Poultry number counting method and system

Abstract

The invention discloses a poultry quantity counting method and system, and belongs to the technical field of quantity counting methods and systems based on a deep neural network, and the method comprises the steps: obtaining video data of a loading and unloading area; performing frame extraction processing on the video data to obtain a static image; each frame of static image is analyzed based on an artificial intelligence target detection algorithm to identify a to-be-detected target in the static image, and the to-be-detected target comprises poultry; identifying an identification result on the static image; performing de-duplication processing on the identified static images of different frames based on a de-duplication target detection algorithm so as to remove the to-be-detected target which is repeatedly identified; and counting the poultry subjected to the de-weighting treatment to obtain the quantity of the loaded and unloaded poultry. According to the invention, the detection is not interfered by the high-humidity and high-dust environment of the slaughter house area, and accurate counting of the number of poultry can be ensured.

Description

Technical Field

[0001] The present invention relates to the field of poultry counting, and in particular to a poultry counting method and system. Background Art

[0002] In the daily operations of slaughterhouses, accurately tracking the number of poultry slaughtered each day is crucial. However, existing counting methods rely on installing infrared counting devices (e.g., infrared sensors) on hanging chains. These devices detect when poultry blocks the infrared sensor's beam. However, the loading and unloading areas of slaughterhouses are subject to extremely harsh environmental conditions. High humidity and high dust levels often prevent infrared counting devices from functioning properly, significantly reducing their lifespan. This results in extremely high errors in their counting function, or even failure, causing significant challenges for factory management.

[0003] This technical challenge is an urgent problem that needs to be solved for slaughterhouses. Summary of the Invention

[0004] The present invention aims to provide a method and system for accurately counting the number of poultry without being disturbed by the high humidity and high dust environment in the slaughterhouse.

[0005] To achieve the above object, the technical solution of the present invention is:

[0006] A poultry counting method comprising:

[0007] Obtain video data of the loading and unloading area;

[0008] Performing frame extraction processing on the video data to obtain a static image;

[0009] Analyzing each frame of the static image based on an artificial intelligence target detection algorithm to identify a target to be detected in the static image, wherein the target to be detected includes poultry;

[0010] Marking the recognition result on the static image;

[0011] Deduplication is performed on the static images of different frames after identification based on the deduplication target detection algorithm to remove the repeated identified targets to be detected;

[0012] The poultry were counted after weight removal to obtain the number of poultry loaded and unloaded.

[0013] Optionally, video data of the loading and unloading area is obtained based on a high-definition camera.

[0014] Optionally, the performing frame extraction processing on the video data to obtain a static image includes:

[0015] Frame extraction is performed on the video data to convert continuous video data into a series of static images.

[0016] Optionally, analyzing each frame of the static image based on an artificial intelligence target detection algorithm to identify the target to be detected in the static image includes:

[0017] Inputting each frame of the static image into the target detection model;

[0018] The target detection model identifies and locates each frame of the static image based on the artificial intelligence target detection algorithm.

[0019] Optionally, marking the recognition result on the static image includes:

[0020] The recognition result is drawn back onto the static image in the form of a graphic or a mark.

[0021] Optionally, the deduplication target detection algorithm is used to perform deduplication processing on static images of different frames after marking to remove repeatedly identified targets to be detected, including: using a Kalman filter and non-maximum suppression to perform deduplication processing on static images of different frames after marking to remove repeatedly identified targets to be detected.

[0022] Optionally, in the loading and unloading area, the poultry unloaded by the transport vehicle are hung on the hook; the target to be inspected also includes an empty hook; after the deduplication process, the method further includes:

[0023] Count the empty hooks after deduplication;

[0024] If the continuous count of the empty hooks exceeds a preset threshold, the poultry after the deduplication process is recounted.

[0025] Optionally, the poultry counting method further comprises:

[0026] Repeat the above steps until all counting tasks for the day are completed.

[0027] A poultry counting system comprising:

[0028] a video data acquisition device for acquiring video data of the loading and unloading area;

[0029] An edge computing device is used to perform frame extraction processing on the video data to obtain a static image; analyze each frame of the static image based on an artificial intelligence target detection algorithm to identify the target to be detected in the static image, wherein the target to be detected includes poultry; mark the recognition result on the static image; perform deduplication processing on the static images of different frames after marking based on a deduplication target detection algorithm to remove repeatedly identified targets to be detected; and count the poultry after deduplication processing to obtain the number of poultry loaded and unloaded.

