A whole-process management method and system for broiler slaughtering based on a segmentation model

Abstract

The present invention relates to the technical field of computer-aided agricultural data supervision and processing. A whole-process management method and system for broiler slaughtering based on a segmentation model provided by the present invention. By acquiring the whole-process data of broiler slaughtering and preprocessing the image data, and using the segmentation model for automatic segmentation and recognition, the automatic and intelligent management of the broiler slaughtering process is realized. In addition, the system also introduces a self-optimization and learning mechanism, and optimizes and trains the model through a genetic algorithm, improving the recognition accuracy of the model. The present invention also uses a conditional random field method to post-process the segmentation results, optimize the segmentation effect, extract the region of interest, and perform further processing and analysis, such as calculating the area and shape, etc. Effectively solving the problems of poor image quality and difficult acquisition of labeled data in the traditional broiler slaughtering process, improving the processing efficiency and accuracy, and providing a more efficient and safe product management method for enterprises.

Description

Technical Field

[0001] The present invention relates to the field of computer-aided agricultural data supervision and processing, and particularly to a whole-process management method and system for broiler slaughtering based on a segmentation model. Background Art

[0002] The whole-process management of broiler slaughtering realizes real-time monitoring and data collection of the whole process of broiler slaughtering by introducing advanced digital production equipment and monitoring systems. A perfect information system is established to collect, analyze and process production data in real time, and provide decision-making support for managers.

[0003] The whole-process digital management of broiler slaughtering covers all links from the entry of broilers to the outbound of products. By integrating information technology and automation technology, the intelligence, efficiency and safety of the slaughtering process are realized. The main links include: receiving and tracing of live chickens to ensure the clear origin and safety of each broiler; automated slaughtering, using automated equipment such as stunning, slaughtering, scalding and defeathering to improve slaughtering efficiency; internal organs and head processing, accurately removing and processing internal organs and chicken heads; precooling and segmentation, using a digital control system to quickly cool down, and then automatically segmenting according to preset specifications; quality inspection and classification, using image recognition and AI technology to eliminate unqualified products; and finally packaging, quick-freezing and refrigeration to ensure product quality and extend the shelf life.

[0004] In the whole-process digital management of broiler slaughtering, image processing technology is an important technology in the quality inspection and classification link. However, there are the following technical pain points in the application process of image processing technology in the broiler slaughtering scenario. In order to improve the accuracy of image recognition, a large amount of labeled data is required for model training. However, during the broiler slaughtering process, affected by factors such as environment, light and equipment, the quality of the collected images is poor, such as insufficient or excessive light, blurred images, etc., resulting in time-consuming and labor-intensive workload for the acquisition and collation of labeled data, thus affecting the accuracy of subsequent processing. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention provides a whole-process management method and system for broiler slaughtering based on a segmentation model, which solves the problem that during the broiler slaughtering process, affected by factors such as environment, light and equipment, the quality of the collected images is poor, such as insufficient or excessive light, blurred images, etc., resulting in time-consuming and labor-intensive workload for the acquisition and collation of labeled data, thus affecting the accuracy of subsequent processing.

[0006] To solve the above technical problems, the specific technical solutions of the present invention are as follows:

[0007] In the first aspect, the present invention provides a whole-process management method for broiler slaughtering based on a segmentation model, including:

[0008] In step S101, the whole-process data of broiler slaughter is obtained. The whole-process data of broiler slaughter includes broiler slaughter quality standard data, the operation data of each device during broiler slaughter, and broiler slaughter image data. The broiler slaughter image data is stored in the cloud database, and the broiler slaughter quality standard data is stored in the local server database;

[0009] In step S102, according to the operation data of each device during broiler slaughter, the broiler slaughter process steps are determined, and the broiler slaughter image data is preprocessed to obtain the preprocessed broiler slaughter image data;

[0010] In step S103, a group of recognition models for each process step of broiler slaughter images is established, and the preprocessed broiler slaughter image data is subjected to image recognition through the group of recognition models for each process step of broiler slaughter images to obtain the image recognition results of each process step;

[0011] In step S104, the image data is randomly retrieved from the cloud database, and the broiler sampling inspection image recognition results are matched with the image recognition results of each process step. If the match is unsuccessful, a model recognition fault warning message is generated;

[0012] In step S105, according to the model recognition fault warning message, the broiler slaughter image recognition model corresponding to the fault warning message is determined, the original broiler slaughter data stored in the cloud database is retrieved as the training set data, the image recognition result data corresponding to the original broiler slaughter data is retrieved as the validation set, the error limit value of the broiler slaughter image recognition model corresponding to the fault warning message is determined according to the broiler slaughter quality standard data, and the original broiler slaughter data and the broiler slaughter image recognition model corresponding to the fault warning message are trained and optimized through the genetic algorithm model to obtain the optimized broiler slaughter image recognition model corresponding to the fault warning message, and the optimized broiler slaughter image recognition model corresponding to the fault warning message replaces the broiler slaughter image recognition model corresponding to the fault warning message.

