A multi-layer caged waterfowl feeding method, system and device based on behavioral feedback

Abstract

This application discloses a multi-layer caged waterfowl feeding method, system, and device based on behavioral feedback, belonging to the field of waterfowl farming technology. The method comprises: a high-definition camera on a mobile feed trough collects waterfowl image data and the QR code label of the breeding cage in real time; a core processing module accurately locates the cage number by decoding the QR code and uses a multi-channel fusion method to segment the waterfowl body area image and the beak area image; based on the segmented beak area and body area, the waterfowl's feeding desire index and body condition score are calculated, and the waterfowl is fed based on the feeding desire index and body condition score; a high-definition camera installed at the terminal of the feed trough running track collects feed trough images; the core processing module analyzes the pollution image of the mobile feed trough and determines the pollution level, and based on the pollution level determination, automatically generates a feed cart cleaning prompt. This method achieves reasonable feeding of waterfowl, reduces feed waste and pollution, and improves farming efficiency and automation level.

Description

Technical Field

[0001] The present application relates to the technical field of waterfowl breeding, and in particular to a method, system and device for feeding multi-layer caged waterfowl based on behavioral feedback. Background Art

[0002] Currently, waterfowl farming is primarily divided into two models: caged and non-caged. Compared to non-caged models, caged waterfowl farming allows for better control of feed intake and feeding schedules, ensuring proper nutrition for the birds while effectively managing feed costs. Furthermore, caged farming facilitates easier cleaning, improves the farming environment, prevents water pollution, strengthens disease prevention and control, and enhances animal welfare.

[0003] With the rapid development of the waterfowl farming industry, traditional waterfowl feeding methods face numerous technical challenges. Currently, large-scale farms commonly use manual feeding and pipe feeding, but both methods have significant drawbacks: manual feeding is labor-intensive, prone to feed contamination, and uneven distribution; pipe feeding is prone to clogging and feed residue. More critically, existing waterfowl feeding equipment is mostly a direct copy of broiler / laying chicken feeding equipment. This design fails to fully consider the unique physiological characteristics of waterfowl, such as their high activity, large size, and distinctive flat-billed feeding pattern. As a result, in practical applications, problems such as feed accumulation, uneven feeding, and trough contamination often occur, seriously impacting feeding efficiency and waterfowl performance. Especially in high-density farming environments, traditional equipment suffers from poor feed delivery, high clogging rates, and uneven distribution, significantly compromising feeding effectiveness. Furthermore, most waterfowl feeding equipment in China currently lacks intelligent control. Many devices still utilize fixed feeding strategies that cannot dynamically adjust based on environmental parameters and waterfowl health, making precise, individualized feeding difficult. This not only exacerbates feed waste and pollution, but also hinders the improvement of farming efficiency. Therefore, the development of intelligent feeding equipment that adapts to the specific needs of waterfowl, improves feeding efficiency, reduces waste, and improves their health has become an urgent need for the industry. Furthermore, cage farming models require high stocking densities and significant economies of scale, necessitating regular daily inspections to check the health and growth of waterfowl. This, in particular, requires repeated feeding and supplemental feeding, a task that is repetitive.

[0004] To address the above-mentioned issues, researchers have designed intelligent feeding methods and developed corresponding devices, but some problems still exist. The prior art has designed a stepped chicken cage feeding device. The stepped design allows a feed cart to feed multiple feed troughs simultaneously, effectively reducing the equipment's footprint on the farm. However, it lacks intelligent food storage and cannot automatically adjust the feed intake, potentially leading to the accumulation of excess feed. Furthermore, the prior art has also designed a guide-rail cage poultry automatic feeding system and control method, which can implement simple food storage detection and dynamically adjust the feed intake. However, it lacks intelligent monitoring and alarm functions for waterfowl feeding desire, health status, and contaminated feed. Summary of the Invention

[0005] In response to the above-mentioned deficiencies in the prior art, the present application provides a multi-layer caged waterfowl feeding method, system and device based on behavioral feedback, so as to dynamically adjust the feed amount according to the feeding desire and health status of the waterfowl, as well as intelligent monitoring and alarm of contaminated feed, thereby reducing feed waste and pollution and significantly improving the breeding efficiency and automation level.

[0006] In order to achieve the above-mentioned invention objectives, the technical solutions adopted in this application are:

[0007] In a first aspect, the present application provides a multi-layer caged waterfowl feeding method based on behavioral feedback, comprising:

[0008] S1: The high-definition camera on the mobile feeder collects real-time image data of waterfowl in front of the feeder and the QR code labels on the breeding cages;

[0009] S2: The core processing module decodes the collected QR code tags to obtain the cage number of the collected waterfowl, and uses a multi-channel fusion method to segment the waterfowl image data into the waterfowl body area image and the beak area image;

[0010] S3: Calculate the feeding desire index of waterfowl based on the segmented mouth area image;

[0011] S4: extracting key image features of the waterfowl's body structure based on the segmented waterfowl body region image, constructing a body condition scoring model, and controlling feeding of the waterfowl by combining the feeding desire index and the body condition score obtained by the body condition scoring model;

[0012] S5: A high-definition camera installed at the terminal of the trough running track collects the trough image. The core processing module analyzes the pollution image and determines the pollution degree of the mobile trough. Based on the pollution degree judgment, it will automatically generate a material cart cleaning prompt.

