CN116646082A

CN116646082A - Disease early warning traceability system based on abnormal chicken manure under laminated cage raising mode

Info

Publication number: CN116646082A
Application number: CN202310469620.9A
Authority: CN
Inventors: 郑炜超; 秦文祥; 魏永祥; 李保明; 童勤
Original assignee: China Agricultural University
Current assignee: China Agricultural University
Priority date: 2023-04-27
Filing date: 2023-04-27
Publication date: 2023-08-25

Abstract

The invention relates to a disease early warning traceability system based on abnormal chicken manure under a laminated cage raising mode, which comprises an image acquisition platform, an abnormal manure traceability system and a cage raising chicken abnormal manure detection platform; the image acquisition platform is used for acquiring chicken manure images and sending the acquired chicken manure images to the abnormal manure tracing system; the abnormal excrement tracing system reversely pushes cage positions generated by the chicken manure based on the identified positions of the abnormal chicken manure, so that only the chicken producing the abnormal excrement is positioned; the abnormal excrement detection platform is used for acquiring abnormal excrement traceability system data and visually monitoring the overall process of the raising health of the cage-reared chickens. According to the invention, the image acquisition platform is utilized to acquire the chicken manure image in real time, the deep learning neural network technology is utilized to establish the recognition and positioning model, and the error correction unit is utilized to assist in calibration, so that the tracing of abnormal manure and the early warning of diseases are realized.

Description

Disease early warning traceability system based on abnormal chicken manure under laminated cage raising mode

Technical Field

The invention relates to a disease early warning and tracing system based on abnormal feces of chickens in a stacked cage raising mode, and relates to the field of health monitoring of chickens in a stacked cage raising mode.

Background

In the large-scale poultry raising process, the detection, prevention and control of chicken diseases are a non-negligible part. In the further optimization and upgrading process of the poultry raising industry, how to intelligently monitor the diseases of chickens in the stacked cage raising mode becomes a bottleneck for restricting the development of the industry. According to statistics, the number of diseases which threaten and cause harm to poultry industry in China is 80 or more, and the characteristics of high spreading speed and high incidence are directly related to the death rate, feed conversion ratio and feed conversion ratio of chickens. Therefore, it is necessary to provide early warning by intelligent detection of chicken diseases in cages, thereby reducing economic losses due to death of chickens.

The difficulty in identifying individual chickens in the cage raising environment is high, the characteristics are fuzzy, and the disease analysis is inaccurate. The existing intelligent disease detection scheme focuses on researching and detecting the form and sound of the poultry, and ignores the indication effect of chicken manure on poultry abnormality. Monitoring abnormal chicken manure is an important way for detecting chicken diseases, and normal chicken manure is usually brown, grey and has white solid at the top and is cylindrical in shape. When a chicken is ill, feces can show abnormality at the first time, and diseases with different causes can cause feces with different abnormal phenomena, such as: abnormal color feces, abnormal shape feces, abnormal water content, etc.

The prior art discloses a method and a system for detecting feces of cage chickens, which are characterized in that a chicken cage is placed above a conveyor belt, chicken feces in the chicken cage fall on the conveyor belt to be conveyed along the conveyor belt, images of the chicken feces on the conveyor belt are sequentially collected through a CCD camera, the color and the thickness degree of the feces on the conveyor belt are obtained through image processing, monitoring data are obtained, the types of the feces are obtained according to the comparison between the color and the thickness degree of the feces and the known feces-disease relationship, and health of chicken flocks in each chicken cage is monitored. The technology can accurately judge the health state of the poultry, but can not lock the area where the abnormal poultry is located, and can not implement targeted solving measures. The prior art also discloses a disease early warning system and method based on abnormal feces and anatomical images of the cage chickens, the categories of diseases are identified and judged by combining the abnormal feces images and the digestive system images of the sick chickens, the range of the abnormal feces is positioned by installing an RFID tag on the back of the feces cleaning belt, the final result is uploaded to a client, and treatment suggestions provided by specialist doctors are obtained remotely through a web module. However, the technology needs to install an RFID tag on the back of the manure cleaning belt, which can cause serious influence on the manure cleaning belt transmission mechanism and even can cause the manure cleaning belt to deviate; meanwhile, the situation that the RFID label falls off to cause positioning failure can occur, and the system cannot stably operate for a long time.

In conclusion, if the health state of the chickens can be intelligently judged in early stages of infection diseases of the chickens through chicken manure, and the cage position of the sick chickens is accurately positioned for accurate treatment, the method has important significance in promoting the green high-quality transformation of the poultry industry and promoting the intelligent development of the poultry farming industry.

Disclosure of Invention

The present invention aims to solve at least one of the technical problems existing in the prior art. Therefore, the invention aims to provide the disease early warning and tracing system based on the abnormal feces of the chickens in the stacked cage raising mode, which is simple to operate, and achieves the disease early warning of the chickens in the cage raising mode by identifying the abnormal feces and positioning the abnormal feces.

In order to achieve the aim of the invention, the invention adopts the following technical scheme: an early disease early warning and tracing system based on abnormal chicken manure under a laminated cage raising mode comprises an image acquisition platform, an abnormal manure tracing system and a cage raising chicken abnormal manure detection platform;

the image acquisition platform is used for acquiring chicken manure images and sending the acquired chicken manure images to the abnormal manure traceability system;

the abnormal excrement tracing system is used for identifying abnormal chicken manure, and reversely pushing cage positions generated by the chicken manure based on the identified positions of the abnormal chicken manure, so that only the chicken producing the abnormal excrement is positioned;

The abnormal excrement detection platform is used for acquiring data of the abnormal excrement traceability system and visually monitoring the overall process of the raising health of the cage-reared chickens.

Further, the image acquisition platform comprises an aluminum profile frame, a motion track, an industrial camera and an auxiliary light source;

the chicken coop frame is fixedly arranged at a manure outlet of a manure cleaning belt at the tail end of the chicken coop frame, the center of the aluminum profile frame is provided with a moving track, an industrial camera is arranged above a manure conveying belt and is arranged on the moving track in a sliding manner, two sides of the aluminum profile frame are provided with auxiliary light sources, and the auxiliary light sources are used for providing illumination to facilitate the industrial camera to photograph chicken manure.