[0013] Furthermore, S102, according to the operation data of each device during broiler slaughter, determines the broiler slaughter process steps, preprocesses the broiler slaughter image data, and obtains the preprocessed broiler slaughter image data, including:

[0014] According to the operation data of each device in broiler slaughtering, determine the broiler slaughtering process steps. Construct an image preprocessing module group according to the broiler slaughtering process steps. Sort the broiler slaughtering image data stored in the cloud database according to the image generation time to generate a broiler slaughtering image data sequence list. Match the broiler slaughtering image data sequence list with the image preprocessing modules in the image preprocessing module group to obtain the local image preprocessing module corresponding to the broiler slaughtering image data. The broiler slaughtering image data is preprocessed by the corresponding local image preprocessing module to obtain the preprocessed broiler slaughtering image data.

[0015] Further, S103: Establish a recognition model group for each process step of broiler slaughtering images. Perform image recognition on the preprocessed broiler slaughtering image data through the recognition model group for each process step of broiler slaughtering images to obtain the image recognition results for each process step, including:

[0016] The recognition model group for each process step of broiler slaughtering images includes a hanging live chicken image recognition model, a broiler quality image recognition model, a standardized slaughter process image recognition model, a scalding and defeathering image recognition model, an internal organs and head processing image recognition model, a product quality image recognition model, a product classification image recognition model, and a product packaging image recognition model. Extract the attribute features of the preprocessed broiler slaughtering image data to obtain the broiler slaughtering steps corresponding to the broiler slaughtering images, and establish corresponding broiler slaughtering image data step labels for the preprocessed broiler slaughtering image data. Match the broiler slaughtering image data step labels with the recognition model group for each process step of broiler slaughtering images to obtain the matching results of the broiler slaughtering image recognition model group. After transmitting the preprocessed broiler slaughtering image data to the corresponding broiler slaughtering image recognition model, perform image segmentation on the preprocessed broiler slaughtering image data through the segmentation model, and recognize the segmented broiler slaughtering image through the corresponding broiler slaughtering image recognition model to obtain the image recognition results for each process step.

[0017] Further, S104: Randomly retrieve the image data retrieved from the cloud database, and match the broiler sampling inspection image recognition results with the image recognition results for each process step. If the match is unsuccessful, generate a model recognition failure warning message, including:

[0018] Randomly retrieve the generation time of the original data in the pre - processed broiler slaughter image data, retrieve the image data from the cloud database according to the original data generation time to obtain the sampled broiler slaughter images, determine the original image detection standard based on the broiler slaughter quality standard data, pre - process the sampled broiler slaughter images according to the original image detection standard to obtain the pre - processed sampled broiler slaughter images, and then match the pre - processed sampled broiler slaughter images with the pre - processed broiler slaughter image data. If the match is successful, substitute the pre - processed sampled broiler slaughter images into the broiler slaughter image recognition model group for image recognition to obtain the broiler sampling image recognition result, and match the broiler sampling image recognition result with the image recognition results of each process step. If the match is not successful, generate a model recognition fault warning message.

[0019] Furthermore, the present invention provides a method for the whole - process management of broiler slaughter based on a segmentation model, which further includes:

[0020] Determine the information of the equipment currently used in the whole process of broiler slaughter through the operation data of each equipment in broiler slaughter, match the currently used equipment information in the preset broiler slaughter process flow to determine the broiler slaughter process steps, screen the steps that need image recognition in the broiler slaughter process steps to obtain the process steps to be image - recognized, and establish an image pre - processing module group according to the process steps to be image - recognized.

[0021] Furthermore, the present invention provides a method for the whole - process management of broiler slaughter based on a segmentation model, which further includes:

[0022] The image pre - processing module group constructed for the broiler slaughter process steps includes a pre - processing module for the image of live chickens hanging, a pre - processing module for the image of broiler quality, a pre - processing module for the standardized image of the slaughter process, a pre - processing module for the scalding and defeathering image, a pre - processing module for the image of internal organs and head processing, a pre - processing module for the image of product quality, a pre - processing module for the image of product classification, and a pre - processing module for the image of product packaging.

[0023] Furthermore, the present invention provides a method for the whole - process management of broiler slaughter based on a segmentation model, which further includes:

[0024] Establish a recognition model group for each process step of broiler slaughter images according to the broiler slaughter process steps. The recognition model group for each process step of broiler slaughter images is used to perform feature annotation on the image data generated in each step of the broiler slaughter process steps, and transmit the annotated images to the corresponding segmentation model of the recognition model group for each process step of broiler slaughter images for segmented image recognition.

[0025] Furthermore, the present invention provides a method for the whole - process management of broiler slaughter based on a segmentation model, which further includes:

[0026] Before inputting the annotated image into the segmentation model, denoise, enhance the contrast, and adjust the brightness of the annotated image to obtain the preprocessed annotated image;

[0027] Select the U-Net convolutional neural network as the training model, and use the preprocessed annotated image to train the model to obtain a trained segmentation model;

[0028] Input the preprocessed broiler slaughter image data into the trained segmentation model, and the model automatically segments the image into different regions, including chicken, chicken skin, bones, chicken head, chicken legs, and chicken necks;

[0029] Post-process the results output by the segmentation model, optimize the segmentation results through the conditional random field method, extract the regions of interest, and perform further processing and analysis on them to calculate the area and shape.