[0013] Furthermore, the multi-channel fusion method described in S2 is used to segment the waterfowl image data into the waterfowl body and beak, specifically including:

[0014] A1: Convert the collected image data into HSV color space and Lab color space, extract the S channel in the HSV color space to identify the waterfowl body area image; extract the a channel in the Lab color space to extract the waterfowl beak area image;

[0015] A2: Binarize the waterfowl body area image in the S channel and the waterfowl beak area image in the a channel, and use morphological filtering to remove noise from the binarized images;

[0016] A3: The CV model is used to accurately segment the waterfowl body area image, and grayscale morphological processing and Gaussian filtering are used to optimize the edge of the segmented image; the RSF model, local area fitting and narrowband optimization algorithm are used to optimize the extraction of the waterfowl beak boundary in the beak area image.

[0017] Furthermore, the S3 specifically includes:

[0018] S301: Analyzing the opening and closing movements of the waterfowl's mouth and the changes in the distance from the feed trough in consecutive frames based on the segmented mouth area image;

[0019] S302: Using the movement frequency, proximity, and beak speed changes as indicators, construct a waterfowl feeding desire index:

[0020]

[0021] in, is the feeding desire index, is the frequency of mouth opening and closing per unit time, is the distance from the center of mass of the mouth to the front edge of the trough, is the mouth movement speed, in pixels / frame, 、 and is the empirical coefficient.

[0022] Furthermore, the S4 specifically includes:

[0023] S401: extracting a binary mask image of the waterfowl body region image based on the segmented waterfowl body region image;

[0024] S402: extracting key image features of the waterfowl body structure based on the segmented waterfowl body region image;

[0025] S403: performing feature normalization processing on the key image features to obtain a score value of the standardized features of each key image feature;

[0026] S404: The scoring values ​​of each standardized feature are integrated to obtain a body condition scoring model:

[0027]

[0028] in, Scoring waterfowl body condition, Body area The rating value of is the body aspect ratio The rating value of Edge curvature fluctuation The rating value of is the tail height ratio The rating value of is the symmetry coefficient The rating value of is the weight coefficient;

[0029] S405: Control feeding of waterfowl by combining the feeding desire index and the body condition score obtained by the body condition score model: and , extend the time the feed cart stays and increase feeding; if and , do not feed, start the RDHL-DETR model to detect dead waterfowl; if and , shorten the material car residence time, among which, is the lowest threshold, is the highest threshold, is the minimum weight threshold, The maximum weight threshold.

[0030] Furthermore, the RDHL-DETR model specifically includes:

[0031] The standard convolution module in the original DETR model is replaced with the RepConv module; the multi-head self-attention mechanism in the traditional AIFI module is replaced with a deformable multi-head attention mechanism; and a local feature fusion module is introduced.

[0032] Furthermore, the RDHL-DETR model is trained, specifically including:

[0033] B1: Collect image data of waterfowl deaths in different scenarios and annotate the image data using annotation tools;

[0034] B2: Divide the labeled image data into training set, validation set and test set;

[0035] B3: Use the training set to train the RDHL-DETR model;

[0036] B4: Use the validation set to evaluate the trained RDHL-DETR model and save the optimal network training weights based on the validation results;

[0037] B5: Use the test set to evaluate the generalization ability of the RDHL-DETR model and ensure the stability of the RDHL-DETR model.

[0038] Furthermore, the S5 specifically includes:

[0039] S501: A high-definition camera installed at the terminal of the trough running track collects the trough image;

[0040] S502: performing brightness equalization and geometric distortion correction processing on the collected trough image;

[0041] S503: extracting a main visual area image including the trough from the processed trough image according to a preset trough structure template;

[0042] S504: Detecting the contaminated area of ​​the main visual area image of the trough using a fusion algorithm of color space, edge structure, and texture complexity to obtain a contamination score index;

[0043] S505: Perform pollution judgment based on the pollution score index and the set pollution threshold: If the pollution score index is greater than the pollution threshold, the material tank is judged to be seriously polluted, and a manual or automatic cleaning operation is required. A pollution alarm record is generated including the material tank number, detection time, pollution score value and image screenshot.