Further, the image acquisition platform is further provided with a camera self-cleaning device, the camera self-cleaning device comprises a fixed base, a direct current motor, a swing arm structure, a cleaning brush and a controller, the fixed base is arranged on one side of the industrial camera, the direct current motor is arranged on one side of the fixed base, the output end of the direct current motor is connected with the swing arm structure, the cleaning brush is arranged on the swing arm structure, and the controller is used for driving the direct current motor to work at regular time.

Further, the abnormal excrement tracing system comprises an abnormal excrement identification and classification unit, an abnormal excrement tracing positioning unit and an error correction unit;

the abnormal excrement identification and classification unit is configured to identify the category and the position of the abnormal excrement according to a pre-established abnormal excrement identification and classification model based on the chicken excrement image;

the abnormal excrement backtracking positioning unit is configured to reversely push cage positions generated by chicken manure to perform space tracing based on the positions of the identified abnormal chicken manure through a pre-established abnormal excrement backtracking positioning model;

the error correction unit is configured to eliminate a part of accumulated error generated in the abnormal stool backtracking positioning.

Further, the process for establishing the abnormal stool identification classification model comprises the following steps:

data set preparation, comprising: according to the characteristics of chicken manure, the collected various chicken manure image data are divided into three types of abnormal shape, abnormal color and abnormal water content, and image marking software labelimg is utilized for marking data to generate an xml file containing coordinate point information, category names and length and width information of a marking frame; preparing an abnormal chicken manure data set from the original image and the xml file according to the data set format of the PASCAL VOC;

Model training, comprising: building a yolov5 use system environment; modifying network parameters; compiling network codes; loading a pre-training model; loading a data set for iterative training; and stopping training when the loss converges to a preset value or the iteration number reaches the maximum, and obtaining the abnormal fecal identification classification model after training is completed.

Further, the establishment of the abnormal fecal backtracking positioning model comprises the following steps:

building a SuperPoint and SuperGlue use system environment, and replacing a feature detection head of a SuperGlue structure with a SuperPoint network; modifying network parameters and compiling network codes;

inputting the preprocessed images into a built network for processing to obtain pixel offset distances of two adjacent images;

obtaining conversion parameters of the pixel distance and the real world distance through industrial camera calibration;

establishing a model according to the position of the on-site chicken coop frame, and obtaining the position of an abnormal excrement generation cage position according to the pixel offset distance of the adjacent images as X:

wherein S is the total offset distance of abnormal feces based on image pixels, M is the sum of pixels in the X-axis direction of the image size obtained by industry, D is the actual distance from the placement position of the industrial camera to the last chicken coop, N is the width of a single chicken coop, and K is the conversion coefficient of the calibrated pixel distance and the actual distance of the industrial camera.

Further, the error correction unit takes marks which are manually marked on the manure transfer belt in a certain and known mode in pairs as real references, codes are written in an OpenCV library to detect and identify the manual marks in pictures acquired in the manure transfer belt running manure cleaning process, moving distance information of the marks and the number of the detected marks are recorded, measurement errors are obtained through comparing actual distances between data generated by an algorithm and the marks, and the real distances between the marks and the distances generated by calculation are mutually fused by utilizing a data fusion idea so as to eliminate accumulated errors.

Further, writing the code through the OpenCV library, detecting and identifying the mark includes: converting the RGB image into HSV color gamut space through format conversion; screening the characteristic colors of the marks; filtering non-target noise in the screened binarized mask image by using image morphology operation and filtering operation; and calculating a minimum circumscribed rectangle through searching the outline, optimizing and combining the multi-target circumscribed rectangle frames aiming at partial chicken manure shielding marks, and positioning the mark positions. Further, if abnormal chicken manure X _n+1 Between the nth mark and the (n+1) th mark, after error correction, the distance T between the abnormal feces and the camera is the same _d The method comprises the following steps:

T _d ＝FR _d +C _d +(n-1)· _I d

wherein C is _d For the distance of abnormal feces from the last mark or between cameras, I _d T is the true distance between the marks _d For the distance of abnormal feces from a camera before feces cleaning, namely the retrospective positioning position, FR _d The location is located for the backtracking of the first marker.

Further, the abnormal feces detection platform of the cage-rearing chicken builds an abnormal feces detection web interface by using the VUE framework, deploys the abnormal feces detection web interface on the cloud server, and the cloud server is responsible for carrying out data analysis processing on the acquired image data, packaging the abnormal results, transmitting the data interface to a background database and displaying the data interface.

The invention adopts the technical proposal and has the following characteristics:

1. the method utilizes the image acquisition platform to acquire the images of the chicken manure on the manure cleaning belt below the cage-raising chicken cage in real time, utilizes the deep learning neural network technology to establish an abnormal manure recognition model, researches the sequence image analysis based on the deep learning to realize the determination of the space position of the abnormal manure, and utilizes the error correction unit to assist in calibration, thereby finally realizing the tracing of the abnormal manure and the early warning of diseases.

2. According to the invention, the angle of intelligent detection of health of the layered cage chickens is transferred from the observation and identification cage chickens to the observation excrement, the health judgment of the chickens is visual, the accuracy of identifying abnormal chicken manure reaches more than 95%, and the problems of large individual identification difficulty, fuzzy characteristics and inaccurate disease analysis of chickens in the cage raising environment are effectively avoided.

3. According to the invention, the abnormal chicken cage position generating abnormal feces is retrospectively positioned, the positioning accuracy is 94.7%, and the accurate treatment of diseases in the chicken raising process in the laminated cage raising mode can be realized.

4. The intelligent continuous monitoring of abnormal feces can avoid the conditions of group infection and the like caused by delayed disease diagnosis, missed treatment window time and the like of chickens, reduce economic loss caused by death of chickens, and also reduce the working intensity of breeding staff.

5. The deep learning model related by the invention can be subjected to model fusion and deployed into a project framework, hardware simplification is realized, and the recognition and classification of abnormal chicken manure, the backtracking positioning of abnormal manure and error correction can be realized by shooting sequence images above the manure transfer belt by using one industrial camera, so that the operation is simple, and the machine arrangement of a production site is not required to be destroyed.

6. The invention establishes sequence image matching by deep learning, realizes the monitoring of the running distance of the conveyor belt, and reduces the problem of large accumulated error in the long-distance monitoring process by a data fusion mode.

In conclusion, the invention can realize the high-efficiency, low-cost and high-precision abnormal excrement recognition, health monitoring, abnormal tracing and disease early warning of the cage-raised chickens, is beneficial to promoting the intelligent development of the chicken raising industry in China, provides key equipment support for intelligent raising of the cage-raised chickens, and can be widely applied to early warning and tracing of the disease of the cage-raised chickens.