[0030] In a second aspect, the present invention provides a whole-process management system for broiler slaughter based on a segmentation model, applying the whole-process management method for broiler slaughter based on a segmentation model, including: a local server, a cloud server, a device end, and a background control end. The local server establishes communication connections with the cloud server, the device end, and the background control end. The local server includes:

[0031] An acquisition unit that acquires the whole-process data of broiler slaughter. The whole-process data of broiler slaughter includes broiler slaughter quality standard data, the operation data of each device in broiler slaughter, and broiler slaughter image data. Store the broiler slaughter image data in the cloud database and store the broiler slaughter quality standard data in the local server database;

[0032] An image preprocessing unit that determines the broiler slaughter process steps according to the operation data of each device in broiler slaughter, and preprocesses the broiler slaughter image data to obtain the preprocessed broiler slaughter image data;

[0033] An image segmentation and recognition unit that establishes a recognition model group for each process step of broiler slaughter images, and performs image recognition on the preprocessed broiler slaughter image data through the recognition model group for each process step of broiler slaughter images to obtain the image recognition results for each process step;

[0034] A quality inspection unit that randomly retrieves image data from the cloud database, matches the broiler sampling inspection image recognition results with the image recognition results for each process step. If the match is unsuccessful, a model recognition failure warning message is generated;

[0035] The model optimization unit identifies the fault warning information according to the model, determines the broiler slaughter image recognition model corresponding to the fault warning information, retrieves the original broiler slaughter data stored in the cloud database as the training set data, retrieves the image recognition result data corresponding to the original broiler slaughter data as the validation set, determines the error limit value of the broiler slaughter image recognition model corresponding to the fault warning information according to the chicken slaughter quality standard data, trains and optimizes the original broiler slaughter data and the broiler slaughter image recognition model corresponding to the fault warning information through the genetic algorithm model, obtains the optimized broiler slaughter image recognition model corresponding to the fault warning information, and replaces the broiler slaughter image recognition model corresponding to the fault warning information with the optimized broiler slaughter image recognition model.

[0036] The beneficial effects of the present invention are as follows:

[0037] Improve image quality: By preprocessing the collected broiler slaughter image data through multiple groups of image preprocessing modules, the image quality can be effectively improved, the influence of factors such as environment, light, and equipment on the image can be reduced, and a higher-quality data basis can be provided for subsequent image segmentation and recognition.

[0038] Automation and intelligent processing: The present invention automatically segments the preprocessed image through the segmentation model and combines the image recognition models of each process step for recognition, realizing the automation and intelligent processing of the entire broiler slaughter process, and improving the processing efficiency and accuracy.

[0039] Real-time monitoring and sampling inspection mechanism: By randomly retrieving the preprocessed broiler slaughter image data for sampling inspection and comparing it with the original image for verification, the accuracy of image recognition can be monitored in real time, and possible errors can be detected and corrected in a timely manner.

[0040] Self-optimization and learning ability: When a model recognition fault is found, the system can automatically perform model optimization training to improve the recognition ability of the model, enabling the system to have the ability of self-optimization and learning.

[0041] Improve management efficiency: Through the whole-process management method of the present invention, enterprises can manage the broiler slaughter process more efficiently, reduce manual intervention, lower costs, and improve product quality and safety at the same time.

[0042] Fault warning and timely processing: When the system detects a model recognition fault, it will generate a warning information in a timely manner to remind the management personnel to process it, ensuring the smooth progress of the entire slaughter process.

[0043] In summary, the whole-process management method and system for broiler slaughter based on the segmentation model provided by the present invention can not only improve the accuracy and efficiency of image recognition, but also realize automated and intelligent management. Brief Description of the Drawings

[0044] To more clearly illustrate the technical solution of the present invention, the accompanying drawings required for use in the embodiments will be briefly introduced below. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0045] Figure 1 It is a schematic diagram of the process method provided by the embodiment of the present invention. Specific embodiments

[0046] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention. The technical solutions provided by each embodiment of the present invention will be described in detail below with reference to the drawings.

[0047] To better understand the objectives of the present invention, the present invention will be further described in detail below.

[0048] In the first aspect, please refer to Figure 1 , the present invention provides a whole-process management method for broiler slaughtering based on a segmentation model, including:

[0049] In step S101, obtain the whole-process data of broiler slaughtering. The whole-process data of broiler slaughtering includes broiler slaughter quality standard data, operation data of each device during broiler slaughtering, and broiler slaughter image data. Store the broiler slaughter image data in the cloud database and store the broiler slaughter quality standard data in the local server database;

[0050] In step S101, first comprehensively obtain the relevant data of the whole process of broiler slaughtering. These data mainly include three aspects: broiler slaughter quality standard data, operation data of each device during broiler slaughtering, and broiler slaughter image data.

