based on the number of chicken images with abnormal size and / or weight parameters in the chicken coop images, it determines whether to issue an evaluation alarm signal. Therefore, by combining image recognition with depth information, the position of the image sensor in the chicken coop is no longer constrained, improving scene adaptability. Simultaneously, by generating a 3D model of each chicken to correct its volume and determining the weight parameters based on the conversion relationship between model unit volume and weight, the accuracy of body size and weight assessment for large numbers of chickens in the chicken coop is improved.

[0063] In a preferred embodiment, the image acquisition and processing module specifically comprises: an image acquisition unit; wherein the image acquisition unit is configured to acquire the images of the target henhouse collected by the plurality of image sensors arranged at different positions of the target henhouse; wherein the images of the target henhouse collected by the plurality of image acquisition units are configured as a plurality of images covering different ranges of the target henhouse.

[0064] Further, the image acquisition and processing module specifically comprises: an image segmentation unit; wherein the image segmentation unit is configured to identify and segment the chicken images in the plurality of images of the target henhouse using the PIDNet model trained based on the chicken images, to obtain the chicken image detection results of each image; wherein the chicken image detection results are specifically: , ; wherein, represents the chicken image detection results of the Mth image of the target henhouse, represents the mask of the Nth chicken detected in the Mth image of the target henhouse.

[0065] Further, the image acquisition and processing module specifically comprises: a depth estimation unit; wherein the depth estimation unit is configured to extract the depth information of each image of the target henhouse using the SideRT model, and store the depth information in the form of a depth map.

[0066] In this embodiment, the acquisition of the images of the target henhouse is performed by the image sensors arranged at different positions of the target henhouse and covering different ranges of the target henhouse. After a plurality of images of the target henhouse are acquired, the PIDNet (i.e., a real-time semantic segmentation model) is used to identify the chickens in the plurality of images of the target henhouse (as shown in Figure 2 In this embodiment, the 4 images of the target henhouse collected by the 4 image sensors A, B, C and D arranged at the midpoints of the 4 sides of the rectangular henhouse , , , ) are identified, and the chicken image segmentation is performed according to the identification detection results, to obtain a series of detection results under each image , , , , represents the detection results of , represents the detection results of ​The detected mask of the first chicken is used to extract the depth information of each chicken coop image using the SideRT model (i.e., a Transformer structure model based on single image depth estimation) after that, so as to calculate the size of the chickens in the target chicken coop and provide estimated data support. Compared with the traditional weight detection realized by image information only, the precision is higher. At the same time, the embodiment adopts the method of collecting internal images from the region edge of the target chicken coop, so that the placement position of the image sensor is no longer constrained, and the scene adaptability is improved (the existing method needs to make the image sensor shooting angle vertical downward, so the visual angle shooting area is limited).

[0067] In a preferred embodiment, the size calculation module specifically comprises a 2D feature extraction unit; wherein the 2D feature extraction unit is configured to scale the information extracted based on the mask generated based on each chicken image based on the depth information, and calculate the 2D features of each chicken image; wherein the 2D features include the projection area S, the width W of the chicken image and the perimeter L of the chicken image, and the expression is specifically:

[0068] ;

[0069] ;

[0070] ;

[0071] wherein, is the area of the mask, is the width of the mask, is the perimeter of the mask, D is the average depth value of the position of the mask, and k is a constant for adjusting the nonlinearity, and b are proportion factors for adjusting the influence of depth on area, and c is an offset for preventing the area from being scaled too large when the depth value is too small.

[0072] Further, the size calculation module specifically comprises a 3D feature extraction unit; wherein the 3D feature extraction unit is configured to estimate the 3D features of each chicken image according to the depth information of each chicken coop image and the projection area S; wherein the 3D features include the volume V, and the expression is specifically:

[0073] ;

[0074] wherein V represents the estimated volume, represents the average depth value of the peripheral line of the mask of each chicken image, represents the depth value of the i-th pixel corresponding to the mask of each chicken image.

[0075] In this embodiment, the 2D feature extraction unit uses the filled mask as the information extraction target and scales it using depth information, thereby calculating the 2D features of each chicken image. The 2D features include the projected area S, the width W of the chicken image, and the perimeter L of the chicken image. The projected area is the area of the mask scaled by depth information, the width of the chicken is the maximum width of the mask scaled by depth, and the perimeter of the chicken is the length of the outer line of the mask scaled by depth. After obtaining the 2D features of the chicken image, the volume of the chicken is estimated by the depth information of the corresponding region of the mask, and the 3D features of the chicken image, i.e., the volume parameter of the chicken, are obtained. Compared with the traditional weight detection based only on image information, the accuracy of the body shape and weight evaluation of a large number of chickens in the chicken coop is improved.