Drawings

Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the invention. Like parts are designated with like reference numerals throughout the drawings. In the drawings:

FIG. 1 is a schematic diagram of a system architecture according to an embodiment of the present invention;

fig. 2 is a structural diagram of an image acquisition platform according to an embodiment of the present invention:

FIG. 3 is a schematic view of field acquisition according to an embodiment of the present invention;

FIG. 4 is a diagram of a YOLOv5 network architecture in accordance with an embodiment of the present invention;

FIG. 5 is a flow chart of constructing an optimal abnormal stool recognition and classification model according to an embodiment of the present invention;

FIG. 6 is a block diagram of a SuperPoint network in accordance with an embodiment of the present invention;

FIG. 7 is a Superglue network architecture diagram according to an embodiment of the present invention;

FIG. 8 is a schematic diagram of an error correction algorithm according to an embodiment of the present invention;

FIG. 9 is a block diagram of a cage chicken abnormal feces detection platform according to an embodiment of the invention;

FIG. 10 is a screen shot of a login interface of a caged chicken abnormal feces detection platform according to an embodiment of the invention;

FIG. 11 is a diagram illustrating an abnormal stool detection result interface screen shot according to an embodiment of the present invention;

FIG. 12 is a detailed screenshot of the results of abnormal stool detection in an embodiment of the present invention.

Detailed Description

It is to be understood that the terminology used herein is for the purpose of describing particular example embodiments only, and is not intended to be limiting. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. The terms "comprises," "comprising," "includes," "including," and "having" are inclusive and therefore specify the presence of stated features, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, steps, operations, elements, components, and/or groups thereof. The method steps, processes, and operations described herein are not to be construed as necessarily requiring their performance in the particular order described or illustrated, unless an order of performance is explicitly stated. It should also be appreciated that additional or alternative steps may be used.

Although the terms first, second, third, etc. may be used herein to describe various elements, components, regions, layers and/or sections, these elements, components, regions, layers and/or sections should not be limited by these terms. These terms may be only used to distinguish one element, component, region, layer or section from another region, layer or section. Terms such as "first," "second," and other numerical terms when used herein do not imply a sequence or order unless clearly indicated by the context. Thus, a first element, component, region, layer or section discussed below could be termed a second element, component, region, layer or section without departing from the teachings of the example embodiments.

For ease of description, spatially relative terms, such as "inner," "outer," "lower," "upper," and the like, may be used herein to describe one element or feature's relationship to another element or feature as illustrated in the figures. Such spatially relative terms are intended to encompass different orientations of the device in use or operation in addition to the orientation depicted in the figures.

Because the prior art does not intelligently judge the health state of chickens in early stages of infection diseases through chicken manure, the cage position of the sick chickens cannot be accurately positioned. The invention provides a disease early warning traceability system based on abnormal chicken manure under a stacked cage raising mode, which comprises an image acquisition platform, an abnormal manure traceability system and a cage raising chicken abnormal manure detection platform. The image acquisition platform is used for acquiring chicken manure images and sending the acquired chicken manure images to the abnormal manure traceability system; the abnormal excrement tracing system is arranged at the cloud server and is used for identifying abnormal chicken manures and reversely pushing cage positions generated by the chicken manures based on the identified positions of the abnormal chicken manures to accurately position only the chicken producing the abnormal manures; the abnormal excrement detection platform for the cage-reared chickens is a client and is used for acquiring abnormal excrement traceability system data, and the whole process of the cage-reared chickens are monitored in a visual mode, so that the prevention and treatment of chicken diseases by the rearing staff are conveniently guided. Therefore, the invention can efficiently and accurately realize abnormal feces identification, abnormal tracing and disease early warning of the cage chickens with low cost.

Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

As shown in fig. 1, the disease early warning traceability system based on abnormal chicken manure under a stacked cage raising mode provided in this embodiment includes an image acquisition platform 1, an abnormal manure traceability system 2 and a cage raising chicken abnormal manure detection platform 3.

The image acquisition platform 1 is used for acquiring chicken manure images and sending the acquired chicken manure images to the abnormal manure traceability system 2;

the abnormal excrement tracing system 2 is arranged at the cloud server and is used for reversely pushing cage positions generated by the chicken manure based on the identified positions of the abnormal chicken manure so as to accurately position the chicken with the abnormal excrement;

the abnormal excrement detection platform 3 of the cage-rearing chickens is a client and is used for acquiring data of the abnormal excrement traceability system 2, visually monitoring the whole process of the rearing health of the cage-rearing chickens, and conveniently guiding the raising personnel to prevent and treat chicken diseases and trace the source for positioning.

In a preferred embodiment, as shown in fig. 2 and 3, the image acquisition platform 1 comprises an aluminum profile frame 11, a movement track 12, an industrial camera 13 and an auxiliary light source 14.

The aluminum profile frame 11 is fixedly erected at the manure outlet of the manure cleaning port at the tail end of the chicken coop frame, the moving track 12 is arranged at the center of the aluminum profile frame 11, the industrial camera 13 is arranged above the manure transfer belt, the industrial camera 13 is arranged on the moving track 12 in a sliding manner through the moving sliding block 131, the industrial camera 13 can move on the moving track 12, the imaging range is ensured to cover the manure transfer belt surface at the manure cleaning port, and the imaging range is used for collecting chicken manure images. Both sides of the aluminum profile frame 11 are provided with auxiliary light sources 14, and the auxiliary light sources 14 are used for providing illumination so that the industrial camera 13 can take pictures of chicken manure conveniently. When the manure transfer belt does not work, the auxiliary light source 14 and the industrial camera 13 are in a standby dormant state; when the manure transfer belt starts to start manure cleaning, the auxiliary light source 14 and the industrial camera 13 start to work simultaneously to collect images of chicken manure, and the images of manure on the manure transfer belt are continuously collected when the conveyor belt moves.

Further, the industrial camera 13 may be fixed in the camera protection device 15, having a waterproof and dustproof function.

Further, the auxiliary light source 14 may employ an industrial LED explosion-proof lamp with adjustable illumination angle and illumination brightness.