[0051] Broiler slaughter quality standard data: These data are important bases for evaluating the broiler slaughter process and product quality. It may include various quality standards such as the weight range of broilers, meat color, fat content, and microbial indicators. These standard data will be stored in the local server database for quick access and reference.

[0052] Operation data of each device in broiler slaughtering: This type of data reflects the operating status and parameters of each device during the slaughtering process, such as the speed of the slaughter line, the blade temperature of the cutting machine, the temperature and time settings of the cooling equipment, etc. These data are crucial for monitoring the efficiency and safety of the slaughtering process.

[0053] Broiler slaughtering image data: These are real-time pictures or videos of the broiler slaughtering process captured by cameras or other imaging devices. The image data can provide intuitive visual information to help managers understand every detail of the slaughtering process, thus ensuring operation specifications and product quality. This type of data will be stored in the cloud database for long-term preservation and remote access.

[0054] The advantage of storing broiler slaughtering image data in the cloud database is that it can achieve centralized management and remote access of the data. At the same time, cloud storage also provides better data backup and recovery capabilities. While storing chicken slaughter quality standard data in the local server database is to ensure that these key quality standard information can be accessed quickly and stably when needed, so as to monitor and adjust the slaughtering process in a timely manner.

[0055] In step S102, according to the operation data of each device in broiler slaughtering, determine the broiler slaughtering process steps, and preprocess the broiler slaughtering image data to obtain the preprocessed broiler slaughtering image data;

[0056] In step S102, the operation is mainly divided into two parts: First, determine the current slaughtering process steps according to the operation data of each device during the broiler slaughtering process. These data may include parameters such as the operating status, speed, and temperature of the device. By analyzing these data, the system can judge the current slaughtering process steps.

[0057] Next is the preprocessing stage of broiler slaughtering image data. This process first obtains the broiler slaughtering image data from the cloud database and sorts it according to the time sequence of image generation to generate a sequence list of broiler slaughtering image data. Then, the system constructs a group of image preprocessing modules that match the broiler slaughtering process steps. These modules perform specific image preprocessing operations for different slaughtering steps.

[0058] Subsequently, the system matches the sequence list of broiler slaughtering image data with the modules in the group of image preprocessing modules to find the preprocessing module corresponding to each segment of image data. Finally, each segment of broiler slaughtering image data will be processed by its corresponding preprocessing module to obtain the preprocessed broiler slaughtering image data. These preprocessed data will be more conducive to subsequent analysis and recognition, improving the accuracy and efficiency of the entire system.

[0059] In step S103, an identification model group for each process step of broiler slaughter images is established. The pre - processed broiler slaughter image data is subjected to image recognition through the identification model group for each process step of broiler slaughter images, and the image recognition results for each process step are obtained;

[0060] In step S103, first, an identification model group for each process step of broiler slaughter images is established. These models include, but are not limited to, image recognition models such as hanging live chickens, broiler quality, standardization of the slaughter process, scalding and defeathering, internal organ and head processing, product quality, product classification, and product packaging.

[0061] Next, the pre - processed broiler slaughter image data is subjected to image recognition using these model groups. This process first involves extracting attribute features to determine the broiler slaughter step corresponding to the image and tagging it with the corresponding data step label. Then, these labels are matched with the previously established identification model group for each process step of broiler slaughter images to find the most suitable identification model.

[0062] After that, the pre - processed broiler slaughter image data is transmitted to the corresponding identification model, and image segmentation is performed through a dedicated segmentation model. Finally, the segmented images are subjected to detailed recognition through the corresponding identification models, thereby obtaining accurate image recognition results for each process step. This series of processes ensures the accuracy and traceability of each link in the broiler slaughter process.

[0063] In step S104, image data is randomly retrieved from the cloud database, and the image recognition results of broiler sampling inspection are matched with the image recognition results of each process step. If the match is unsuccessful, a model recognition fault warning message is generated;

[0064] The main content of step S104 is to verify the model and give a fault warning. First, the system randomly retrieves broiler slaughter image data from the cloud database. Then, the image recognition results of broiler sampling inspection in these data are matched with the image recognition results of each previous process step. If the match is unsuccessful, it means that there may be a problem with the model recognition. At this time, the system generates a model recognition fault warning message. This warning message is very important as it reminds users or managers to check or optimize the current model to ensure the accuracy of recognition and the stable operation of the system.

[0065] In step S105, based on the model to identify the fault warning information, determine the broiler slaughter image recognition model corresponding to the fault warning information, retrieve the original broiler slaughter data stored in the cloud database as the training set data, retrieve the image recognition result data corresponding to the original broiler slaughter data as the validation set, determine the error limit value of the broiler slaughter image recognition model corresponding to the fault warning information according to the chicken slaughter quality standard data, and train and optimize the original broiler slaughter data and the broiler slaughter image recognition model corresponding to the fault warning information through the genetic algorithm model to obtain the optimized broiler slaughter image recognition model corresponding to the fault warning information, and replace the broiler slaughter image recognition model corresponding to the fault warning information with the optimized broiler slaughter image recognition model corresponding to the fault warning information.