[0076] In a preferred embodiment, the weight estimation module specifically includes: a 3D model generation unit configured to generate a feature data set based on the 2D features and the 3D features of each chicken in each chicken coop image, input the feature data set into a neural radiance field model to obtain a feature latent representation, and input the feature latent representation into a CLIP-Mesh model to obtain a 3D model of each chicken; a volume correction unit configured to correct the 3D features of each chicken to the volume corresponding to the 3D model of each chicken according to the 3D model of each chicken.

[0077] Further, the weight estimation module specifically includes: a weight calculation unit; wherein the weight calculation unit is configured to calculate the weight parameter of each chicken according to the 3D model of each chicken and a preset ratio of weight to unit volume.

[0078] In this embodiment, considering the use of image recognition combined with depth information to measure and evaluate the body shape and weight of chickens in the target chicken coop, although it has certain advantages compared to traditional techniques, the measurement accuracy is still not very high. Therefore, after calculating the 2D features and 3D features of the chicken image based on image recognition and depth information technology, the latent representation of chicken-related information is generated using NeRF, and then the corrected volume is obtained using CLIP-Mesh and the weight is calculated by a preset rule. Specifically, first, the 3D model generation unit obtains the projected area S, the width W of the chicken, the perimeter L of the chicken, and the volume V, which form the data μ, and then the data μ is input into the neural radiance field model (such as the MipNeRF model) to obtain the latent representation: , (θ is the angle between the image sensor and the horizontal plane, and x is the latent representation), after which the latent representation x and the text "Chicken" of "chicken" are input into the CLIP-Mesh (a model structure based on CLIP and Mesh neural networks) to obtain a 3D model of the chicken corresponding to the latent representation x, finally, the volume parameters of the 3D model are used to replace the volume parameters calculated by the 2D features and the 3D features before, and the weight parameters of each chicken are calculated in the form of "weight / unit volume" according to the 3D model, thereby providing a chicken body type and weight calculation method with high accuracy.

[0079] In a preferred embodiment, the result feedback module specifically comprises: an abnormality judgment unit configured to judge whether the size parameter and / or the weight parameter of each chicken is an abnormal value, and if so, the chicken is determined as an abnormal target; an evaluation alarm unit configured to judge whether the number of chickens determined as abnormal targets in each chicken house image collected at each time exceeds a preset number, and if so, an evaluation alarm signal is fed back.

[0080] In this embodiment, considering that in actual application, there are a large number of chickens in each area of the target chicken house, if full coverage monitoring of the target chicken house is performed and each chicken is detected and tracked, a large amount of calculation overhead and high equipment cost will be caused. Therefore, this embodiment discards the idea of full-area full-target tracking detection, and adopts probability estimation based on the number of abnormal targets in a single area, that is, whether the number of chickens with abnormal body type and / or weight (the body type is measured by the 2D feature or the volume parameter corrected by the 3D model, and the weight is obtained by the weight parameter calculated by the 3D model) in the chicken house image collected by each image sensor at each time exceeds a preset number, if so, it indicates that there is a high possibility of abnormal body type and weight in the target chicken house, at this time, the breeder needs to adjust and optimize the feeding strategy. Specifically, as shown in Figure 3 , when there are unqualified chickens exceeding the threshold in the area captured by the image sensor A during use, an alarm information is sent to the user, and the approximate range of the unqualified chickens is given according to the depth information, for example, if the depth value is 3m, the unqualified chickens are located at a horizontal distance of 3m from the image sensor A. Thus, on the basis of ensuring the accuracy of chicken body type and weight monitoring and evaluation, the device operation load and cost budget are reduced.

[0081] Embodiment 2:

[0082] Referring to Figure 4 , Figure 4 A flowchart of a chicken body type and weight evaluation method based on image recognition provided by the embodiment of the present application is shown in the figure.