Further, as shown in fig. 2 (b), the industrial camera 13 is further provided with a camera self-cleaning device 16, which comprises a fixed base 161, a direct current motor 162, a swing arm structure 163, a cleaning brush 164 and a controller 165, wherein the fixed base 161 is arranged on one side of the industrial camera 13, the direct current motor 162 is arranged on one side of the fixed base 161, the output end of the direct current motor 162 is connected with the swing arm structure 163, the swing arm structure 163 is provided with the cleaning brush 165, and the swing arm structure 163 is driven by the direct current motor 162 to drive the cleaning brush 165 to swing back and forth to clean dust on the camera protection device 15 during starting. Therefore, the camera self-cleaning device 16 can regularly clean dust of the lens of the industrial camera, ensure clear imaging, and effectively solve the problem that the camera cannot work for a long time in the production process due to the severe environment of the henhouse.

In a preferred embodiment, the abnormal stool tracing system 2 comprises an abnormal stool identification and classification unit, an abnormal stool backtracking positioning unit and an error correction unit.

The abnormal excrement identification and classification unit is configured to identify and classify abnormal chicken manure based on the chicken manure image, and identify abnormal excrement and occurrence time of the abnormal excrement.

The abnormal excrement backtracking positioning unit is configured to reversely push cage positions generated by the chicken manure based on the identified positions of the abnormal chicken manure to perform space backtracking so as to realize accurate positioning of abnormal chickens generating abnormal excrement;

the error correction unit is configured to eliminate partial accumulated errors generated in the abnormal fecal backtracking and positioning, and ensure the accuracy of the backtracking and positioning.

Further, the abnormal manure recognition and classification unit performs data analysis processing on the collected group of chicken manure sequence images by using a deep learning target detection frame YOLOV5 to realize recognition and classification of abnormal chicken manure, and the process comprises the following steps:

1) Setting an image to be acquired in one second and one frame, and preprocessing the image;

2) Establishing a deep learning model;

3) And detecting the preprocessed image by using a deep learning model, analyzing the result to obtain an abnormal stool image, and recording the occurrence time and detection result of the abnormal stool.

Specifically, the abnormal feces appearance time is recorded and used for judging and analyzing the current health condition of the abnormal chicken, so that the overall analysis of the diseases is facilitated; meanwhile, the historical data of the abnormal time is stored, so that the statistical analysis of the health condition of the chickens in the whole henhouse is facilitated. The detection result comprises abnormal excrement type information and position information of the abnormal excrement in the image, wherein the abnormal excrement type information is respectively abnormal color, abnormal shape and abnormal water content.

In this embodiment, the preprocessing of the image includes image enhancement and image cropping.

The image enhancement is specifically: and manually screening the collected pictures by taking obvious color contrast, proper brightness and clearness as standards, and processing the pictures with lower brightness or poor contrast by using a histogram equalization technology, wherein the adjustment parameters are cliplimit=2 and tilegridsize= (16, 16), so that the quality and the quantity of the picture data are ensured.

The picture cutting is specifically as follows: square cutting is used on the basis of a 2592 x 2048ppi picture, invalid areas or areas with interference to detection are removed, and the fact that the length and width dimensions of the picture are the same and abnormal chicken manure areas are contained is guaranteed. The square picture is scaled to ensure that it is 640 x 640 pixels in size.

In this embodiment, the deep learning model is built, including the following steps:

1. data set preparation, comprising: 11 According to the characteristics of chicken manure, classifying the collected various chicken manure image data into three types of abnormal shape, abnormal color and abnormal water content; 12 Using image labeling software labelimg to label data and generating an xml file containing coordinate point information, category names, length and width of pictures and the like of a labeling frame; 13 Image original and xml file, and preparing an abnormal chicken manure data set comprising a training set, a verification set and a test set according to the data set format of the PASCAL VOC (The PASCAL Visual Object Classes).

2. Model training, comprising: 21 Building a yolov5 use system environment; 22 Modifying the network parameters; 23 Compiling the network code; 24 Loading a pre-training model; 25 Loading a training set for iterative training; 26 When the loss converges to a preset value or the iteration number reaches the maximum, stopping training to obtain an abnormal fecal identification classification model after training is completed;

3. model verification, comprising:

31 Determining an evaluation index, comprising: introducing three indexes of accuracy, recall rate and F1 score for evaluation; the accuracy P is the ratio of the number of classified correct samples to the total number of samples for a given dataset, namely:

recall R, also known as recall, is used to describe the ratio of positive examples determined to be true to total positive examples in the classifier,

namely:

f1 is the harmonic mean of the precision and recall:

wherein P is the accuracy, R is the recall, TP is the true for positive prediction, FN is the false for positive prediction, FP is the true for negative prediction, TN is the false for negative prediction;

32 The test set data and the images acquired by the image acquisition platform are utilized to verify, the 3 evaluation indexes are utilized to evaluate the abnormal fecal identification classification model obtained through training, the P-R curve is utilized, the accuracy P is taken as the Y axis, the recall rate R is taken as the X axis, and the F1 evaluation index is combined to select the abnormal fecal identification classification model with the optimal effect to apply and detect.

In model verification, calculating the area surrounded by the P-R curve and the coordinate axis, and selecting the abnormal fecal identification classification model with large area as the optimal identification model; if the areas surrounded by the P-R curves of the models are equal, F1 evaluation indexes corresponding to the models are compared, and the abnormal stool identification classification model corresponding to the large F1 value is selected as the optimal abnormal stool identification classification model.

Further, the abnormal manure backtracking positioning unit realizes non-contact manure belt operation information monitoring, and analyzes and monitors the movement distance of the manure belt surface by constructing a deep learning characteristic point recognition and characteristic matching network superpoint-superglue, so as to obtain the manure belt operation distance, reduce the monitoring difficulty of the manure belt operation, eliminate the possibility of damage to the manure belt in the monitoring process, monitor and record the type, quantity and space position information of the abnormal manure, specifically analyze the movement distance of the manure belt in an image acquisition interval by a deep learning neural network model through a motion sequence image, reversely push the cage position generated by chicken manure, realize the accurate positioning of abnormal chicken producing the abnormal manure, and achieve the effect of disease early warning, and comprise: building a SuperPoint and SuperGlue use system environment, and replacing a feature detection head of a SuperGlue structure with a SuperPoint network; modifying network parameters and compiling network codes; inputting the preprocessed image into a SuperPoint-Superglue network; acquiring pixel offset distances of two adjacent pictures: the SuperPoint network analyzes feature points and feature descriptors of two adjacent pictures, transmits the feature points and feature descriptors to the SuperGlue network, the SuperGlue network performs feature matching and item filtering exclusion, outputs a matching relation between image features, calculates the difference of coordinates of all points with the same matching relation in the two adjacent pictures, and performs summation and averaging to obtain pixel offset distances of the two adjacent pictures.