[0066] Specifically, S102: Determine the broiler slaughter process steps according to the operation data of each device in broiler slaughter, preprocess the broiler slaughter image data to obtain the preprocessed broiler slaughter image data, including:

[0067] Determine the broiler slaughter process steps according to the operation data of each device in broiler slaughter, construct an image preprocessing module group according to the broiler slaughter process steps, sort the broiler slaughter image data stored in the cloud database by the image generation time to generate a broiler slaughter image data sequence table, match the broiler slaughter image data sequence table with the image preprocessing modules in the image preprocessing module group to obtain the local image preprocessing module corresponding to the broiler slaughter image data, and preprocess the broiler slaughter image data through the corresponding local image preprocessing module to obtain the preprocessed broiler slaughter image data.

[0068] Specifically, S103: Establish a recognition model group for each process step of the broiler slaughter image, and perform image recognition on the preprocessed broiler slaughter image data through the recognition model group for each process step of the broiler slaughter image to obtain the image recognition results of each process step, including:

[0069] The identification model group for each process step of broiler slaughter images includes a hanging chicken image recognition model, a broiler quality image recognition model, a standardized image recognition model for the slaughter process, a scalding and defeathering image recognition model, an internal organ and head processing image recognition model, a product quality image recognition model, a product classification image recognition model, and a product packaging image recognition model. Extract the attribute features of the preprocessed broiler slaughter image data to obtain the broiler slaughter steps corresponding to the broiler slaughter images, and establish corresponding step labels for the broiler slaughter image data for the preprocessed broiler slaughter image data. Match the step labels of the broiler slaughter image data with the identification model group for each process step of the broiler slaughter images to obtain the matching result of the broiler slaughter image recognition model group. After transmitting the preprocessed broiler slaughter image data to the corresponding broiler slaughter image recognition model, perform image segmentation on the preprocessed broiler slaughter image data through a segmentation model, and identify the segmented broiler slaughter image through the corresponding broiler slaughter image recognition model to obtain the image recognition results for each process step.

[0070] Specifically, S104: Randomly retrieve the image data from the cloud database, and match the recognition results of the broiler sampling inspection images with the image recognition results of each process step. If the match is unsuccessful, generate a model recognition failure warning message, including:

[0071] Randomly retrieve the generation time of the original data in the preprocessed broiler slaughter image data, retrieve the image data from the cloud database according to the generation time of the original data to obtain the sampled broiler slaughter images, determine the original image detection standard based on the broiler slaughter quality standard data, preprocess the sampled broiler slaughter images according to the original image detection standard to obtain the preprocessed sampled broiler slaughter images, and then match the preprocessed sampled broiler slaughter images with the preprocessed broiler slaughter image data. If the match is successful, substitute the preprocessed sampled broiler slaughter images into the broiler slaughter image recognition model group for image recognition to obtain the recognition results of the broiler sampling inspection images, and match the recognition results of the broiler sampling inspection images with the image recognition results of each process step. If the match is unsuccessful, generate a model recognition failure warning message.

[0072] Specifically, the present invention provides a method for managing the whole process of broiler slaughter based on a segmentation model, further including:

[0073] Determine the equipment information currently used in the whole process of broiler slaughter through the operation data of each device in broiler slaughter, match the currently used equipment information in the preset broiler slaughter process flow to determine the broiler slaughter process steps, screen the steps that need image recognition from the broiler slaughter process steps to obtain the process steps to be image recognized, and establish an image preprocessing module group according to the process steps to be image recognized.

[0074] Specifically, the present invention provides a whole-process management method for broiler slaughtering based on a segmentation model, further including:

[0075] The image preprocessing module group for constructing broiler slaughtering process steps includes a preprocessing module for the image of live chickens being hung, a preprocessing module for the image of broiler quality, a preprocessing module for the standardized image of the slaughtering process, a preprocessing module for the scalding and defeathering image, a preprocessing module for the image of internal organs and head processing, a preprocessing module for the product quality image, a preprocessing module for the product classification image, and a preprocessing module for the product packaging image.

[0076] Specifically, the present invention provides a whole-process management method for broiler slaughtering based on a segmentation model, further including:

[0077] According to the broiler slaughtering process steps, an identification model group for each process step of broiler slaughtering images is established. The identification model group for each process step of broiler slaughtering images is used to perform feature annotation on the image data generated in each step of the broiler slaughtering process, and transmit the annotated image to the corresponding segmentation model of the identification model group for each process step of broiler slaughtering images for segmented image recognition.

[0078] Specifically, the present invention provides a whole-process management method for broiler slaughtering based on a segmentation model, further including:

[0079] Before inputting the annotated image into the segmentation model, denoise, enhance the contrast, and adjust the brightness of the annotated image to obtain the preprocessed annotated image;

[0080] Select the U-Net convolutional neural network as the training model, and use the preprocessed annotated image to train the model to obtain a trained segmentation model;

[0081] Input the preprocessed broiler slaughtering image data into the trained segmentation model, and the model automatically segments the image into different regions, including chicken meat, chicken skin, bones, chicken head, chicken legs, and chicken necks;

[0082] Perform post-processing on the results output by the segmentation model, optimize the segmentation results through the conditional random field method, extract the regions of interest, and perform further processing and analysis on them to calculate the area and shape.