[0044] Furthermore, in S504, the contaminated area is detected on the main visual area image of the short trough using a fusion algorithm of color space, edge structure and texture complexity to obtain a contamination score index, which specifically includes:

[0045] C1: Convert the main visual area image into Lab color space;

[0046] C2: Enhance the abnormal saturation areas in the image in Lab color space;

[0047] C3: Utilizes the local gradient information of the image to extract the trough edge and patch outline, and uses color features to construct a pollution mask image;

[0048] C4: Based on the pollution mask image, a pollution scoring function is constructed to quantify the pollution degree. The pollution scoring function is:

[0049]

[0050] in, is the pollution score index, is the area of ​​the detected contaminated area, is the total area of ​​the trough, is the average saturation concentration in the polluted area, is the local information entropy of the polluted area, 、 and An empirically selected coefficient that is set based on the proportion of manual needs.

[0051] In a second aspect, the present application provides a multi-layer caged waterfowl feeding system based on behavioral feedback, comprising:

[0052] The data acquisition device includes a camera and a wireless transmission module. The camera collects waterfowl image data, QR code labels on breeding cages, and feed trough images, and then transmits them to the edge image processing module through the wireless transmission module;

[0053] The edge image processing module deploys a multi-channel color space image segmentation model and an image processing model, and transmits the processed image to the intelligent recognition module;

[0054] The intelligent recognition module deploys a feeding desire assessment module, a waterfowl body condition scoring module, an RDHL-DETR module, and a feed trough contamination identification module. The waterfowl body area images and waterfowl beak area images segmented by the edge image processing module are transmitted to the feeding desire assessment module and the waterfowl body condition scoring module for calculation, and the calculation results are transmitted to the intelligent feeding control module as feeding instructions; the RDHL-DETR module identifies whether the waterfowl is dead based on the output of the feeding desire assessment module; the feed trough image processed by the edge image processing module is used as the input of the feed trough contamination identification module, and the feed trough contamination identification module transmits the identification results to the waterfowl intelligent feeding platform;

[0055] An intelligent feeding control module, which controls the residence time of the mobile short feed trough according to feeding instructions;

[0056] The waterfowl intelligent feeding platform collects and analyzes data related to waterfowl feeding behavior, waterfowl mortality judgment data, and feed trough pollution identification data, and generates relevant reports for display.

[0057] In a third aspect, the present application provides a multi-layer caged waterfowl feeding device based on behavioral feedback, comprising:

[0058] Several breeding cages, egg conveyor belts, QR code labels, mobile short feed troughs, waterfowl inspection HD cameras, feed trough pollution detection HD cameras and a core processing module; multiple waterfowl inspection HD cameras are mounted on the mobile short feed troughs, the feed trough pollution detection HD cameras are installed at the feed trough inspection terminal, and the QR code labels are located at the front ends of the several breeding cages.

[0059] The beneficial effects of this application are:

[0060] The present application provides a multi-layer caged waterfowl feeding method, system and device based on behavioral feedback, which dynamically regulates the feed amount by adjusting the waterfowl's feeding desire and body condition, thereby achieving reasonable feeding of waterfowl and reducing feed waste. In addition, feed contamination is reduced by identifying and cleaning feed trough contamination, significantly improving breeding efficiency and automation level. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other embodiments can also be obtained based on these drawings.

[0062] Figure 1 A schematic flow chart of a multi-layer caged waterfowl feeding method based on behavioral feedback provided in an embodiment of the present application.

[0063] Figure 2 A schematic diagram of a multi-layer caged waterfowl feeding system based on behavioral feedback provided in an embodiment of the present application.

[0064] Figure 3 A schematic structural diagram of a multi-layer caged waterfowl feeding device based on behavioral feedback provided in an embodiment of the present application.

[0065] Among them: 1-breeding cage, 2-egg conveyor belt, 3-QR code label, 4-mobile short feed trough, 5-waterfowl inspection HD camera, 6-feed trough pollution detection HD camera, 7-core processing module. DETAILED DESCRIPTION

[0066] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field based on this application are within the scope of protection of this application.

[0067] Example 1:

[0068] The present application embodiment provides a multi-layer caged waterfowl feeding method based on behavioral feedback, which can be found in Figure 1 , Figure 1 The figure shows a flow chart of a multi-layer caged waterfowl feeding method based on behavioral feedback provided by an embodiment of the present application, including:

[0069] S1: The high-definition camera on the mobile feed trough collects image data of waterfowl in front of the feed trough and the QR code label on the breeding cage in real time.

[0070] In one embodiment of the present application, a camera is fixedly mounted at the front end of a feed cart. As the feed cart moves horizontally between cage positions along a track, image data of waterfowl in front of each feed trough is collected in real time. Image collection and individual positioning are bound synchronously with the QR code encoding of each cage position, eliminating the need for multiple camera deployments, reducing costs, and improving the traceability of waterfowl behavior tracking and physical condition data.

[0071] S2: The core processing module decodes the collected QR code label to obtain the cage number of the collected waterfowl, and uses a multi-channel fusion method to segment the waterfowl image data into the waterfowl body area image and the mouth area image.