Establishing a model according to the position of the on-site chicken coop frame, which comprises the following steps: after the average pixel displacement of the sequence images is calculated based on the SuperPoint-SuperGlue model, the average pixel displacement is converted into the actual offset distance S of the abnormal feces by a conversion coefficient obtained by linear calibration of an industrial camera, the total offset distance S of the abnormal feces based on the image pixels is analyzed, the pixels and M of the industrial camera in the X-axis direction of the image size are obtained, the actual distance D of the industrial camera placement position from the last chicken coops is the width N of the single chicken coops, the conversion coefficient K of the industrial camera calibration pixel distance and the actual distance is calculated according to the formula, the position X of the abnormal feces is X, and the position X can be calculated according to the formula:

according to the image pixel offset distance during abnormal excrement identification and the fact that abnormal excrement appears in the left and right areas of the excrement cleaning belt, specific cage positions of chickens producing abnormal excrement and the left and right sides of a rearing cage where the chickens producing abnormal excrement are located can be judged, and space tracing of the abnormal excrement is achieved.

In this embodiment, the industrial camera calibrates conversion parameters of the acquired pixel distance and the real world distance: and (3) performing linear calibration of the industrial camera, horizontally placing a tape measure in the field of view of the industrial camera, recording the actual measurement value of the tape measure in the field of view, and performing proportional conversion with the pixel size of the image to obtain the conversion coefficient of the pixel and the actual distance.

Further, the error correction unit uses manual marks, namely marks which are manually marked on the excrement transfer belt in a certain and known mode, as real references (the marks can be painted marks or reflective material strips), compares and analyzes errors of an abnormal excrement backtracking positioning algorithm, fuses real data of the marks with data generated by the algorithm by utilizing the data fusion idea, eliminates partial accumulated errors generated in the abnormal excrement backtracking positioning algorithm, and ensures the accuracy of backtracking positioning, and the error correction unit specifically comprises:

a group of identification marks with known intervals are marked on a conveyor belt, the marks are identified through code detection written in an OpenCV library, moving distance information of the marks and the number of the detected marks are recorded, measurement errors are obtained through comparing actual distances between data generated by an algorithm and the marks, and the actual distances between the marks and the distances generated by the algorithm are mutually fused by utilizing a data fusion idea so as to eliminate accumulated errors.

In this embodiment, writing a code through an OpenCV library, detecting and identifying the mark, that is, writing a mark through a Python language by using an OpenCV library, includes: firstly converting an RGB image into an HSV color gamut space through format conversion, screening the characteristic colors of the marks, filtering non-target noise in the screened binarization mask image by using image morphological operation and filtering operation, calculating the minimum circumscribed rectangle through finding the outline, and positioning the mark positions. The situation that the target is discontinuous occurs because chicken manure falls on the mark frequently in daily production and the mark part in the binarization mask image is blocked is caused, so that the selection rule of the minimum circumscribed rectangle is optimized in actual operation in a multi-target contour merging mode.

In this embodiment, the method for fusing the true distance between the marks and the distance generated by the algorithm by using the data fusion concept includes: the sources of the running data of the manure transfer belt in the model are two, one is obtained by identifying marks, the other is obtained by reasoning through a deep learning model, and because the distance data sampling based on the marks is different from the data sampling frequency based on the deep learning model (the data sampling based on the marks can be carried out once every 3.6m and the data sampling of the deep learning model can be carried out once every 1 s), the low frequency data are inserted into the high frequency data in a time alignment mode for primary fusion, and then the Kalman filtering is carried out on the data between every two marks to obtain continuous and more accurate error correction.

In a preferred embodiment, the abnormal excrement detection platform for the cage-raised chickens uses a VUE framework to link a MySQL database for statistically analyzing historical abnormal data, displaying abnormal excrement identification and classification results, tracing cage positions and other information, and is convenient for the uniform management of the farmers.

In summary, the abnormal excrement identifying and classifying unit, the abnormal excrement backtracking and positioning unit and the error correcting unit share the image acquisition platform 1, and the data processing algorithms in the three units are deeply fused to realize end-to-end abnormal information output, and the abnormal information of chicken manure, including the types, the number, the spatial positions, the generation time and the like of the anomalies, can be analyzed through a group of sequence images acquired by the image acquisition platform 1, so that the requirements and the operation difficulty of hardware equipment are simplified.

The specific application of the disease early warning traceability system based on abnormal chicken manure in a stacked cage raising mode provided by the invention is described in detail below through specific embodiments.

According to the embodiment, the angle of intelligent health detection of poultry is transferred from observing poultry to observing poultry excrement, according to the abnormal characteristics of excrement after poultry is ill, an image acquisition platform 1 is arranged at the upper end of a manure outlet of a manure belt at the tail end of a stacked cage chicken cage, images of chicken manure on the manure belt below the cage chicken cage are continuously acquired, an abnormal manure identification model and an abnormal manure tracing model are established by utilizing a deep learning technology, and under the assistance of an error correction unit, early warning and tracing of health detection of the cage chicken are completed, and the method specifically comprises the following steps:

s1, continuously acquiring chicken manure images on a manure transfer belt when the conveyor belt moves by the image acquisition platform 1, and transmitting the acquired sequence images to a cloud server.

The on-site pictures transmitted by the picture acquisition platform 1 can be used for making a data set through pretreatment under the influence of on-site environmental factors, and the picture pretreatment steps mainly comprise brightness adjustment, contrast adjustment, size cutting and the like, so that chicken manure targets are more outstanding, and the detection effect is improved.

Specifically, brightness adjustment and contrast adjustment: because the conveying belt is made of PP materials, if the light is too bright under the illumination condition, the background exposure is easy to cause, and the picture quality is affected, so that excessive light is avoided in the experimental process. Under the setting, the contrast of chicken manure of different colors and outlines in the obtained picture is not high, firstly, the contrast of the different colors and outlines in the picture is enhanced by using limiting contrast adaptive histogram equalization (CLAHE) on an original image, and the picture is adaptively processed by using a CLAHE method in an Opencv library, wherein the setting parameters are clipLimit=2 and tileGridSize= (16, 16).