[0030] Optionally, in the loading and unloading area, the poultry unloaded by the transport vehicle are mounted on hooks; the target to be inspected also includes empty hooks; the edge computing device is also used to count the empty hooks after the deduplication processing; if the continuous count of the empty hooks exceeds a preset threshold, the poultry after the deduplication processing is recounted.

[0031] This application solves the problem of how to accurately detect, locate and allocate poultry through image recognition technology, so that the detection is not interfered with by the high humidity and high dust environment in the slaughterhouse area, and can ensure the accurate counting of the number of poultry, providing a highly efficient and reliable management tool for the slaughterhouse.

[0032] In order to make the above features and advantages of the present application more obvious and easy to understand, embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a flow chart of a poultry counting method provided in one embodiment of the present application.

[0034] Figure 2 This is a schematic diagram of a poultry counting system provided in another embodiment of the present application.

[0035] In the drawings, like reference numerals refer to the same drawing elements. DETAILED DESCRIPTION

[0036] To make the purpose and technical solutions of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0037] Accurately tracking the number of poultry slaughtered each day is crucial for daily slaughterhouse operations. One method relies on installing infrared counting devices (e.g., infrared sensors) on hanging chains. These devices detect when poultry blocks the sensor's beam. However, the loading and unloading areas of slaughterhouses often experience harsh environmental conditions. High humidity and dust levels often prevent infrared counting devices from functioning properly, significantly reducing their lifespan. This results in extremely high errors and even failure, creating significant challenges for factory management.

[0038] Furthermore, the number of poultry loaded on each truck is crucial data for slaughterhouses, as it directly impacts the settlement of fees between slaughterhouses and farmers. However, infrared sensor counting equipment cannot distinguish between poultry loaded on different vehicles and can only provide a total slaughter count for the day, failing to meet the need for individual vehicle counts. This technical challenge is a pressing issue for slaughterhouses.

[0039] In one embodiment, see Figure 1 The present application provides a poultry counting method, which includes the following steps S10 to S60.

[0040] S10: Acquire video data of the loading and unloading area.

[0041] S20: performing frame extraction processing on the video data to obtain a static image.

[0042] S30: Analyze each frame of the static image based on an artificial intelligence target detection algorithm to identify the target to be detected in the static image, where the target to be detected includes poultry.

[0043] S40: Marking the recognition result on the static image.

[0044] S50: performing deduplication processing on the marked static images of different frames based on a deduplication target detection algorithm to remove the repeatedly identified targets to be detected.

[0045] S60: Counting the poultry after the deduplication process to obtain the number of poultry loaded and unloaded.

[0046] The poultry counting method of the present application solves the problem of how to accurately detect, locate and allocate poultry through image recognition technology, so that the detection is not interfered with by the high humidity and high dust environment of the slaughterhouse, and can ensure the accurate counting of the number of poultry, providing a highly efficient and reliable management tool for the slaughterhouse.

[0047] In step S10, refer to Figure 1 In step S10, video data of the loading and unloading area is obtained.

[0048] As an example, in step S10, obtaining video data of the loading and unloading area includes: obtaining video data of the loading and unloading area based on a high-definition camera.

[0049] As an example, high-definition cameras are installed at key locations in the loading and unloading area to capture high-definition video streams of the loading and unloading area in real time, and high-bandwidth, low-latency data transmission of the video stream is processed through a high-speed network to ensure the integrity and real-time nature of the data.

[0050] As an example, the key position includes a suitable position where a high-definition camera can capture the target to be inspected. Capturing the target to be inspected includes making the target to be inspected appear at a suitable position in the video image without distortion, for example, directly in front of or diagonally above the target to be inspected. The target to be inspected includes poultry.

[0051] In step S20, refer to Figure 1 In step S20, the video data is subjected to frame extraction processing to obtain a static image.

[0052] As an example, in step S20, the frame extraction processing of the video data to obtain a static image includes: frame extraction of the video data to convert continuous video data into a series of static images for subsequent analysis of the target to be inspected in the static image.

[0053] In step S30, refer to Figure 1 In step S30, each frame of the static image is analyzed based on an artificial intelligence target detection algorithm to identify the target to be detected in the static image, where the target to be detected includes poultry.

[0054] As an example, in step S30, the artificial intelligence target detection algorithm is used to analyze each frame of the static image to identify the target to be detected in the static image, including the following steps S301 to S302.

[0055] S301: Input each frame of static image into the target detection model.

[0056] S302: The target detection model identifies and locates each frame of static image based on an artificial intelligence target detection algorithm.

