Real-time livestock barn floor analysis system

Abstract

A real-time livestock barn floor analysis system for livestock barn floor management is disclosed, comprising: a shooting unit that generates an image containing status information of bedding distributed on the floor of a livestock barn; and an analysis unit that analyzes the status information of the bedding by area of ​​the livestock barn and generates distribution information regarding the impregnation status by area.

Description

Technology Field

[0001] The present invention relates to livestock barn floor management technology, and more specifically, to a smart livestock barn management system that utilizes AI-based computer vision technology and IoT sensors to analyze the condition of bedding distributed on the barn floor in real time and determines an optimized stirring pattern through the analysis of livestock behavior patterns. Additionally, the present invention may relate to a technology field that performs consistent floor condition diagnosis regardless of changes in illumination and time by fusing metadata of image data with environmental sensor data, and may also be related to autonomous driving-based stirring robot control technology and cloud-based remote livestock barn management systems. Background Technology

[0002] In the modern livestock industry, hygiene management within livestock barns is essential for improving livestock health and productivity, and in particular, the condition of the bedding laid on the barn floor is recognized as a critical factor directly linked to the prevention of livestock diseases. Since bedding impregnated with livestock manure causes the generation of ammonia gas and bacterial proliferation, maintaining the overall dryness of the bedding through effective mixing with dry bedding can be a key task in barn management.

[0003] To address these problems, various technical approaches have been attempted in the past. Existing livestock barn floor management technologies have primarily focused on hardware-centric physical approaches, such as structural improvements to bedding materials, the development of physical collection devices, the installation of mechanical agitators, modifications to flooring structures, and manure removal using scrapers. However, these conventional technologies reveal fundamental limitations in solving the complex problems that occur in actual livestock barn environments.

[0004] First, conventional mechanical agitation methods treat the entire livestock barn uniformly, failing to account for the varying impregnation states of bedding in different areas. For instance, even when livestock defecate intensively in specific zones, agitating the entire barn in the same manner can lead to energy waste and unnecessary damage to the bedding. Second, conventional physical devices suffer from a critical flaw: they fail to adapt to environmental changes within the barn. While various environmental factors—such as seasonal light intensity, humidity fluctuations due to weather, and temperature variations over time—directly affect the drying speed and impregnation patterns of the bedding, existing technologies remain static treatment methods that completely disregard these dynamic environmental changes. Third, existing technologies have failed to identify the correlation between livestock behavior patterns and changes in floor conditions, making it impossible to pinpoint the root causes of problems and respond proactively. For instance, it is often impossible to distinguish whether persistent moisture in a specific area is due to livestock defecation habits or structural issues like leakage or poor ventilation, frequently resulting in only temporary, stopgap measures.

[0005] Due to these limitations, livestock barn managers are still forced to rely on visual observation and experience to assess bedding conditions and determine the timing and patterns of mixing, which leads to increased management costs and instability in barn hygiene. Particularly in large-scale barns, accurately assessing the overall floor condition is physically challenging, potentially resulting in a failure to detect hygiene deterioration in specific areas at an early stage.

[0006] Therefore, there is an urgent need for the development of new technologies that provide an intelligent monitoring system capable of accurately and objectively analyzing the actual condition of the barn floor, dynamic management techniques capable of adaptively responding to environmental changes, and an integrated approach that even considers the behavioral patterns of livestock. Prior art literature

[0007] Registered Patent 10-1512243 (April 22, 2015) The problem to be solved

[0008] This invention aims to solve the limitations of conventional technology in livestock barn floor management.

[0009] Conventional livestock barn floor management methods rely primarily on mechanical devices and are limited to simple physical treatments, making it difficult to accurately analyze floor conditions due to various environmental variables. In particular, there was a limitation in that consistent analysis was impossible because visual perception results varied even for the same floor condition due to environmental factors such as changes in lighting, seasonal factors, weather changes, and the time of day. Furthermore, the presence of cattle and other objects within the barn caused confusion in object recognition and misjudgment of the condition, leading to a decrease in the accuracy of floor condition diagnosis.