[0072] Furthermore, the multi-channel fusion method described in S2 is used to segment the waterfowl image data into the waterfowl body and beak, specifically including:

[0073] A1: Color channel processing. The acquired image data is converted into HSV and Lab color spaces. The S channel is extracted from the HSV color space to identify the body region of the waterfowl. The a channel is extracted from the Lab color space to specifically extract the beak region. This method effectively distinguishes different regions and prepares for subsequent fine segmentation.

[0074] A2: Coarse segmentation. After extracting the color channels, the system binarizes the images in the S and a channels. The image is thresholded using the Otsu method, with a threshold of 0.6. Morphological filtering (dilation and erosion operations) is then used to remove noise while preserving the main features of the image. Next, the largest connected region in the image is extracted as the initial active contour, preparing for the next step of refined segmentation.

[0075] A3: Fine Segmentation. Building on the coarse segmentation, the system uses the CV model to precisely segment the waterfowl's body area, combining grayscale morphological processing and Gaussian filtering to optimize edges, ensuring even more accurate segmentation results. Furthermore, for the waterfowl's beak area, the system employs the RSF model, combining local region fitting with a narrowband optimization algorithm to improve the accuracy of beak boundary extraction, providing clear image data support for accurately determining the waterfowl's feeding behavior and health status.

[0076] It can be understood that the introduction of HSV-S and Lab-a color channels for body and mouth recognition, respectively, effectively improves the edge segmentation accuracy and system recognition robustness under complex backgrounds.

[0077] S3: Calculate the feeding desire index of the waterfowl based on the segmented mouth area image.

[0078] Specifically:

[0079] S301: After the mouth area is segmented, the mouth opening and closing movements and the changes in the distance from the trough are tracked in consecutive frames;

[0080] S302: Setting the movement frequency, proximity, and mouth speed change as indicators, integrating and evaluating the feeding desire, the waterfowl feeding desire index is:

[0081]

[0082] in, is the feeding desire index, is the frequency of mouth opening and closing per unit time, is the distance from the center of mass of the mouth to the front edge of the trough, is the mouth movement speed, in pixels / frame, 、 and is the empirical coefficient, which is adjusted according to the training set.

[0083] It can be understood that by quantifying feeding behavior based on parameters such as mouth opening and closing frequency, displacement distance, and proximity to the feed trough, dynamic and accurate identification of individual feeding desire and feeding response control can be achieved.

[0084] S4: Based on the segmented waterfowl body area image, key image features of the waterfowl body structure are extracted, a body condition scoring model is constructed, and the feeding of the waterfowl is controlled by combining the feeding desire index and the body condition score obtained by the body condition scoring model.

[0085] Specifically:

[0086] S401: After the body image is acquired and the image segmentation is completed, a binary mask image of the waterfowl body region is extracted as input for body condition scoring. The body region should have a complete boundary and exclude interference from other waterfowl or occlusions.

[0087] S402: Based on the segmented waterfowl body region image, extract key image features of the waterfowl body structure: body area : refers to the number of pixels within the body contour area, indicating the total volume; body aspect ratio : The width-to-height ratio of the segmented contour is calculated by the minimum enclosing rectangle of the segmented contour; the edge curvature fluctuates : Measure the contour smoothness by changing the first-order derivative of the edge curvature; tail height ratio : Calculate the ratio of the tail area height to the total body height; symmetry coefficient : Assess bilateral symmetry based on the difference in area on both sides of the image axis;

[0088] S403: Feature Normalization Each of the above features is normalized and converted into a score value in the interval [0, 1], which is denoted as:

[0089] , , , ,

[0090] Among them, the normalization method adopts Gaussian function fitting;

[0091] S404: The scoring values ​​of each standardized feature are integrated into a comprehensive score to obtain a body condition scoring model:

[0092]

[0093] in, The body condition score of waterfowl is , Body area The rating value of is the body aspect ratio The rating value of Edge curvature fluctuation The rating value of is the tail height ratio The rating value of is the symmetry coefficient The rating value of is the weight coefficient, obtained based on expert experience;

[0094] S405: Control feeding of waterfowl by combining the feeding desire index and the body condition score obtained by the body condition score model: and , indicating that the waterfowl has a strong desire to eat and is in good physical condition, extend the time the feed cart stays and increase the feeding time; if and , not feeding, indicating that the waterfowl has health risks such as abnormal body shape or malnutrition, the RDHL-DETR model is activated to detect dead waterfowl; if and , indicating that the waterfowl may be overweight, shortening the time the feed truck stays to adjust the feeding time of the waterfowl, among which, is the lowest threshold, is the highest threshold, is the minimum weight threshold, The maximum weight threshold.