[0057] As an example, each static image frame is input into a target detection model. The target detection model uses an artificial intelligence target detection algorithm, such as the YOLO (You Only Look Once) algorithm or the SSD (Single Shot MultiBox Detector) algorithm. Through the principle of deep learning, the target to be detected in the static image is quickly and accurately identified and located, and the category information of the target to be detected and the precise location in the static image are obtained. The target detection algorithm of this embodiment mainly involves the YOLO5 algorithm.

[0058] Furthermore, in order to ensure that the target detection model can run efficiently, the target detection model is optimized by using technologies such as optimizing the model structure, replacing the model activation function, and quantization to achieve optimization of the computational efficiency and resource usage of the target detection model.

[0059] For example, due to the high speed of the one-way chain of the poultry hanging table, the inference speed of the target detection model must keep up with the detection speed. The replacement model activation function includes: the original silu activation function of the target detection model was replaced by relu, which solves the gradient saturation problem of the sigmoid and tanh functions and accelerates the training of the target detection model.

[0060] As an example, the weight results trained by the Pytorch training framework are 32-bit precision by default. In this application, when the target detection model is deployed, the 32-bit precision is quantized to a 16-bit precision target detection model to improve the inference calculation speed of the target detection model.

[0061] As an example, the Yolov5 algorithm used in this embodiment can be divided into depth and width in terms of structural dimensions. At some operator layers, the target detection model sets a larger dimension in the width direction to enhance the extraction of data features. Since the number of categories detected by this application is small, the target detection model of this application does not need to set a larger width dimension to extract better features. At the same time, reducing the width dimension greatly reduces the number of model parameters and increases the calculation speed of the target detection model.

[0062] As an example, after model optimization, dynamic adjustments are also included. The dynamic adjustments include: dynamically adjusting model parameters according to different environmental conditions and poultry morphology, improving the robustness and adaptability of detection, and improving the accuracy and real-time performance of detection results. Because high-definition cameras are installed in different positions at different poultry hanging tables in slaughterhouses, the images of poultry are different. This application performs some data enhancement on the captured video stream, including: randomly zooming in and out of the targets to be inspected in the loading and unloading area, randomly converting the color space, randomly converting the brightness of the image, randomly setting image noise points, histogram equalization, and other image enhancement methods.

[0063] Specifically, for different application scenarios, the sizes of the detected targets are inconsistent. In order to avoid repeated counting of the same target to be inspected and missing counts of different targets to be inspected, a counting line is set in the loading and unloading area. The targets to be inspected are filtered within a certain width range on the left or right or above or below the counting line. The certain range here means that no other targets to be inspected can appear at the same time in the counting area. Due to the different distances where the high-definition cameras are installed, the widths of the shooting areas of the high-definition cameras in different scenarios are different. If the shooting area is too small, it cannot cope with the lack of counting in the case of video frame skipping. If the shooting area is too large, it will easily lead to the lack of counts of multiple targets to be inspected. Therefore, it is necessary to adaptively adjust the counting area according to the width value of the detection frame of different application scenarios, and then design a filtering algorithm to process the noise data of the targets to be inspected in the shooting area.

[0064] In step S40, refer to Figure 1 In step S40, the recognition result is marked on the static image.

[0065] As an example, in step S40 , marking the recognition result on the static image includes: drawing the recognition result back on the static image in the form of a graphic or a mark.

[0066] In step S50, refer to Figure 1 In step S50, deduplication processing is performed on the marked static images of different frames based on a deduplication target detection algorithm to remove the repeatedly identified targets to be detected.

[0067] As an example, in step S50, the deduplication target detection algorithm is used to perform deduplication processing on the static images of different frames after marking to remove the repeatedly identified targets to be detected, including: using a Kalman filter and non-maximum suppression to perform deduplication processing on the static images of different frames after marking to remove the repeatedly identified targets to be detected, effectively eliminating the problem of repeated counting caused by target movement or environmental factors, and achieving accurate counting of the targets to be detected.

[0068] In step S60, refer to Figure 1 In step S60, the poultry after deduplication processing is counted to obtain the number of poultry loaded and unloaded.

[0069] For example, in the loading and unloading area, poultry unloaded from transport vehicles are mounted on hooks. The objects to be inspected also include empty hooks, meaning that the objects to be inspected include poultry and empty hooks without poultry mounted on them. When the objects to be inspected include poultry and empty hooks, the objects to be inspected identified in step S30 also include empty hooks, and the deduplication process in step S40 also includes removing duplicated empty hooks.