[0010] Furthermore, conventional technology has been limited to a static approach that analyzes only the current state, making it impossible to diagnose complex causes by considering the correlation between livestock defecation patterns and changes in floor conditions. Consequently, it has been difficult to predict and respond to issues such as leakage, undisturbed areas, and structural problems in advance, thereby hindering the efficiency of livestock management and the improvement of the hygienic environment for livestock.

[0011] Therefore, there is a need for a smart livestock barn floor analysis system that enables consistent floor condition analysis regardless of environmental variables and allows for accurate diagnosis and predictive management by utilizing AI-based computer vision technology.

[0012] The problems that this disclosure aims to solve are not limited to those mentioned above, and other unmentioned problems will be clearly understood by a person skilled in the art from the description below. means of solving the problem

[0013] A real-time livestock barn floor analysis system for managing a livestock barn floor according to one aspect of the present invention for solving the above problem comprises: a shooting unit that generates an image containing state information of bedding distributed on the floor of a livestock barn; and an analysis unit that analyzes the state information of the bedding by area of ​​the livestock barn and generates distribution information regarding the impregnation state by area.

[0014] Meanwhile, the above image includes metadata regarding illumination or the date and time of shooting, and the analysis unit can analyze the state information of the mat by considering the metadata.

[0015] In addition, it further includes a collection unit for collecting internal environment data of the livestock barn; and the analysis unit can analyze the state information of the bedding by considering the internal environment data.

[0016] Additionally, it may further include a display unit that displays a standardized image that has been corrected and converted to a color value at a reference point in time considering the date or time of the image.

[0017] Additionally, it may further include a conversion unit that converts an input image into the standardized image using an artificial intelligence model based on multiple training images captured at different times.

[0018] Additionally, the system further includes a recognition unit that analyzes the behavioral patterns of livestock within the barn to identify the defecation posture of the livestock and analyzes the distribution of the defecation preference area based on the position of the livestock at the time of defecation; and the analysis unit may consider the distribution of the defecation preference area when determining the impregnation state for each area.

[0019] Additionally, it may further include a comparison unit that determines an abnormal section by comparing the distribution information on the region-specific impregnation state generated by the analysis unit with the distribution of the defecation preference area to determine a non-overlapping area.

[0020] In addition, it may further include a processing unit that determines the bedding stirring pattern by considering the above-mentioned abnormal section.

[0021] Additionally, it may further include a stirring unit that moves along a predetermined path and stirs the bedding based on the above bedding stirring pattern.

[0022] In addition, the analysis unit may generate the distribution information by considering the shading information of the floor created as a small object within the image or the unidentified area of ​​the floor due to the small object. Furthermore, a computer program stored on a computer-readable recording medium for executing the present disclosure may be further provided.

[0023] In addition, a computer-readable recording medium for recording a computer program for executing a method for implementing the present disclosure may be further provided. Effects of the invention

[0024] According to the present invention, by accurately diagnosing the condition of the livestock barn floor in real time through AI-based computer vision technology and multi-sensor fusion analysis, significant technical effects can be provided compared to conventional simple mechanical device methods. In particular, through image standardization technology that provides consistent analysis results regardless of environmental variables such as illumination, time, and season, the technical challenge of analysis errors caused by environmental variables, which could not be solved by existing technologies, can be resolved.

[0025] Furthermore, the correlation comparison function between livestock behavior pattern analysis and floor condition analysis of the present invention enables the diagnosis of complex causes beyond simple analysis of the current state, thereby allowing for the prediction of issues such as leakage, undisturbed areas, and structural problems in advance, and providing preventive management effects through preemptive responses. This offers advantageous effects, such as reduced labor costs for livestock barn management and the continuous improvement of the livestock hygiene environment.