[0095] It is understandable that the minimum weight threshold and the maximum weight threshold can be set according to the farm managers, and the waterfowl body condition score is calculated by the structural features in the image (such as area, aspect ratio, edge curvature, etc.), combined with the desire index to achieve automated visual assessment of individual nutritional status and health level.

[0096] Furthermore, the waterfowl mortality recognition mechanism triggered by low feeding desire is automatically judged using the improved end-to-end deep detection model RDHL-DETR, specifically:

[0097] In one embodiment of the present application, (1) in the feature extraction stage, RDHL-DETR replaces the standard convolution module (HGBlock) in the original DETR framework with the RepConv module. This improvement improves the computational efficiency of feature extraction and enhances the generalization ability of the model when processing waterfowl images with different structures. The RepConv module can effectively extract more detailed information while reducing computational costs, providing more efficient and accurate feature data for the subsequent feature fusion stage; (2) in order to further improve the model's processing ability for high-resolution images, RDHL-DETR introduces a deformable multi-head attention mechanism (DMHA) in the traditional AIFI module, replacing the original traditional multi-head self-attention (MHSA) mechanism. DMHA can dynamically perceive the target deformation and position offset in the image, and is particularly robust when detecting the postures of dead waterfowl such as "belly up" or "head drooping". This improvement significantly improves the recognition accuracy of the model in occluded or highly dynamic scenes; (3) when processing complex backgrounds or partial occlusions, traditional cross-scale feature fusion methods (such as CCFF) may cause the loss of small target information. To overcome this problem, RDHL-DETR introduced a local feature fusion module (LFFM). This module selectively expands the receptive field and adaptively adjusts the model's perception of the focus area, effectively extracting detailed features of dead waterfowl, such as "fluffy feathers" and "straight legs", significantly improving the model's recognition accuracy in complex environments.

[0098] As can be seen, to improve the accuracy and speed of dead waterfowl detection in complex environments, an improved DETR (Detection Transformer) model framework, RDHL-DETR, was proposed. This model incorporates several technical improvements based on DETR, effectively addressing the impact of environmental factors such as occlusion, uneven lighting, and complex backgrounds on detection accuracy. By integrating RepConv, a deformable multi-head attention mechanism (DMHA), and a local feature fusion module (LFFM), RDHL-DETR addresses the low recognition accuracy of traditional models under conditions of occlusion, complex lighting, and varying postures. This significantly improves the performance and robustness of the detection model, particularly for identifying small targets (such as dead waterfowl) in dynamic environments.

[0099] Furthermore, the RDHL-DETR model is trained, specifically including:

[0100] First, image data containing waterfowl deaths in different scenarios were collected, and the images were accurately labeled using the Labelme annotation tool; then, the dataset was divided into training, validation, and test sets in a ratio of 7:2:1. To ensure the training effect, the size of all images was uniformly adjusted to 640x640 pixels. During the training process, RDHL-DETR was used as the model architecture and 200 rounds of training were performed. After each round of training, the model was evaluated using the validation set, and the best network training weights were saved based on the verification results. Finally, the generalization ability of the model was evaluated using the test set to ensure that the model performed stably in practical applications and avoid overfitting. This training process helps to improve the accuracy and robustness of the model in stain detection tasks. At the same time, through the evaluation of the validation set and test set, it is ensured that the model has good generalization ability and can work stably in different waterfowl death scenarios.

[0101] S5: A high-definition camera installed at the terminal of the trough running track collects the trough image. The core processing module analyzes the pollution image and determines the pollution degree of the mobile trough. Based on the pollution degree judgment, it will automatically generate a material cart cleaning prompt.

[0102] Specifically:

[0103] S501: The camera is fixedly installed at the terminal detection position of the feed cart running track. When the feed cart completes the entire feeding process, each feed trough is aligned with the terminal camera in turn, and accurate image acquisition is achieved through motor or guide rail positioning;

[0104] S502: The captured image is first processed for brightness equalization and geometric distortion correction to remove visual interference caused by illumination fluctuations or angle errors;

[0105] S503: Based on the preset trough structure template, the main visual region (ROI) containing the short trough is extracted from the image for subsequent contamination analysis. The region extraction can be completed by combining structure projection and grayscale matching technology to ensure the accuracy and consistency of positioning;

[0106] S504: After obtaining the trough ROI area image, the system uses a fusion algorithm based on color space, edge structure and texture complexity to detect the contaminated area. Specifically:

[0107] First, in the Lab color space, we enhance the image's saturation abnormalities. By setting a threshold, we extract dark, brown, or blackish areas as candidate contamination pixels. Secondly, we use the image's local gradient information to extract the trough's edges and patch outlines, combining this with color features to construct a contamination mask image. Finally, we quantify the degree of contamination using the following contamination scoring function:

[0108]

[0109] in, is the pollution score index, is the area of ​​the detected contaminated area, is the total area of ​​the trough, is the average saturation concentration (color concentration index) of the polluted area, is the local information entropy of the polluted area, reflecting the texture complexity and distribution discreteness). 、 and This is an empirically selected coefficient based on manual adjustments to balance the impact of various indicators on the final score. This score comprehensively considers the extent of contamination, color characteristics, and distribution structure, effectively distinguishing between different levels of contamination, such as light residue and heavy manure coverage.