[0083] As Figure 4As shown, a chicken body shape and weight evaluation method based on image recognition is used in the chicken body shape and weight evaluation system based on image recognition as described above, and comprises the following steps:

[0084] S1: Obtain a plurality of chicken house images of a target chicken house, identify and segment a plurality of chicken images in each chicken house image, and extract depth information in the chicken house image;

[0085] S2: Based on the segmented plurality of chicken images and the depth information, the size parameters of each chicken in each chicken house image are calculated; wherein the size parameters include 2D features and 3D features of each chicken;

[0086] S3: Based on the 2D features and 3D features of each chicken in each chicken house image, a 3D model of each chicken is generated, the size parameters of each chicken are corrected, and the weight parameters of each chicken are calculated;

[0087] S4: Determine whether there are chicken images with abnormal size parameters and / or weight parameters exceeding a preset number in each chicken house image collected at each time, and if so, feedback evaluation warning information.

[0088] The embodiment provides a chicken body shape and weight evaluation method based on image recognition, which extracts a plurality of chicken images from a chicken house image, calculates the 2D features and 3D features of the corresponding chicken of each chicken image using the depth information of the chicken house image, then generates a 3D model of each chicken using the 2D features and 3D features and calculates the weight parameters, and finally determines whether to feedback an evaluation warning signal according to the number of chicken images with abnormal size parameters and / or weight parameters in a plurality of chicken house images. Therefore, the combination of image recognition and depth information makes the position of the image sensor in the chicken house no longer constrained, improving the scene adaptability. At the same time, the volume is corrected by generating a 3D model of each chicken, and the weight parameters are determined based on the conversion relationship between the model unit volume and the weight, improving the body shape and weight evaluation accuracy of a large number of chickens in the chicken house.

[0089] The specific implementation of the chicken body shape and weight evaluation method based on image recognition of the present application is basically the same as the above-mentioned chicken body shape and weight evaluation system based on image recognition, and will not be repeated here.

[0090] In the description of the embodiments of the present application, it needs to be understood that the terms "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "center", "top", "bottom", "top", "bottom", "inner", "outer", "inner side", "outer side" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. Among them, "inside" refers to the inside or enclosed area or space. "Periphery" refers to the area around a particular component or a particular area.

[0091] In the description of the embodiments of the present application, the terms "first", "second", "third", "fourth" are only used for description purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second", "third", "fourth" can be explicitly or implicitly include one or more of the features. In the description of the present application, unless otherwise specified, the meaning of "a plurality of" is two or more.

[0092] In the description of the embodiments of the present application, it needs to be noted that, unless otherwise explicitly specified and limited, the terms "mounting", "connecting", "connecting", "assembling" should be understood in a broad sense, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be directly connected, or indirectly connected through an intermediate medium; it can be the communication inside two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0093] In the description of the embodiments of the present application, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0094] In the description of the embodiments of the present application, it needs to be understood that "-" and "~" represent the range between two values, and the range includes the endpoints. For example: "A-B" represents a range greater than or equal to A and less than or equal to B. "A~B" represents a range greater than or equal to A and less than or equal to B.

[0095] In the description of the embodiments of the present application, the term "and / or" herein is only a description of the association relationship between the associated objects, which means that there can be three relationships, for example, A and / or B can represent the three cases of A alone, A and B together, and B alone. In addition, the character " / " herein generally represents that the front and rear associated objects have an "or" relationship.

[0096] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications and adaptions can be effected without departing from the spirit and scope of the present application, which is not limited to the exact construction and arrangement described. It is intended, therefore, to cover all modifications and adaptions that fall within the scope of the claims and their equivalents.

Claims

1. A chicken body shape and weight assessment system based on image recognition, characterized in that, The system includes: The image acquisition and processing module is configured to acquire several images of the target chicken coop, identify and segment several chicken images in each chicken coop image, and extract depth information from the chicken coop images. The size calculation module is configured to calculate the size parameters of each chicken in each chicken coop image based on several chicken images obtained from segmentation and the depth information; wherein the size parameters include the 2D features and 3D features of each chicken. The weight estimation module is configured to generate a 3D model of each chicken based on the 2D and 3D features of each chicken in each chicken coop image, correct the size parameters of each chicken, and calculate the weight parameters of each chicken. The result feedback module is configured to determine whether there are more than a preset number of chicken images with abnormal size parameters and / or weight parameters in each chicken coop image collected at each time moment; if so, it will provide feedback evaluation alarm information. The weight estimation module specifically includes: A 3D model generation unit is configured to generate a feature data set based on the 2D and 3D features of each chicken in each chicken coop image, input the feature data set into a neural radiation field model to obtain a latent feature representation, and input the latent feature representation into a CLIP-Mesh model to obtain a 3D model of each chicken. A volume correction unit is configured to correct the 3D features of each chicken to the volume corresponding to the 3D model of each chicken, based on the 3D model of each chicken. The image acquisition and processing module specifically includes: an image acquisition unit; the image acquisition unit is configured to acquire chicken house images acquired by four image sensors located at the midpoints of the four sides of the target chicken house; the chicken house images acquired by the four image acquisition units are configured as four images covering different areas of the target chicken house; The image acquisition and processing module specifically includes: an image segmentation unit; the image segmentation unit is configured to use a PIDNet model trained on chicken images to identify and segment chicken images in four images of the target chicken coop, obtaining a chicken image detection result for each image; the chicken image detection result is specifically: , … ; This represents the chicken image detection result for the Mth chicken coop image. The mask represents the Nth chicken detected in the Mth chicken coop image.