[0026] Furthermore, the AI-based stirring pattern optimization algorithm of the present invention enables the determination of an efficient work path through real-time data analysis, thereby achieving unexpected synergistic effects that significantly improve work efficiency and optimize energy consumption compared to conventional uniform stirring methods. Through this, it is possible to provide industrial ripple effects that promote the digital transformation of the smart farm ecosystem and contribute to the sustainable development of the livestock industry.

[0027] The effects of the present disclosure are not limited to those mentioned above, and other unmentioned effects will be clearly understood by a person skilled in the art from the description below. Brief explanation of the drawing

[0028] FIG. 1 is a conceptual diagram of a real-time livestock barn floor analysis system according to one embodiment of the present disclosure. FIG. 2 is a conceptual diagram of a real-time livestock barn floor analysis system according to another embodiment of the present disclosure. FIG. 3 is a conceptual diagram of a real-time livestock barn floor analysis system according to another embodiment of the present disclosure. FIG. 4 is a conceptual diagram of a real-time livestock barn floor analysis system according to another embodiment of the present disclosure. FIG. 5 is a conceptual diagram of a real-time livestock barn floor analysis system according to another embodiment of the present disclosure. Specific details for implementing the invention

[0029] Specific structural or functional descriptions regarding embodiments according to the concept of the present invention disclosed herein are provided merely for the purpose of explaining embodiments according to the concept of the present invention, and embodiments according to the concept of the present invention may be implemented in various forms and are not limited to the embodiments described herein.

[0030] The terms used in this specification will be briefly explained, and the present invention will be described in detail. The terms used in this specification have been selected to be as generally used as possible, taking into account their functions in the present invention; however, these terms may vary depending on the intent of those skilled in the art, case law, the emergence of new technologies, etc. Additionally, in specific cases, terms have been arbitrarily selected by the applicant, and in such cases, their meanings will be described in detail in the relevant description of the invention. Therefore, the terms used in the present invention should be defined not merely by their names, but based on their meanings and the overall content of the present invention.

[0031] Throughout this specification, when a part is described as "comprising" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components. Furthermore, terms such as "...part," "...module," etc., as used in this specification refer to a unit that processes at least one function or operation, and this may be implemented in hardware or software, or as a combination of hardware and software.

[0032] The present invention relates to a smart livestock barn floor management system capable of effectively maintaining the hygienic and dry conditions of the barn floor by utilizing AI-based computer vision technology to analyze the condition of the bedding in real time, determining an optimized stirring pattern based on the analysis, and automatically stirring the bedding. Conventional livestock barn management methods relied on simple mechanical devices or manpower, making it difficult to accurately assess the condition of the floor; in particular, analysis errors caused by various environmental variables and inefficient stirring operations have been pointed out as problems. To solve these problems, the present invention provides an intelligent management system that integrates real-time video analysis, environmental data fusion, and livestock behavior pattern analysis.

[0033] Referring to FIG. 1, a real-time livestock barn floor analysis system (100) according to one embodiment of the present invention may basically include a shooting unit (110), an analysis unit (120), and a processing unit (130). The shooting unit (110) can generate an image containing status information of the bedding distributed on the floor of the livestock barn, and the analysis unit (120) can analyze the status information of the bedding by area of ​​the livestock barn and generate distribution information regarding the impregnation status by area. The processing unit (130) can determine a bedding stirring pattern for drying the bedding based on the distribution information. These components can be organically linked to each other to implement an integrated system capable of monitoring the condition of the livestock barn floor in real time and performing optimized management tasks.