[0110] S505: Pollution warning judgment and prompt. When the calculated pollution score Greater than the set threshold If the contamination score is 0.65 (e.g., 0.65), the feed trough is considered severely contaminated and requires manual or automatic cleaning. At this point, the waterfowl smart feeding system automatically generates a contamination alarm record, including the feed trough number, detection time, contamination score, and image screenshots. This alarm information is then sent via a pop-up window on the management terminal, push notifications to the app, or synchronized with the backend management system, alerting farmers to promptly address the issue. The system also records the historical contamination score curve for each feed trough, which can be used to inform subsequent decisions on feed trough structure improvements, position adjustments, or cleaning frequency optimization.

[0111] Example 2:

[0112] The present application provides a multi-layer caged waterfowl feeding system based on behavioral feedback, which is as follows: Figure 2 As shown, Figure 2 A schematic diagram of a multi-layer caged waterfowl feeding system based on behavioral feedback provided in an embodiment of the present application includes:

[0113] The data acquisition device includes a camera and a wireless transmission module. The camera collects waterfowl image data, QR code labels on breeding cages, and feed trough images, and then transmits them to the edge image processing module through the wireless transmission module;

[0114] The edge image processing module deploys a multi-channel color space image segmentation model and an image processing model, and transmits the processed image to the intelligent recognition module;

[0115] The intelligent recognition module deploys a feeding desire assessment module, a waterfowl body condition scoring module, an RDHL-DETR module, and a feed trough contamination identification module. The waterfowl body area images and waterfowl beak area images segmented by the edge image processing module are transmitted to the feeding desire assessment module and the waterfowl body condition scoring module for calculation, and the calculation results are transmitted to the intelligent feeding control module as feeding instructions; the RDHL-DETR module identifies whether the waterfowl is dead based on the output of the feeding desire assessment module; the feed trough image processed by the edge image processing module is used as the input of the feed trough contamination identification module, and the feed trough contamination identification module transmits the identification results to the waterfowl intelligent feeding platform;

[0116] An intelligent feeding control module, which controls the residence time of the mobile short feed trough according to feeding instructions;

[0117] The waterfowl intelligent feeding platform collects and analyzes data related to waterfowl feeding behavior, waterfowl mortality judgment data, and feed trough pollution identification data, and generates relevant reports for display.

[0118] In one embodiment of the present application, the waterfowl intelligent feeding platform will record the historical pollution score change curve of each feed trough, which can be used for subsequent feed trough structure improvement, position adjustment or cleaning frequency optimization decision-making; and the waterfowl intelligent feeding platform relies on core technologies such as camera visual recognition, image analysis, and individual behavior modeling to record the feeding time, feeding desire index, cage position, feed cart stay time, actual feeding amount, feed remaining, feeding behavior image and body condition score of each waterfowl, and form an individual waterfowl feeding behavior report; form a group waterfowl feeding performance report to display the overall feeding trend, average feed amount, residual feed rate and body condition score distribution of waterfowl in each area, assist in optimizing feeding strategies and feed cart operation paths, and improve group feeding efficiency and resource utilization; when the feeding desire index or body condition score continues to be abnormal, the system calls the death recognition model to automatically judge and alarm, and generates a death report containing location information, recognition screenshots, confidence levels and other fields to assist in accurate cleaning Management and health control; when the feed truck completes a round of feeding, the terminal camera identifies and scores the pollution of all short feed troughs in turn. If the pollution level is higher than the set threshold, the system automatically generates a pollution warning report to prompt the administrator to arrange cleaning and maintenance in time to avoid feed waste and disease transmission; users are supported to feedback problems such as abnormal feeding, missed death, feed trough pollution or equipment failure through the mini program. The waterfowl intelligent feeding platform automatically summarizes and generates processing reports to improve the feedback closed-loop efficiency and service quality; provides operation management functions: provides feeding desire and body condition score threshold setting, equipment operation status monitoring, regional feeding parameter configuration, feed trough pollution score management and individual file management, etc., supports zoning, grouping and stage-by-stage refined breeding control; supports graphical display of multi-dimensional data, including feeding behavior, body condition changes, death warnings, pollution scores, equipment operation status and user feedback processing, etc., forming a large visual data screen to achieve "one picture to control the entire operation situation".