And (3) size cutting: because the resolution of the video data acquired by the image acquisition platform is 2592×2048ppi, the calculation amount can be greatly increased by directly calculating, and the hardware cost is increased. Therefore, the images need to be resized to 640 x 640ppi uniformly, which can increase the model training efficiency. And (3) square cutting is used on the basis of the picture with the original size of 2592 x 2048ppi, invalid areas or areas with interference on detection are removed, the same length and width size of the picture is ensured, and then the picture is uniformly scaled to 640 x 640ppi.

S2, an abnormal manure recognition and classification unit performs data analysis processing by utilizing an improved deep learning target detection frame yolov5 to recognize and classify abnormal chicken manure, and when the abnormal chicken manure appears, abnormal information is transmitted to a MySQL database for guiding production management, and the method comprises the following steps:

S21, establishing a picture data set: and (5) acquiring an abnormal chicken manure picture on site, and manufacturing an abnormal chicken manure data set.

S22, labeling abnormal chicken manure in the image into abnormal colors, abnormal shapes and abnormal water content by using Labelimg software, and generating corresponding xml files by labeling information of each labeled picture.

S23, preparing an abnormal chicken manure data set from the picture and a corresponding xml file according to a data set format of a PASCAL VOC (The PASCAL Visual Object Classes), wherein the abnormal chicken manure data set comprises a training set, a verification set and a test set, the ratio is 8:1:1, and the deep learning target detection model is established, and the method comprises the following steps:

as shown in fig. 4 and 5, a deep-learning abnormal stool recognition classification model is established by using a deep convolutional neural network yolov5, and the structure of the model is mainly divided into an input end, a back bone (Backbone network), a neg network and a prediction end (output end). The input end is a series of incoming chicken manure images; the backbox selects a Focus structure and a CSP structure as network frames, and the key in the Focus structure is slicing operation. The 640×640×3 three-channel image is input into the Focus structure, and is first changed into a 320×320×12 feature map through slicing operation, and then is finally changed into a 320×320×32 feature map through convolution operation using 32 convolution kernels. There are two CSPs in YOLOv5, the csp1_x structure in the Backbone network of the backhaul, and another csp2_x structure in the neg. For the Backbone network structure of the Backbone, the convolution kernel sizes in the CSP module are 3*3, the step value is 2, and if the input image size is 640 x 640, the law of its feature map change is: 640 x 640- >320 x 320- >160 x 160- >80 x 80- >40 x 40- >20 x 20, and finally obtaining a 20 x 20 size feature map. The Focus structure and the CSP structure can reduce the calculation bottleneck, enhance the learning ability of the network and ensure the operation speed and accuracy of the model. SPP structures have been used to solve the multi-scale detection problem, which use convolution kernels of three different sizes, 5×5,9×9,13×13, to increase the receptive field of the network. And performing convolution operation on the input feature map by using the convolution kernels with the three different scales, using padding operation to ensure that w and h dimensions of the output features are the same, and then splicing the original feature map with the three different output feature maps to serve as input of a lower network. The SPP structure is insensitive to aspect ratio and size characteristics of the input image, improving the scale-invariance of the image and reducing over-fitting; the Neck network mainly generates a feature pyramid which can enhance the detection of the model on objects with different scaling scales, so that the same object with different sizes and scales can be identified. The PANET network is based on Mask R-CNN and FPN frameworks, while enhancing information propagation, and is considered to be the most suitable feature fusion network for YOLO, thus aggregating features using PANET as neg. The feature extractor of the network employs a new FPN structure that enhances the bottom-up path, improving the propagation of low-level features. Each stage of the three paths takes as input the feature map of the previous stage and processes with a 3x3 convolutional layer. The output is added to the same stage profile of the top-down path through the cross-connect, which provides information for the next stage. And simultaneously, recovering the destroyed information path between each candidate region and all feature layers by using adaptive feature pooling (Adaptive feature pooling), and aggregating each candidate region on each feature layer to avoid being arbitrarily allocated. The output layer directly outputs the identification result, namely the type and the position of the chicken manure to be predicted.

YOLO-V5 adopts the GIoU loss value as the loss function of the binding box, but the chicken manure on the conveyor belt are mutually stacked, the abnormal chicken manure is easy to be blocked and disturbed, in order to improve the detection performance of the model on the blocked abnormal chicken manure, the GIoU loss value is replaced by the CIoU loss value which is more suitable for the detection of a blocking target as the loss function of the frame loss, and the calculation formula of the CIoU loss value is as follows:

wherein the method comprises the steps of

Wherein L is _CIoU The CIoU loss value, A is a prediction frame, B is a real frame, ρ is Euclidean distance, p is the center point of the prediction frame, p ^gt For the center point of the target frame, c is the diagonal distance of the minimum circumscribed rectangle between frames, alpha is the weight function, ioU is the intersection ratio of the predicted frame and the real frame, w is the width of the predicted frame, h is the height of the predicted frame, and w ^gt For the width of the target frame, h ^gt Is the high of the target frame.

S24, establishing an abnormal excrement identification classification model: building a yolov5 use system environment; modifying network parameters; compiling network codes; loading a pre-training model; loading a data set for iterative training; and stopping training when the loss converges to a preset value or the iteration number reaches the maximum, and obtaining the abnormal fecal identification classification model.

S25, verifying the images acquired by using the test set data and the image acquisition platform, evaluating the training-obtained abnormal stool identification classification model by using the 3 evaluation indexes, and using a P-R curve: taking the accuracy P as a Y axis and the recall rate R as an X axis; and combining F1 evaluation indexes, selecting an abnormal stool identification classification model with optimal effect, and performing application detection.

S3, the abnormal manure backtracking and positioning unit transmits the motion sequence images of the surface of the manure transfer belt collected during manure cleaning into a trained deep learning feature matching model SuperPoint-SuperGlue, analyzes the moving distance of the manure transfer belt in an image acquisition interval, and when abnormal manure is identified, reversely pushes out the space position generated by the abnormal manure according to the moving distance of the current manure transfer belt and the chicken coop arrangement on the henhouse site to realize space tracing and disease positioning of the abnormal manure.