[0034] The shooting unit (110) is a core component for collecting high-resolution video data to accurately determine the condition of the bedding on the floor of the livestock barn, and can be designed to provide images of consistent quality even in various shooting environments. In one embodiment of the present invention, the shooting unit (110) may include a plurality of IP cameras installed on the ceiling of the livestock barn, and each camera may capture high-definition video of, for example, 4K resolution or higher, but is not limited thereto, and may be appropriately selected in the range of 2K to 8K depending on the usage environment. The cameras may be placed at intervals of, for example, 3 meters to 5 meters so as to cover the entire area of ​​the livestock barn without omission, but may be adjusted in the range of 2 meters to 8 meters depending on the size and shape of the livestock barn.

[0035] One of the important features of the shooting unit (110) is that it can automatically include metadata in the captured image. As mentioned in claim 2, the image may include metadata regarding illuminance or the time of shooting, and such metadata may enable accurate state analysis considering environmental variables during a subsequent analysis process. Specifically, illuminance information may be measured through an illuminance sensor built into the camera, and may be measured in a range of, for example, 0 lux to 100,000 lux, but is not limited thereto. The time of shooting information may have an accurate timestamp recorded based on a GPS-synchronized system clock, which can be used to analyze the pattern of changes in the floor condition by season and time of day.

[0036] According to some other embodiments, the imaging unit (110) may further include an infrared illumination device for night imaging. The infrared illumination may use near-infrared light with a wavelength of, for example, 850 nm, but may be selected from a range of 750 nm to 950 nm depending on the environment of use. Such infrared illumination can enable continuous 24-hour monitoring without causing stress to livestock. Additionally, the imaging unit (110) may include a housing with waterproof and dustproof functions, and may satisfy, for example, a protection rating of IP65 or higher, but is not limited thereto.

[0037] Video data generated by the shooting unit (110) can be simultaneously streamed in real time and stored. For real-time analysis, the video can be transmitted at a frame rate of, for example, 30fps, but can be adjusted within a range of 15fps to 60fps depending on system performance and network bandwidth. Video compression can use H.264 or H.265 codecs and can be optimized by considering the balance between compression ratio and image quality. The stored video data can be used as subsequent training data or for historical analysis.

[0038] The analysis unit (120) can be implemented as an AI-based computer vision system that precisely analyzes the condition of the bedding on the floor of the livestock barn based on image data collected from the shooting unit (110). The analysis unit (120) can be largely composed of an image preprocessing unit, an object detection unit, a state classification unit, and a distribution generation unit, and each unit can function independently while also interacting with one another to derive a comprehensive analysis result.

[0039] The image preprocessing unit can perform the role of converting the original image received from the shooting unit (110) into a form suitable for analysis. As mentioned in claim 4, the analysis unit (120) can analyze the condition information of the bedding in each area of ​​the livestock barn based on a standardized image in which the image is corrected and converted into a color value at a reference point in time considering the date and time. This standardization process may be essential to obtain consistent analysis results by minimizing the influence of changes in illumination, shadows, color temperature, etc., that may occur in various shooting environments.

[0040] Specifically, the standardization process may include the following steps. First, the illuminance value at the time of shooting and the shooting date and time information can be extracted from the metadata of the original image. Next, correction parameters can be determined by calculating the difference from reference conditions (e.g., 12 noon, illuminance 50,000 lux, clear weather). Subsequently, color values ​​can be standardized by applying techniques such as histogram equalization, gamma correction, and color temperature adjustment. In some embodiments of the present invention, a machine learning-based automatic correction algorithm may be applied to this standardization process, which can automatically determine optimized correction parameters through a large amount of training data.

[0041] Meanwhile, a real-time livestock barn floor analysis system (100) according to one embodiment of the present invention may further include a display unit (not shown) that displays a standardized image for manager monitoring.

[0042] The object detection unit can perform the role of accurately distinguishing bedding, livestock, and other objects in a standardized image. As mentioned in claim 10, the analysis unit (120) can generate the distribution information by considering the shading information of the floor created by the cow object within the image or the unidentified area of ​​the floor caused by the cow object. This may be an important function for solving the problem in a livestock barn environment where livestock or other objects may obscure the floor or create shadows, thereby hindering accurate state analysis.