[0119] Example 3:

[0120] The present application provides a multi-layer caged waterfowl feeding device based on behavioral feedback, which is as follows: Figure 3 As shown, Figure 3 A schematic diagram of a multi-layer caged waterfowl feeding device based on behavioral feedback provided in an embodiment of the present application includes:

[0121] Several breeding cages 1, egg conveyor belts 2, QR code labels 3, mobile short feed troughs 4, waterfowl inspection HD cameras 5, feed trough pollution detection HD cameras 6 and a core processing module 7; multiple waterfowl inspection HD cameras 5 are mounted on the mobile short feed troughs 4, the feed trough pollution detection HD camera 6 is installed at the feed trough inspection terminal, and the QR code labels 3 are located at the front ends of the several breeding cages 1.

[0122] The present application provides a multi-layer caged waterfowl feeding method, system and device based on behavioral feedback. By integrating a camera at the front end of a mobile short feed trough and combining it with QR code positioning, it realizes the simultaneous binding of image acquisition and individual information, eliminating the need for multiple camera deployment, reducing costs while improving the traceability of waterfowl behavior tracking and body condition data; introducing HSV-S and Lab-a color channels for body and mouth recognition, respectively, effectively improving edge segmentation accuracy and system recognition robustness under complex backgrounds; combining waterfowl body condition index and feeding desire to accurately control feed feeding amount; at the same time, automatically enabling the death recognition model when the feeding desire continues to be abnormal, and using the improved DETR structure combined with the attention mechanism to achieve high-accuracy recognition of dead waterfowl, significantly reducing missed detections and false alarms. In addition, after the inspection of the feed truck, the terminal camera is used to detect the degree of contamination in the feed trough. If the contamination score exceeds the threshold, a cleaning reminder is triggered, thereby improving the efficiency of environmental sanitation management.

[0123] It should be noted that those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of this application, and it should be understood that the scope of protection of this application is not limited to such specific descriptions and embodiments. Those skilled in the art can make various other specific variations and combinations based on the technical teachings disclosed in this application without departing from the essence of this application, and such variations and combinations are still within the scope of protection of this application.

Claims

1. A multi-layer caged waterfowl feeding method based on behavioral feedback, characterized in that: include: S1: The high-definition camera on the mobile feeder collects real-time image data of waterfowl in front of the feeder and the QR code labels on the breeding cages; S2: The core processing module decodes the collected QR code tags to obtain the cage number of the collected waterfowl, and uses a multi-channel fusion method to segment the waterfowl image data into the waterfowl body area image and the beak area image; S3: Calculate the feeding desire index of waterfowl based on the segmented mouth area image; S4: extracting key image features of the waterfowl's body structure based on the segmented waterfowl body region image, constructing a body condition scoring model, and controlling feeding of the waterfowl by combining the feeding desire index and the body condition score obtained by the body condition scoring model; S5: A high-definition camera installed at the end of the trough running track collects trough images. The core processing module analyzes the pollution image and determines the pollution level of the mobile trough. Based on the pollution level judgment, it automatically generates a material cart cleaning prompt. Said S3 specifically includes: S301: Analyzing the opening and closing movements of the waterfowl's mouth and the changes in the distance from the feed trough in consecutive frames based on the segmented mouth area image; S302: Using the movement frequency, proximity, and beak speed changes as indicators, construct a waterfowl feeding desire index: in, is the feeding desire index, is the frequency of mouth opening and closing per unit time, is the distance from the center of mass of the mouth to the front edge of the trough, is the mouth movement speed, in pixels / frame, 、 and is the empirical coefficient; The S4 specifically includes: S401: extracting a binary mask image of the waterfowl body region image based on the segmented waterfowl body region image; S402: extracting key image features of the waterfowl body structure based on the segmented waterfowl body region image; S403: performing feature normalization processing on the key image features to obtain a score value of the standardized features of each key image feature; S404: The scoring values ​​of each standardized feature are integrated to obtain a body condition scoring model: in, Scoring waterfowl body condition, Body area The rating value of is the body aspect ratio The rating value of Edge curvature fluctuation The rating value of is the tail height ratio The rating value of is the symmetry coefficient The rating value of is the weight coefficient; S405: Control feeding of waterfowl by combining the feeding desire index and the body condition score obtained by the body condition score model: and , extend the time the feed cart stays and increase feeding; if and , do not feed, start the RDHL-DETR model to detect dead waterfowl; if and , shorten the material car residence time, among which, is the lowest threshold, is the highest threshold, is the minimum weight threshold, is the highest weight threshold; The RDHL-DETR model specifically includes: The standard convolution module in the original DETR model is replaced with the RepConv module; the multi-head self-attention mechanism in the traditional AIFI module is replaced with a deformable multi-head attention mechanism; and a local feature fusion module is introduced.