In order to reduce hardware requirements and simplify monitoring operation, the invention utilizes two deep learning network frameworks Superpoint and Superglue, uses an image feature point detection and feature matching method to analyze the sequence images acquired by the image acquisition module to obtain average moving pixel values of the acquired sequence images of the movement of the manure transfer belt, then carries out camera calibration, and converts the pixel movement of the image of the manure transfer belt into physical movement in the real world according to the installation pose of an industrial camera and internal parameters and external parameters, thereby obtaining the actual movement distance of the manure transfer belt. If abnormal chicken manure is detected, the cage position for generating the abnormal chicken manure can be reversely pushed out by the movement of the manure conveying belt, and backtracking positioning is completed.

Specifically, the traditional image feature point detection and feature matching method is generally obtained according to manual design, and with the continuous progress of the deep learning technology, the image feature points are extracted by using a deep neural network and then the feature points are matched with each other, so that higher accuracy can be obtained. Experiments prove that SuperPoint+SuperGlue has higher matching accuracy for henhouse conveyor belt distance monitoring, so that deep learning networks SuperPoint and SuperGlue are selected to be built to establish an abnormal fecal backtracking positioning model. According to the invention, the characteristic points of the two acquired adjacent pictures are extracted by using the Superpoint network, redundant image information is removed, and the calculation load is reduced. And transmitting the compressed image information to a Superglue network for feature matching, calculating average pixel displacement of two adjacent images, and converting the pixel displacement into actual displacement of the manure belt by using camera calibration parameters, so that the effect of monitoring the movement of the manure belt is achieved.

As shown in fig. 6, the SuperPoint algorithm is a deep learning method based on feature point detection and descriptor extraction, and is used for self-supervision training. The network structure is an encoder-decoder structure, a complete picture is input, deep features of the picture are extracted through a shared VGG network encoder, and then the feature points and the descriptors are output through two decoders of the feature points and the descriptors respectively. Unlike conventional manual design algorithms, the feature points and descriptors of the network are generated in parallel. As shown in fig. 7, the superflue network can perform feature matching and term filtering exclusion simultaneously, the input of which is feature points and descriptors (manual features or deep learning features) in two images, and the output of which is a matching relationship between image features. The feature matching is solved by solving the differentiable optimal transfer problem, and a cost matrix is constructed by utilizing the feature value output by the GNN (graph neural network). The whole frame consists of two main modules: attention GNN, and a best matching layer. Wherein, the attention GNN encodes the input feature points and descriptors into a feature matching vector, and then the feature matching performance of the vector f is repeatedly enhanced (repeated L times) by utilizing self attention and cross attention; then entering an optimal matching layer, obtaining a matching degree score matrix by calculating the inner product of the feature matching vector, then calculating an optimal feature distribution matrix by a sink horn algorithm (iterated T times), and solving the condition that no feature point exists in the image by setting a durtbin channel in the score matrix S so as to achieve the effect of filtering the outer point.

The specific process of constructing the abnormal fecal backtracking and positioning model is as follows: building a SuperPoint and SuperGlue use system environment; modifying network parameters; compiling network codes; inputting the original image acquired in the abnormal excrement recognition and classification module into a network; acquiring pixel offset distances of two adjacent pictures; obtaining conversion parameters of the pixel distance and the real world distance through camera calibration; establishing a model according to the position of the on-site chicken coop frame; the method comprises the steps that abnormal feces are analyzed based on total offset distance S of image pixels, pixels and M in the X-axis direction of the image size obtained by a camera, actual distance D between the placement position of the camera and the last chicken coop, width N of the single chicken coop, conversion coefficient K between the calibrated pixel distance of the camera and the actual distance, the position of the abnormal feces generation coop is X, and the X can be calculated according to a formula:

S4, the error correction unit takes marks with certain and known distances from each other, which are manually marked on the manure transfer belt, as real references, codes are written in an OpenCV library in pictures acquired in the manure transfer belt running manure cleaning process to detect and identify the manual marks, the moving distance information of the marks and the number of the detected marks are recorded, the actual distances between data generated by a comparison algorithm and the marks are used for acquiring measurement errors, and the actual distances between the marks and the distances generated by the algorithm are mutually fused by utilizing a data fusion idea so as to eliminate accumulated errors.

In particular, since the henhouse cage is generally 70 to 90 meters in length, the retrospective positioning can accumulate measurement errors during long-distance monitoring, and the accuracy of positioning is greatly affected. Therefore, the present embodiment has devised error correction for solving the problem of large accumulated error generated under long-distance monitoring. A set of artificial marks are added on the manure transfer belt, and the marks are equidistant from each other and have known distances. When the manure transfer belt runs, the algorithm starts to record the running distance of the manure transfer belt after the picture in the visual field of the industrial camera 13 changes, as shown in fig. 8, when the first mark enters the picture of the industrial camera, the tracing positioning position of the first mark is calculated and recorded as FR by using the distance between the mark center and the image center and the total running distance of the manure transfer belt recorded at the moment _d While the subsequent operating distance is monitored and recorded with the first mark as a reference. When the second mark enters the camera picture, repeating the steps to calculate the monitoring distance between the second mark and the first mark as E _d1 Monitoring and recording the subsequent operation distance by taking the second mark as a reference after the calculation is finished, and the like, wherein when the nth mark appears in the industrial camera picture, the monitoring distance between the nth mark and the n-1 th mark is E _dn-1 Assuming that the true distance between two adjacent marks is I _d The accumulated error of the whole system is E _d1 +E _d2 +……+E _dn-1 -n·I _d . If abnormal stool is detected during short-range monitoring, e.g. X in FIG. 8 ₁ ，X ₂ Where the cumulative error is small, within an acceptable range. If abnormal feces are detected during long distance monitoring, e.g. X in the figure ₃ Or longer distances, the accumulated error may pass through the true distance I between the marks _d Distance E from the calculated mark _d Mutually fused, so that a part of accumulated errors are eliminated, and the retrospective positioning accuracy is improved. In the graph X ₁ ，X ₂ ，X ₃ Three abnormal chicken manure positions are taken as an example, C _d T is the distance of abnormal feces from the last mark or between cameras (the part of error depends on the algorithm design itself) _d For the distance between the abnormal feces and the camera before feces cleaning, namely the retrospective positioning position, the retrospective distance calculation formula after error correction is used is shown in table 1.

TABLE 1

From the design, if abnormal chicken manure X _n+1. Between the nth mark and the n+1th mark, after error correction, the calculated distance is:

T _d ＝FR _d +C _d +(n-1)·I _d

the part of the distance where errors may occur is FR _d 、C _d That is, by adding a mark, (n-1) & I can be obtained _d Partial error cancellation brings the accumulated error to within an acceptable range.