2. The chicken body shape and weight assessment system based on image recognition according to claim 1, characterized in that, The image acquisition and processing module specifically includes: Depth estimation unit; The depth estimation unit is configured to extract depth information for each chicken coop image using the SideRT model and store it in the form of a depth map.

3. The chicken body shape and weight assessment system based on image recognition according to claim 1, characterized in that, The dimension calculation module specifically includes: 2D feature extraction unit; The 2D feature extraction unit is configured to scale the information extraction target generated by the mask of each chicken image based on the depth information, and calculate the 2D features of each chicken image. The 2D features include: the projected area S, the width W of the chicken image, and the perimeter L of the chicken image, specifically expressed as: ； ； ； in, The area of ​​the mask. The width of the mask. Let be the perimeter of the mask, D be the average depth at the location of the mask, and k be a constant that adjusts the degree of non-linearity. b is a scaling factor used to adjust the effect of depth on area, and c is an offset used to prevent excessive area scaling when the depth value is too small.

4. The chicken body shape and weight assessment system based on image recognition according to claim 3, characterized in that, The dimension calculation module specifically includes: 3D feature extraction unit; The 3D feature extraction unit is configured to estimate the 3D features of each chicken coop image based on the depth information of each chicken coop image and the projected area S. The 3D features include: volume V, specifically expressed as: ； Where V represents the estimated volume, This represents the average depth value of the mask outline for each chicken image. This represents the depth value of the i-th pixel at the corresponding position in the mask of each chicken image.

5. The image recognition-based chicken body shape and weight assessment system according to claim 4, characterized in that, The weight estimation module specifically includes: Weight calculation unit; The weight calculation unit is configured to calculate the weight parameters of each chicken based on the 3D model of each chicken and a preset ratio of weight to unit volume.

6. The image recognition-based chicken body shape and weight assessment system according to claim 5, characterized in that, The result feedback module specifically includes: An anomaly detection unit is configured to determine whether the size parameter and / or weight parameter of each chicken is an abnormal value; if so, the chicken is determined to be an abnormal target. An evaluation alarm unit is configured to determine whether the number of chickens identified as abnormal targets in each chicken coop image collected at each time exceeds a preset number; if so, an evaluation alarm signal is fed back.

7. A method for assessing chicken body size and weight based on image recognition, characterized in that, The method includes: S1: Acquire several images of the target chicken coop, identify and segment several chicken images in each chicken coop image, and extract the depth information from the chicken coop images; S2: Based on the segmented chicken images and the depth information, the size parameters of each chicken in each chicken coop image are calculated; wherein, the size parameters include the 2D features and 3D features of each chicken; S3: Based on the 2D and 3D features of each chicken in each chicken coop image, generate a 3D model of each chicken, correct the size parameters of each chicken, and calculate the weight parameters of each chicken. S4: Determine whether there are more than a preset number of chicken images with abnormal size parameters and / or weight parameters in each chicken coop image collected at each time moment. If so, provide an evaluation alarm message. The method further includes: generating a feature data set based on the 2D and 3D features of each chicken in each chicken coop image; inputting the feature data set into a neural radiation field model to obtain a latent feature representation; inputting the latent feature representation into a CLIP-Mesh model to obtain a 3D model of each chicken; and correcting the 3D features of each chicken to the volume corresponding to the 3D model based on the 3D model of each chicken. The method further includes: acquiring chicken house images collected by four image sensors positioned at the midpoints of the four sides of the target chicken house; wherein the chicken house images collected by the four image sensors are configured as four images covering different areas of the target chicken house; and, The PIDNet model, trained on chicken images, was used to identify and segment chicken images from four images of the target chicken coop, yielding chicken image detection results for each image. Specifically, the chicken image detection results are as follows: , … ; This represents the chicken image detection result for the Mth chicken coop image. The mask represents the Nth chicken detected in the Mth chicken coop image.

Images