[0070] As an example, after the deduplication process, the following steps may also be performed:

[0071] Count the empty hooks after deduplication;

[0072] If the continuous count of the empty hooks exceeds a preset threshold, the poultry after the deduplication process is recounted.

[0073] Specifically, the data on empty hooks between vehicles on the transport chain comes from the number of empty hooks encountered during the processing by slaughterhouse staff. A slaughterhouse typically has more than one poultry hanging station, each of which can only unload poultry from one truck at a time. The poultry driver transports the poultry to one of the hanging stations, where slaughterhouse staff hang the poultry from the truck onto the transport chain. For example, this process of hanging a truckload of ducks takes approximately 26 to 45 minutes. After the current truckload of poultry is finished hanging, the truck waiting outside the hanging station arrives to begin unloading. The slaughterhouse needs to know how many birds are currently hanging on the current truckload. After the current truckload leaves the poultry hanging station, the next truckload, loaded with poultry, begins unloading. Each truckload of poultry is hung on the one-way transport chain and transported to the slaughterhouse for slaughter. For poultry unloaded between different vehicles, after the poultry on the current vehicle is hung, the slaughterhouse staff will leave several empty hooks on the hanging transport chain to distinguish the poultry from different vehicles, and then proceed to hang the poultry for the next vehicle.

[0074] As an example, the principle of distribution analysis is applied to analyze the continuous empty hook data of the target to be inspected based on the empty hooks between different vehicles in the transport chain.

[0075] As an example, when the number of empty hooks detected exceeds the set threshold, it means that the counting of the poultry on the previous car has ended, and the subsequent counting will be for the poultry on the next car, triggering the end signal of the counting of the poultry on the previous car and the start signal of the counting of the poultry on the next car, that is, the poultry can be re-counted.

[0076] As an example, the count results for the previous load of poultry are automatically saved.

[0077] As another example, after step S60, steps S10 to S60 may be looped, i.e., repeated several times until all counting tasks for the day are completed. This loop ensures that the system can continuously and efficiently process the number of poultry loaded on each truck until the end of the slaughterhouse's day.

[0078] This poultry counting method solves the problem of how to accurately detect, locate, and allocate poultry through image recognition technology. It eliminates interference from the high humidity and dusty environment of the slaughterhouse, allowing accurate counting of the number of poultry loaded on each truck. This provides a more precise and reliable management tool for slaughterhouses.

[0079] In another embodiment, see Figure 2 The present application provides a poultry counting device for a slaughterhouse based on a deep neural network, and the poultry counting device for a slaughterhouse based on a deep neural network comprises:

[0080] Video data acquisition device 2, used to acquire video data of the loading and unloading area 3;

[0081] The edge computing device 1 is used to extract frames from the video data to obtain a static image; analyze each frame of the static image based on an artificial intelligence target detection algorithm to identify the target to be detected in the static image, wherein the target to be detected includes poultry; mark the recognition result on the static image; perform deduplication processing on the static images of different frames after marking based on a deduplication target detection algorithm to remove repeatedly identified targets to be detected; and count the poultry after deduplication processing to obtain the number of poultry loaded and unloaded.

[0082] As an example, the edge computing device 1 is connected to the video data acquisition device 2 via a line.

[0083] As an example, the video data acquisition device 2 includes a high-definition camera. The video data acquisition device 2 can be located in front of or obliquely above the target to be inspected, so that the target to be inspected appears at a suitable position in the video screen without distortion.

[0084] As an example, in the loading and unloading area 3, the poultry unloaded from the transport vehicle are hung on the hooks; wherein, the objects to be inspected also include empty hooks.

[0085] As an example, the edge computing device 1 is also used to: count the empty hooks after deduplication processing; if the continuous count of empty hooks exceeds a preset threshold, the poultry after deduplication processing is recounted. At this time, the number of poultry detected represents the full load number of poultry on the previous car, triggering the end signal of the poultry counting on the previous car, and at the same time starting the start signal of the poultry counting on the next car.

[0086] As an example, the video data acquisition device 2 shoots the loading and unloading area 3, transmits the video data to the edge computing device 1 through the line, extracts frames from the video data, and converts the continuous video stream into a series of static images; based on the artificial intelligence target detection algorithm, each frame of the static image is analyzed to identify the poultry in the static image, so as to achieve accurate detection and positioning of the poultry; the recognition result is then marked on the static image in the form of a graphic or mark; for subsequent analysis and verification, the static images of different frames after marking are deduplicated based on the deduplication target detection algorithm, and the Kalman filter and non-maximum suppression are used to predict the motion trajectory of the target to be inspected and remove duplicate detection data, thereby eliminating the same poultry targets in different static images, effectively eliminating the problem of repeated counting caused by target movement or environmental factors, and counting the poultry after deduplication to obtain the number of poultry loaded and unloaded. Finally, after deduplication, in order to accurately count the number of poultry loaded on each truck, the principle of distribution analysis is applied to analyze the continuous empty hook data based on the remaining empty hook data between different vehicles in the transportation chain. When the continuous increase in the number of empty hook data exceeds the preset threshold, the system will trigger the counting task of the next vehicle, thereby accurately allocating the number of poultry loaded on each truck.