[0043] For object detection, for example, YOLO (You Only Look Once) v8 or higher versions may be used, but other object detection algorithms such as R-CNN or SSD (Single Shot MultiBox Detector) may be applied depending on the usage environment and performance requirements. The detection accuracy of small objects can aim for, for example, 95% or higher, but can be adjusted within the range of 90% to 99% depending on system performance and real-time processing requirements. Bounding boxes and segmentation masks can be generated for the detected small objects, which allow the corresponding areas to be excluded from analysis or for separate processing logic to be applied.

[0044] The state classification unit is a key component for precisely classifying the impregnation state of the bedding, and can utilize a deep learning-based segmentation model. In one embodiment of the present invention, a segmentation model based on the U-Net architecture may be used, but other segmentation models such as Mask R-CNN, DeepLab, and PSPNet may be applied depending on the usage environment. The state of the bedding may be classified into three stages, for example, dry, wet, and saturated, but may be extended to five or seven stages if more detailed analysis is required.

[0045] The classification criteria for each state can be defined as follows. The dry state is a state in which the bedding has absorbed almost no moisture, and the RGB color values ​​may be in the range, for example, R: 150-200, G: 120-170, B: 80-130, but are not limited thereto and may be adjusted according to the type of bedding and lighting conditions. The wet state is a state in which the bedding has partially absorbed moisture, and the RGB color values ​​may be in the range, for example, R: 100-150, G: 80-120, B: 60-100. The saturated state is a state in which the bedding has absorbed maximum moisture and no longer has the ability to absorb moisture, and the RGB color values ​​may be in the range, for example, R: 70-120, G: 60-100, B: 40-80.

[0046] The distribution generation unit can generate impregnation state distribution information for the entire area of ​​the livestock barn based on the state classification results. The livestock barn floor can be divided into a grid of, for example, 1 meter × 1 meter units, but can be adjusted within a range of 0.5 meters × 0.5 meters to 2 meters × 2 meters depending on the size of the livestock barn and analysis precision requirements. For each grid area, the ratio of dry, wet, and saturated states can be calculated and expressed as a percentage. For example, a specific grid area may have a distribution of 60% dry, 30% wet, and 10% saturated.

[0047] Distribution information can be visualized in the form of a two-dimensional heatmap and represented by different colors or densities for each state. For example, dry conditions can be indicated in shades of green, wet conditions in shades of yellow, and saturated conditions in shades of red; however, other color schemes may be applied depending on user preference or visual effects. Additionally, the distribution information can be stored as time-series data and utilized to analyze patterns of floor condition change over time.

[0048] Referring to FIG. 2, a real-time livestock barn floor analysis system (100) according to another embodiment of the present invention may further include a collection unit (140) for collecting internal environment data of the livestock barn. As mentioned in claim 3, the analysis unit (120) may analyze the condition information of the bedding by taking into account the internal environment data. This is intended to derive more accurate and reliable analysis results by comprehensively considering environmental factors that are difficult to identify through simple image analysis alone.

[0049] The collection unit (140) may include a sensor network composed of various IoT sensors, and each sensor may monitor specific environmental parameters in real time. The temperature sensor may measure the temperature inside the livestock barn and may have a precision of, for example, ±0.1℃, but is not limited thereto, and may be selected within a range of ±0.05℃ to ±0.5℃ depending on the usage environment. The measurement range may be, for example, -20℃ to 60℃, but may be adjusted considering the regional characteristics of the livestock barn and seasonal changes.

[0050] The humidity sensor can measure the relative humidity inside the livestock barn and can have an accuracy of, for example, ±1% RH. Since the humidity inside the barn is directly related to the drying speed of the bedding, accurate humidity measurement can significantly improve the accuracy of the analysis of the bedding condition. The measurement range can be, for example, 0% to 100% RH, and the analysis algorithm can reflect the fact that the drying of the bedding can be significantly delayed, particularly in high humidity environments of 80% or higher.