2. The multi-layer caged waterfowl feeding method based on behavioral feedback according to claim 1, characterized in that: The method described in S2 uses a multi-channel fusion method to segment the waterfowl image data into the waterfowl body and beak, specifically including: A1: Convert the collected image data into HSV color space and Lab color space, extract the S channel in the HSV color space to identify the waterfowl body area image; extract the a channel in the Lab color space to extract the waterfowl beak area image; A2: Binarize the waterfowl body area image in the S channel and the waterfowl beak area image in the a channel, and use morphological filtering to remove noise from the binarized images; A3: The CV model is used to accurately segment the waterfowl body area image, and grayscale morphological processing and Gaussian filtering are used to optimize the edge of the segmented image; the RSF model, local area fitting and narrowband optimization algorithm are used to optimize the extraction of the waterfowl beak boundary in the beak area image.

3. The multi-layer caged waterfowl feeding method based on behavioral feedback according to claim 1, characterized in that: Training the RDHL-DETR model specifically includes: B1: Collect image data of waterfowl deaths in different scenarios and annotate the image data using annotation tools; B2: Divide the labeled image data into training set, validation set and test set; B3: Use the training set to train the RDHL-DETR model; B4: Use the validation set to evaluate the trained RDHL-DETR model and save the optimal network training weights based on the validation results; B5: Use the test set to evaluate the generalization ability of the RDHL-DETR model and ensure the stability of the RDHL-DETR model.

4. The multi-layer caged waterfowl feeding method based on behavioral feedback according to claim 1, characterized in that: The S5 specifically includes: S501: A high-definition camera installed at the terminal of the trough running track collects the trough image; S502: performing brightness equalization and geometric distortion correction processing on the collected trough image; S503: extracting a main visual area image including the trough from the processed trough image according to a preset trough structure template; S504: Detecting the contaminated area of ​​the main visual area image of the trough using a fusion algorithm of color space, edge structure, and texture complexity to obtain a contamination score index; S505: Perform pollution judgment based on the pollution score index and the set pollution threshold: If the pollution score index is greater than the pollution threshold, the material tank is judged to be seriously polluted, and a manual or automatic cleaning operation is required. A pollution alarm record is generated including the material tank number, detection time, pollution score value and image screenshot.

5. The multi-layer caged waterfowl feeding method based on behavioral feedback according to claim 4, characterized in that: In S504, the contamination area detection is performed on the main visual area image of the short chute using a fusion algorithm of color space, edge structure and texture complexity to obtain a contamination score index, which specifically includes: C1: Convert the main visual area image into Lab color space; C2: Enhance the abnormal saturation areas in the image in Lab color space; C3: Utilizes the local gradient information of the image to extract the trough edge and patch outline, and uses color features to construct a pollution mask image; C4: Based on the pollution mask image, a pollution scoring function is constructed to quantify the pollution degree. The pollution scoring function is: in, is the pollution score index, is the area of ​​the detected contaminated area, is the total area of ​​the trough, is the average saturation concentration in the polluted area, is the local information entropy of the polluted area, 、 and An empirically selected coefficient that is set based on the proportion of manual needs.

6. A system according to any one of claims 1 to 5, characterized in that: include: The data acquisition device includes a camera and a wireless transmission module. The camera collects waterfowl image data, QR code labels on breeding cages, and feed trough images, and then transmits them to the edge image processing module through the wireless transmission module; The edge image processing module deploys a multi-channel color space image segmentation model and an image processing model, and transmits the processed image to the intelligent recognition module; The intelligent recognition module deploys a feeding desire assessment module, a waterfowl body condition scoring module, an RDHL-DETR module, and a feed trough contamination identification module. The waterfowl body area images and waterfowl beak area images segmented by the edge image processing module are transmitted to the feeding desire assessment module and the waterfowl body condition scoring module for calculation, and the calculation results are transmitted to the intelligent feeding control module as feeding instructions; the RDHL-DETR module identifies whether the waterfowl is dead based on the output of the feeding desire assessment module; the feed trough image processed by the edge image processing module is used as the input of the feed trough contamination identification module, and the feed trough contamination identification module transmits the identification results to the waterfowl intelligent feeding platform; An intelligent feeding control module, which controls the residence time of the mobile short feed trough according to feeding instructions; The waterfowl intelligent feeding platform collects and analyzes data related to waterfowl feeding behavior, waterfowl mortality judgment data, and feed trough pollution identification data, and generates relevant reports for display.

7. A multi-layer caged waterfowl feeding device based on behavioral feedback, characterized in that: include: A plurality of breeding cages (1), an egg conveyor belt (2), a QR code label (3), a mobile short feed trough (4), a waterfowl inspection high-definition camera (5), a feed trough pollution detection high-definition camera (6) and a core processing module (7); a plurality of the waterfowl inspection high-definition cameras (5) are mounted on the mobile short feed trough (4), the feed trough pollution detection high-definition camera (6) is installed at the feed trough inspection terminal, the QR code label (3) is located at the front end of the plurality of breeding cages (1), and the core processing module (7) is used to execute the steps of the method according to any one of claims 1 to 5.

Images