[0083] In the second aspect, the present invention provides a whole-process management system for broiler slaughtering based on a segmentation model, applying the whole-process management method for broiler slaughtering based on a segmentation model, including: a local server side, a cloud server side, a device side, and a background control side. The local server side establishes communication connections with the cloud server side, the device side, and the background control side. The local server side includes:

[0084] An acquisition unit acquires the whole-process data of broiler slaughtering. The whole-process data of broiler slaughtering includes broiler slaughtering quality standard data, the operation data of each device during broiler slaughtering, and broiler slaughtering image data. The broiler slaughtering image data is stored in the cloud database, and the broiler slaughtering quality standard data is stored in the local server database;

[0085] An image preprocessing unit determines the broiler slaughtering process steps according to the operation data of each device during broiler slaughtering, and preprocesses the broiler slaughtering image data to obtain preprocessed broiler slaughtering image data;

[0086] An image segmentation and recognition unit establishes a recognition model group for each process step of broiler slaughtering images, and performs image recognition on the preprocessed broiler slaughtering image data through the recognition model group for each process step of broiler slaughtering images to obtain the image recognition results of each process step;

[0087] A quality inspection unit randomly retrieves image data from the cloud database, matches the broiler sampling inspection image recognition results with the image recognition results of each process step. If the matching is unsuccessful, a model recognition fault warning message is generated;

[0088] A model optimization unit determines the broiler slaughtering image recognition model corresponding to the fault warning message according to the model recognition fault warning message, retrieves the original broiler slaughter data stored in the cloud database as the training set data, retrieves the image recognition result data corresponding to the original broiler slaughter data as the validation set, determines the error limit value of the broiler slaughtering image recognition model corresponding to the fault warning message according to the broiler slaughtering quality standard data, trains and optimizes the original broiler slaughter data and the broiler slaughtering image recognition model corresponding to the fault warning message through a genetic algorithm model to obtain the optimized broiler slaughtering image recognition model corresponding to the fault warning message, and replaces the broiler slaughtering image recognition model corresponding to the fault warning message with the optimized broiler slaughtering image recognition model corresponding to the fault warning message.

[0089] The present invention solves the problem of image quality problems caused by factors such as environment, light, and equipment during broiler slaughtering in the following way:

[0090] Image preprocessing: First, the original broiler slaughtering image data is processed by a special image preprocessing module. The purpose of this step is to improve the image quality, specifically by adjusting brightness, contrast, reducing noise, etc., to reduce the interference caused by the environment, light, and equipment to the image.

[0091] Segmentation model and image recognition: The preprocessed image will then be automatically segmented by the segmentation model and recognized in combination with the image recognition model established for each process step of broiler slaughtering. This method not only improves the accuracy of image recognition but also realizes the automation and intelligent processing of the whole process of broiler slaughtering.

[0092] System self-optimization and learning: The system also introduces mechanisms for self-optimization and learning. When a failure in model recognition is detected, it can automatically trigger the model optimization training process, and adjust the model through techniques such as genetic algorithms to improve its recognition ability.

[0093] Through the above methods, the present invention effectively solves the problem of poor image quality, reduces the difficulty and workload of obtaining and organizing labeled data, and thus improves the accuracy of subsequent image recognition.

[0094] The specific implementation manners of the present invention cover multiple aspects such as algorithms, model selection, training processes, and implementation steps, which are as follows:

[0095] Algorithm and model selection: The genetic algorithm model is used for training optimization. A recognition model group for each process step of broiler slaughter images is specifically established.

[0096] Training process: With the help of the genetic algorithm model, the original broiler slaughter data and the broiler slaughter image recognition model corresponding to the fault warning information are trained and optimized, aiming to obtain an optimized model with better performance.

[0097] After training is completed, the optimized new model replaces the broiler slaughter image recognition model corresponding to the original fault warning information to improve the recognition accuracy of the system.

[0098] The implementation steps are as follows;

[0099] Step S101: First, the system acquires data on the entire process of broiler slaughter, including quality standard data, equipment operation data, and image data of broiler slaughter. The acquired data will be properly stored in the cloud and local databases.

[0100] Step S102: Next, according to the equipment operation data during broiler slaughter, the system determines the specific slaughter process steps and performs corresponding preprocessing on the acquired image data to improve the image quality.

[0101] Step S103: Subsequently, the system establishes a recognition model group for each process step and uses these models to perform accurate image recognition on the preprocessed image data.

[0102] Step S104: To monitor the recognition accuracy of the system, the system randomly retrieves image data from the cloud database for spot checks. These spot-check data are matched and verified with the previous recognition results. If a mismatch is found, the system generates a fault warning information.

[0103] Step S105: When a fault warning message is received, the system will immediately start the genetic algorithm model to train and optimize the image recognition model with problems. After the optimization is completed, the new model will replace the original faulty model to ensure the continuous and efficient operation of the system.

[0104] In the present invention, the use of the genetic algorithm model follows the following detailed steps:

[0105] Determine the model corresponding to the fault warning message: The system will first identify the fault warning message according to the model and accurately locate the broiler slaughter image recognition model that needs to be optimized.