[0051] Ammonia concentration sensors are an important component for monitoring air quality inside livestock barns, capable of measuring the amount of ammonia generated from livestock excrement in real time. The measurement range may be, for example, 0 ppm to 100 ppm, but can be adjusted according to the barn's ventilation system and livestock density. Since areas with high ammonia concentrations are likely to be areas where excrement is concentrated, this information can serve as an important reference for analyzing bedding conditions and determining stirring patterns.

[0052] The placement of sensors can be optimized by considering the size and shape of the livestock barn. For example, sensors can be arranged in a grid pattern at intervals of 5 meters, but can be adjusted within a range of 3 to 10 meters depending on the characteristics of the livestock barn. The installation height of the sensors can be, for example, at a point 2 meters from the ground, but can be selected within a range of 1.5 to 3 meters considering the size of the livestock and the structure of the livestock barn.

[0053] The collection unit (140) may also include a function for collecting external weather information. This can collect real-time weather data through APIs from the Korea Meteorological Administration or other weather service providers, and may include information such as temperature, humidity, sunlight, solar radiation, precipitation, wind speed, and wind direction. Since external weather conditions can have a direct or indirect effect on the internal environment of the livestock barn, more accurate analysis results can be obtained by comprehensively considering this information.

[0054] The data collection cycle may be, for example, at 5-minute intervals, but can be adjusted within a range of 1 to 30 minutes depending on the type of sensor and the rate of change. In cases where rapid environmental changes are expected, the collection cycle can be shortened to perform more detailed monitoring. The collected data can be transmitted to a central processing system via LoRaWAN, Wi-Fi, or a wired network, and encryption protocols may be applied to ensure data integrity and security.

[0055] Referring to FIG. 3, another embodiment of the present invention may further include a conversion unit (150) that converts an input image into a standardized image using an artificial intelligence model based on a plurality of training images taken at different shooting times. This is a configuration mentioned in claim 5 and is a core technology for obtaining consistent analysis results by correcting differences in image quality caused by changes in illumination, seasonal changes, weather changes, etc., which may occur in various shooting environments.

[0056] The conversion unit (150) can be implemented based on a generative AI model, and in particular, a Generative Adversarial Network (GAN) architecture can be effectively utilized. In one embodiment of the present invention, a GAN model dedicated to image conversion, such as CycleGAN or Pix2Pix, may be used, but other generative models such as StyleGAN or BigGAN may be applied depending on the usage environment and performance requirements. These models can learn how to convert images under various shooting conditions into images under reference conditions through a large amount of training data.

[0057] Training data may consist of multiple pairs of images captured at different times. For example, it may include images of the same barn floor area taken in the morning, at noon, in the evening, and at night, as well as images taken on clear, cloudy, and rainy days. Each image may be stored with metadata at the time of capture (illumination, date and time of capture, weather conditions, etc.) to be used as a criterion for conditional transformation during the training process. The scale of the training data may be, for example, 10,000 or more images, but may be adjusted within the range of 1,000 to 100,000 images depending on the complexity and performance requirements of the model.

[0058] The transformation process can consist of the following steps. First, current shooting conditions (illumination, time, weather conditions, etc.) can be extracted from the metadata of the input image. Next, transformation parameters can be determined by calculating the difference from reference conditions (e.g., 12 noon, illumination 50,000 lux, clear weather). Subsequently, the input image can be transformed into a standardized image corresponding to the reference conditions using a trained generative model.

[0059] Various metrics can be used to evaluate the quality of the transformation. The Structural Similarity Index (SSIM) is a metric that measures structural similarity; for example, a value of 0.85 or higher can be targeted, but it can be adjusted within the range of 0.8 to 0.95 depending on the usage environment. The Peak Signal-to-Noise Ratio (PSNR) is a metric that measures the signal-to-noise ratio; for example, a value of 25 dB or higher can be targeted. Additionally, the quality of the transformation can be evaluated by directly measuring the accuracy of litter condition classification in the transformed image.