[0087] As an example, the poultry counting device for a slaughterhouse based on a deep neural network executes the above steps in a loop until the end of work for the day, at which point the poultry counting device for a slaughterhouse based on a deep neural network stops working.

[0088] The poultry counting method and system described above solve the problem of how to accurately detect, locate, and allocate poultry through image recognition technology. This ensures that detection is not affected by the high humidity and dusty environment of the slaughterhouse, and accurately counts the number of poultry loaded on each truck. After testing, the device has achieved an accuracy rate of over 99.5%, providing slaughterhouses with an efficient and reliable management tool.

[0089] Although the present invention has been disclosed above with reference to the embodiments, they are not intended to limit the present invention. Anyone with ordinary skill in the art may make slight changes and modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be determined by the scope of the appended patent applications.

Claims

1. A poultry counting method, characterized in that: include: Obtain video data of the loading and unloading area; Performing frame extraction processing on the video data to obtain a static image; Analyzing each frame of the static image based on an artificial intelligence target detection algorithm to identify a target to be detected in the static image, wherein the target to be detected includes poultry; Marking the recognition result on the static image; Deduplication is performed on the static images of different frames after identification based on the deduplication target detection algorithm to remove the repeated identified targets to be detected; The poultry were counted after weight removal to obtain the number of poultry loaded and unloaded.

2. A poultry counting method as claimed in claim 1, characterized in that: Obtain video data of the loading and unloading area based on high-definition cameras.

3. A poultry counting method as claimed in claim 1, characterized in that: The step of performing frame extraction processing on the video data to obtain a static image includes: Frame extraction is performed on the video data to convert continuous video data into a series of static images.

4. A poultry counting method as claimed in claim 1, characterized in that: The artificial intelligence target detection algorithm is used to analyze each frame of the static image to identify the target to be detected in the static image, including: Inputting each frame of the static image into the target detection model; The target detection model identifies and locates each frame of the static image based on the artificial intelligence target detection algorithm.

5. A poultry counting method as claimed in claim 1, characterized in that: The step of marking the recognition result on the static image includes: The recognition result is drawn back onto the static image in the form of a graphic or a mark.

6. A poultry counting method as claimed in claim 1, characterized in that: The deduplication target detection algorithm is based on which the static images of different frames after marking are deduplicated to remove the repeatedly identified targets to be detected, including: using a Kalman filter and non-maximum suppression to deduplication the static images of different frames after marking to remove the repeatedly identified targets to be detected.

7. A poultry counting method as claimed in claim 1, characterized in that: In the loading and unloading area, the poultry unloaded by the transport vehicle is hung on the hook; the target to be inspected also includes an empty hook; after the deduplication process, the following steps are also included: Count the empty hooks after deduplication; If the continuous count of the empty hooks exceeds a preset threshold, the poultry after the deduplication process is recounted.

8. A poultry counting method according to any one of claims 1 to 7, characterized in that: Also includes: Repeat the above steps until all counting tasks for the day are completed.

9. A poultry counting system, characterized in that: include: a video data acquisition device for acquiring video data of the loading and unloading area; An edge computing device, configured to perform frame extraction processing on the video data to obtain a static image; Based on an artificial intelligence target detection algorithm, each frame of the static image is analyzed to identify the target to be inspected in the static image, wherein the target to be inspected includes poultry; the recognition result is marked on the static image; based on a deduplication target detection algorithm, the static images of different frames after marking are deduplicated to remove the repeatedly identified targets to be inspected; the poultry after deduplication are counted to obtain the number of poultry loaded and unloaded.

10. A poultry counting system as claimed in claim 9, characterized in that: In the loading and unloading area, the poultry unloaded by the transport vehicle is hung on the hooks; the targets to be inspected also include empty hooks; the edge computing device is also used to count the empty hooks after the deduplication processing; if the continuous count of the empty hooks exceeds the preset threshold, the poultry after the deduplication processing is recounted.

Images