[0106] Retrieve the training set and validation set data: Subsequently, the system will retrieve the original broiler slaughter data from the cloud database, and these data will be used as the training set. At the same time, the image recognition result data corresponding to these original data will be used as the validation set.

[0107] Set the error limit value: According to the quality standard data of broiler slaughter, the system will set an error limit value for the image recognition model corresponding to the fault warning message, which is to ensure that the performance of the model after optimization meets the expected standard.

[0108] Model training and optimization: Next, use the genetic algorithm model to train the training set data and optimize the broiler slaughter image recognition model corresponding to the fault warning message. This process aims to improve the recognition accuracy and efficiency of the model.

[0109] Model replacement: Finally, the optimized new model will replace the broiler slaughter image recognition model corresponding to the original fault warning message, thereby ensuring the continuous improvement of the recognition performance of the system.

[0110] Through these steps, the present invention can automatically optimize and train the model with poor recognition performance, significantly improve the recognition ability of the model, and enable the system to have strong self-optimization and learning abilities.

[0111] In the present invention, the conditional random field method is used to optimize the results output by the segmentation model. Specifically, it can finely adjust the initially segmented different regions (such as chicken meat, chicken skin, bones, chicken head, chicken legs, chicken necks, etc.), thereby significantly improving the segmentation accuracy. After the optimization is completed, this method also helps us extract the regions of interest, such as parts of chicken meat with specific quality or chicken legs of specific size, etc. Finally, we conduct in-depth processing and analysis on these extracted regions, calculate their areas, shapes, and other related features, providing strong support for subsequent quality detection and classification. In this way, the conditional random field method enhances the practicality and accuracy of the present invention.

[0112] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these changes and modifications. The above-described embodiments of the present invention do not constitute a limitation on the protection scope of the present invention.

Claims

1. A method for managing the whole process of broiler slaughtering based on a segmentation model, characterized in that: include: S101, obtaining data of the entire broiler slaughtering process, including broiler slaughtering quality standard data, operation data of each device in broiler slaughtering, and broiler slaughtering image data, storing the broiler slaughtering image data in a cloud database, and storing the broiler slaughtering quality standard data in a local server database; S102, determining the broiler slaughtering process steps according to the operation data of each device in the broiler slaughtering, preprocessing the broiler slaughtering image data, and obtaining the preprocessed broiler slaughtering image data; S103, establishing a recognition model group for each process step of broiler slaughtering images, and performing image recognition on the pre-processed broiler slaughtering image data through the recognition model group for each process step of broiler slaughtering images to obtain image recognition results for each process step; S104, randomly retrieve image data from the cloud database, match the broiler random inspection image recognition results with the image recognition results of each process step, and if the match is unsuccessful, generate model recognition fault warning information; S105, identifying the fault warning information according to the model, determining the broiler slaughter image recognition model corresponding to the fault warning information, retrieving the original broiler slaughter data stored in the cloud database as the training set data, retrieving the image recognition result data corresponding to the original broiler slaughter data as the verification set, determining the error limit value of the broiler slaughter image recognition model corresponding to the fault warning information according to the chicken slaughter quality standard data, training and optimizing the original broiler slaughter data and the broiler slaughter image recognition model corresponding to the fault warning information through a genetic algorithm model, obtaining an optimized broiler slaughter image recognition model corresponding to the fault warning information, and replacing the broiler slaughter image recognition model corresponding to the fault warning information with the optimized broiler slaughter image recognition model corresponding to the fault warning information; S102, determining the broiler slaughtering process steps according to the operation data of each device in broiler slaughtering, preprocessing the broiler slaughtering image data, and obtaining the preprocessed broiler slaughtering image data, including: According to the operation data of each device in broiler slaughtering, the broiler slaughtering process steps are determined, and an image preprocessing module group is constructed according to the broiler slaughtering process steps. The broiler slaughtering image data stored in the cloud database are sorted according to the time of image generation to generate a broiler slaughtering image data sequence table, and the broiler slaughtering image data sequence table is matched with the image preprocessing module in the image preprocessing module group to obtain a local image preprocessing module corresponding to the broiler slaughtering image data. The broiler slaughtering image data is image preprocessed by the corresponding local image preprocessing module to obtain the preprocessed broiler slaughtering image data; The operation data of each device in broiler slaughtering is used to determine the equipment information currently used in the whole process of broiler slaughtering, the equipment information currently used is matched in the preset broiler slaughtering process flow, the broiler slaughtering process steps are determined, the steps that need to be image recognized are screened in the broiler slaughtering process steps, the process steps to be image recognized are obtained, and an image preprocessing module group is established according to the process steps to be image recognized; The image preprocessing module group for constructing the broiler slaughtering process steps includes a live chicken hanging image preprocessing module, a broiler quality image preprocessing module, a slaughtering process standardization image preprocessing module, a scalding and hair removal image preprocessing module, an viscera and head processing image preprocessing module, a product quality image preprocessing module, a product classification image preprocessing module, and a product packaging image preprocessing module; S103, establishing a recognition model group for each process step of broiler slaughtering images, and performing image recognition on the pre-processed broiler slaughtering image data through the recognition model group for each process step of broiler slaughtering images to obtain image recognition results for each process step, including: The recognition model group of each process step of broiler slaughtering image includes a live chicken hanging image recognition model, a broiler quality image recognition model, a slaughtering process standardization image recognition model, a scalding and depilation image recognition model, an viscera and head processing image recognition model, a product quality image recognition model, a product classification image recognition model and a product packaging image recognition model. The attribute features of the pre-processed broiler slaughtering image data are extracted to obtain the broiler slaughtering steps corresponding to the broiler slaughtering images, and corresponding broiler slaughtering image data step labels are established for the pre-processed broiler slaughtering image data. The broiler slaughtering image data step labels are matched with the recognition model groups of each process step of the broiler slaughtering image to obtain the matching results of the broiler slaughtering image recognition model group. After the pre-processed broiler slaughtering image data is transmitted to the corresponding broiler slaughtering image recognition model, the pre-processed broiler slaughtering image data is segmented by a segmentation model, and the segmented broiler slaughtering image is recognized by the corresponding broiler slaughtering image recognition model to obtain the image recognition results of each process step. According to the broiler slaughter process steps, a recognition model group for each process step of broiler slaughter images is established. The recognition model group for each process step of broiler slaughter images is used to perform feature annotation on the image data generated by each step in the broiler slaughter process steps, and transmit the annotated image to the segmentation model corresponding to the recognition model group for each process step of broiler slaughter images for image recognition after segmentation; Before inputting the labeled image into the segmentation model, the labeled image is denoised, contrast enhanced, and brightness adjusted to obtain a preprocessed labeled image; Select U-Net convolutional neural network as the training model, use the preprocessed labeled images to train the model, and obtain a trained segmentation model; The pre-processed broiler slaughter image data is input into the trained segmentation model, and the model automatically segments the image into different regions, including chicken meat, chicken skin, bones, chicken head, chicken legs, and chicken neck; Post-process the output of the segmentation model, optimize the segmentation results through the conditional random field method, extract the region of interest, and further process and analyze it to calculate the area and shape; S104, randomly retrieve image data from the cloud database, match the broiler random inspection image recognition results with the image recognition results of each process step, and if the match is unsuccessful, generate model recognition fault warning information, including: The original data generation time in the preprocessed broiler slaughter image data is randomly retrieved, and the image data is retrieved from the cloud database according to the original data generation time to obtain the sampled broiler slaughter image, and the original image detection standard is determined by the broiler slaughter quality standard data. The sampled broiler slaughter image is preprocessed according to the original image detection standard to obtain the preprocessed sampled broiler slaughter image, and the preprocessed sampled broiler slaughter image is matched with the preprocessed broiler slaughter image data. If the match is successful, the preprocessed sampled broiler slaughter image is substituted into the broiler slaughter image recognition model group for image recognition to obtain the broiler sampling image recognition result, and the broiler sampling image recognition result is matched with the image recognition result of each process step. If the match is unsuccessful, a model recognition fault warning information is generated.