[0060] The conversion unit (150) may have a hardware configuration optimized for real-time processing. For example, the conversion speed can be improved through parallel processing using a GPU (Graphics Processing Unit), and high-performance GPUs such as NVIDIA Tesla V100, A100, etc., may be used, but are not limited thereto. The conversion processing time may be, for example, within 1 second per image, but may be adjusted according to image resolution and model complexity.

[0061] According to some other embodiments, the conversion unit (150) may include a multi-scale conversion function. This can perform optimized conversions for images of different resolutions, providing fast processing for low-resolution images and precise conversion for high-resolution images. Additionally, the conversion unit (150) may include an adaptive learning function so that the model can be continuously updated in response to new shooting environments or seasonal changes.

[0062] Referring to FIG. 4, another embodiment of the present invention may further include a recognition unit (160) that analyzes the behavioral patterns of livestock in a barn to identify the defecation posture of the livestock and analyzes the distribution of a defecation preference area based on the position of the livestock at the time of defecation. This is a configuration mentioned in claim 6 and is a key component for implementing a complex diagnostic system that comprehensively considers the behavioral patterns of livestock beyond simple floor condition analysis.

[0063] The recognition unit (160) can be implemented as an intelligent system that combines computer vision technology and behavior analysis algorithms. For livestock object detection, a real-time object detection model based on, for example, YOLO v8 may be used, but other detection algorithms such as Faster R-CNN and SSD may be applied depending on the usage environment. The detection accuracy of cattle can be aimed at, for example, 95% or higher, which can have a direct impact on the accuracy of subsequent behavior analysis.

[0064] Time-series data analysis may be essential for analyzing behavioral patterns. In one embodiment of the present invention, a CNN-LSTM hybrid structure may be used, which can simultaneously perform spatial feature extraction and temporal pattern analysis. The CNN part can analyze the posture and shape of livestock in each frame, and the LSTM part can learn the posture change pattern over time. The analysis window may be set, for example, within a range of 30 seconds to 2 minutes, because the defecation behavior of livestock is generally completed within this time range.

[0065] The characteristics of the defecation posture can be defined as follows. The angle of back arching may be in the range of, for example, 15 to 25 degrees, but can be adjusted within the range of 10 to 30 degrees depending on individual differences and breeds of livestock. The angle of tail elevation may be in the range of, for example, 30 to 45 degrees, and this is a posture that is characteristically observed during defecation. The degree of leg spread, head position, and overall body balance can also be utilized as important characteristics for recognizing the defecation posture.

[0066] The timing of defecation can be detected based on the duration of posture maintenance. Generally, cows can maintain a defecation posture for, for example, 30 to 60 seconds, but this may vary within a range of 15 to 90 seconds depending on individual differences. If the posture maintenance duration exceeds a threshold and a characteristic posture pattern is detected simultaneously, it can be determined as defecation behavior. Additionally, behavioral patterns after defecation (e.g., movement, change of posture, etc.) can also be utilized to improve the accuracy of the determination.

[0067] The distribution analysis of defecation preference zones can be performed by accumulating defecation location data over time. The barn floor can be divided into a grid of, for example, 1 meter × 1 meter units, and the number of defecations occurring in each grid can be recorded. The analysis period may be, for example, one week to one month, but can be adjusted to account for seasonal changes or changes in the rearing environment. Areas with high defecation frequency can be classified as "defecation preference zones," which can be visualized in the form of a heatmap.

[0068] A Gaussian kernel can be applied to generate a heatmap, and an influence range of, for example, 2 meters can be set centered on each defecation point. This takes into account that the impact of defecation is not limited to the exact point but can extend to surrounding areas. The size and shape of the kernel can be adjusted according to the size of the barn and the movement patterns of the livestock.