According to the field actual test, special marks are arranged at 95 different cage positions, wherein 90 positions are correctly positioned, and the positioning accuracy of the abnormal fecal backtracking positioning unit is 94.7%.

S5, the abnormal excrement detection platform for cage chickens is used for displaying the identified abnormal information through the visual platform, and a chicken farm abnormal information user interface is built by using the VUE frame, so that breeders can conveniently grasp chicken disease conditions in time.

The method disclosed by the invention is used for monitoring the health condition of the cage-raised chickens by integrating three functions of identifying and classifying abnormal feces of the cage-raised chickens, tracing and positioning the abnormal feces and correcting errors, and displaying the abnormal monitoring result through a client.

Specifically, the abnormal excrement detection platform for the cage chickens is displayed through the Web interface, and the Web interface is mainly used for visualizing abnormal data, so that the prevention and rapid treatment of diseases by breeding personnel are conveniently guided. The invention builds a set of abnormal feces detection web interfaces by using the VUE framework, and deploys the abnormal feces detection web interfaces on the cloud server. The cloud server is responsible for carrying out data analysis processing on the collected picture data, packaging abnormal results, transmitting the abnormal results to the background database through the data interface, and displaying the abnormal results through the user interface. The abnormal excrement detection platform for the cage-reared chickens can also carry out statistical analysis on historical abnormal data, and the historical abnormal data are displayed in the form of a line graph, a column graph and a pie graph, so that the whole process monitoring of the health of the cage-reared chickens is realized.

As shown in fig. 9, a system website is input in a webpage, a login interface can be seen as shown in fig. 10, after a user name password is input and login is successful, the system home page is entered, the system home page is mainly used for statistically displaying historical abnormal data, abnormal fecal data with specific date inquired about the appointed date is input in the upper left corner, the abnormal fecal data statistics is performed all year down, and a pie chart is used for displaying historical data of different months. The detailed information of the abnormal feces can be checked after clicking to enter the data management interface as shown in fig. 11, including detection time, abnormal types and tracing positions, and the pictures shot by the camera and specific prevention measures can be checked after clicking to check details as shown in fig. 12, and user rights can be added or deleted on the user management interface, so that system management is facilitated.

In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments. In the description of the present specification, reference to the terms "one preferred embodiment," "further," "specifically," "in the present embodiment," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the embodiments of the present specification. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, the different embodiments or examples described in this specification and the features of the different embodiments or examples may be combined and combined by those skilled in the art without contradiction.

Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. The disease early warning and tracing system based on abnormal chicken manure under the stacked cage raising mode is characterized by comprising an image acquisition platform, an abnormal manure tracing system and a cage raising chicken abnormal manure detection platform;

2. The disease early warning and tracing system based on abnormal chicken manure in a stacked cage raising mode according to claim 1, wherein the image acquisition platform comprises an aluminum profile frame, a motion track, an industrial camera and an auxiliary light source;

3. The disease early warning traceability system based on abnormal chicken manure under a laminated cage raising mode according to claim 2, wherein the image acquisition platform is further provided with a camera self-cleaning device, the camera self-cleaning device comprises a fixed base, a direct current motor, a swing arm structure, a cleaning brush and a controller, the fixed base is arranged on one side of an industrial camera, the direct current motor is arranged on one side of the fixed base, the output end of the direct current motor is connected with the swing arm structure, the cleaning brush is arranged on the swing arm structure, and the controller is used for driving the direct current motor to work at regular time.

4. A disease early warning and tracing system based on abnormal chicken manure in a stacked cage raising mode according to any one of claims 1 to 3, wherein the abnormal manure tracing system comprises an abnormal manure recognition and classification unit, an abnormal manure backtracking positioning unit and an error correction unit;

5. The disease early warning and tracing system based on abnormal chicken manure under a stacked cage raising mode of claim 4, wherein the process for establishing the abnormal manure recognition and classification model comprises the following steps:

6. The disease early warning and tracing system based on abnormal chicken manure under a stacked cage raising mode of claim 4, wherein the establishment of the abnormal manure backtracking and positioning model comprises the following steps:

7. The early warning and tracing system for diseases based on abnormal chicken manure under a stacked cage raising mode according to claim 4, wherein the error correction unit takes marks which are manually marked on a manure transfer belt at a certain distance and are known as real references, codes are written in an OpenCV library in pictures acquired in the manure transfer belt operation manure cleaning process to detect and identify the manual marks, moving distance information of the marks and the number of the detected marks are recorded, actual distances between data generated by a comparison algorithm and the marks are used for acquiring measurement errors, and the real distances between the marks and the calculated distances are mutually fused by utilizing a data fusion idea so as to eliminate accumulated errors.

8. The disease early warning and tracing system based on abnormal chicken manure in a stacked cage raising mode of claim 7, wherein the code is written through an OpenCV library, and the mark is detected and identified, comprising: converting the RGB image into HSV color gamut space through format conversion; screening the characteristic colors of the marks; filtering non-target noise in the screened binarized mask image by using image morphology operation and filtering operation; and calculating a minimum circumscribed rectangle through searching the outline, optimizing and combining the multi-target circumscribed rectangle frames aiming at partial chicken manure shielding marks, and positioning the mark positions.

9. A according to claim 8Disease early warning traceability system based on abnormal chicken manure under laminated cage raising mode is characterized in that if abnormal chicken manure X _n+1 Between the nth mark and the (n+1) th mark, after error correction, the distance T between the abnormal feces and the camera is the same _d The method comprises the following steps:

T _d ＝FR _d +C _d +(n-1)·I _d

wherein C is _d For the distance of abnormal feces from the last mark or between cameras, I _d T is the true distance between the marks _d For the distance of abnormal feces from an industrial camera before feces cleaning, namely the retrospective positioning position, FR _d The location is located for the backtracking of the first marker.

10. The disease early warning and tracing system based on abnormal chicken manure under the stacked cage raising mode according to claim 1, wherein the abnormal chicken manure detection platform builds an abnormal manure detection web interface by using a VUE frame and deploys the abnormal manure detection web interface on a cloud server, the cloud server is responsible for carrying out data analysis processing on collected image data, packaging abnormal results and transmitting the abnormal results to a background database through a data interface, and displaying the abnormal results through a user interface.