2. A broiler slaughtering whole process management system based on a segmentation model, using the broiler slaughtering whole process management method based on a segmentation model as claimed in claim 1, characterized in that: include: The local server, cloud server, device and backend control end establish communication connections with the cloud server, device and backend control end. The local server includes: An acquisition unit acquires data of the entire broiler slaughtering process, including broiler slaughtering quality standard data, operation data of each device in broiler slaughtering, and broiler slaughtering image data, stores the broiler slaughtering image data in a cloud database, and stores the broiler slaughtering quality standard data in a local server database; An image preprocessing unit determines the broiler slaughtering process steps according to the operation data of each device in the broiler slaughtering, preprocesses the broiler slaughtering image data, and obtains the preprocessed broiler slaughtering image data; An image segmentation and recognition unit is used to establish a recognition model group for each process step of broiler slaughtering images, and perform image recognition on the preprocessed broiler slaughtering image data through the recognition model group for each process step of broiler slaughtering images to obtain image recognition results for each process step; The quality inspection unit randomly retrieves image data from the cloud database and matches the image recognition results of the broiler random inspection with the image recognition results of each process step. If the match is unsuccessful, a model recognition fault warning information is generated; A model optimization unit identifies fault warning information according to the model, determines a broiler slaughter image recognition model corresponding to the fault warning information, retrieves original broiler slaughter data stored in a cloud database as training set data, retrieves image recognition result data corresponding to the original broiler slaughter data as a verification set, determines an error limit value of the broiler slaughter image recognition model corresponding to the fault warning information according to chicken slaughter quality standard data, trains and optimizes the original broiler slaughter data and the broiler slaughter image recognition model corresponding to the fault warning information through a genetic algorithm model, obtains an optimized broiler slaughter image recognition model corresponding to the fault warning information, and replaces the broiler slaughter image recognition model corresponding to the fault warning information with the optimized broiler slaughter image recognition model corresponding to the fault warning information.

Images