[0069] The recognition unit (160) may also include a function for identifying individual livestock. This can distinguish individual livestock by learning the unique physical characteristics of each livestock (e.g., fur pattern, body shape, size, etc.) and can enable the analysis of individual defecation patterns. The accuracy of individual identification can be aimed at, for example, 90% or more, but can be adjusted according to the number and similarity of the livestock.

[0070] The analysis unit (120) can consider the distribution of the defecation preference area when determining the impregnation state by area.

[0071] As mentioned in claim 7, another embodiment of the present invention may further include a comparison unit (170) that determines an abnormal section by comparing the distribution information of the region-specific impregnation state generated by the analysis unit with the distribution of the defecation preference area to determine a non-overlapping area. This is intended to implement an advanced diagnostic system that can detect abnormal situations early and identify the cause through the correlation between the two analysis results, going beyond simple floor condition analysis or defecation pattern analysis.

[0072] The comparison unit (170) can be implemented as an intelligent comparison system that combines a statistical analysis algorithm and a pattern matching technology. First, a process of normalizing the floor condition heatmap generated by the analysis unit (120) and the defecation preference area heatmap generated by the recognition unit (160) to the same coordinate system and resolution can be performed. This may be an essential preprocessing step for accurate comparison between the two heatmaps.

[0073] Heatmap overlay analysis can be performed at the pixel or grid level. A correlation coefficient between the impregnation level of the floor state and the frequency of defecation at each location can be calculated, using the Pearson correlation coefficient or the Spearman rank correlation coefficient. Under normal circumstances, it can be expected that the impregnation level of the floor state will be high in areas with frequent defecation, so the correlation coefficient may show a high positive correlation, for example, of 0.7 or higher.

[0074] The criteria for determining abnormal sections can be established as follows. Areas with a correlation coefficient of, for example, less than 0.3 may be classified as abnormal sections, but may be adjusted within the range of 0.2 to 0.5 depending on the characteristics and management standards of the livestock barn. Abnormal sections may be further classified in detail; areas with low feces but high impregnation may be classified as "areas suspected of leakage," while areas with high feces but low impregnation may be classified as "unstirred areas" or "areas with poor drainage."

[0075] The cause analysis algorithm can analyze the characteristics of abnormal sections in greater detail. In the case of areas suspected of leakage, it is possible to verify whether a continuous wet state is maintained by analyzing the pattern of impregnation change over time. While impregnation caused by normal waste should gradually dry out over time, impregnation caused by leakage may persist or worsen. In the case of un-stirred areas, it is possible to determine whether sufficient stirring was performed by comparing with past stirring history.

[0076] Structural problem areas can be defined as cases where abnormal conditions repeatedly occur in the same location. For example, if a specific area consistently shows lower-than-expected impregnation despite being a preferred defecation spot, it may suggest a potential problem with the floor structure or drainage facilities in that area. These analysis results can serve as important information for livestock barn managers, indicating the need for facility improvements or maintenance.

[0077] The comparison unit (170) can also analyze pattern changes due to seasonal changes or changes in the rearing environment. For example, during the winter, livestock may tend to concentrate in specific areas, which may also affect defecation patterns. Through an adaptive analysis algorithm that takes these seasonal changes into account, it may be possible to determine abnormal sections more accurately.

[0078] The processing unit (130) is a key component that determines an optimized bedding stirring pattern for drying the bedding based on distribution information generated by the analysis unit (120), and may include various optimization algorithms and decision logic. As mentioned in claim 8, the processing unit (130) can determine the bedding stirring pattern by considering the above-mentioned abnormal section, which can implement a customized stirring strategy suitable for the situation beyond simple uniform stirring.

[0079] The processing unit (130) may be largely composed of a priority determination module, a pattern generation module, a path optimization module, and a scheduling module. The priority determination module can determine the priority of the stirring operation based on the impregnation state distribution information provided by the analysis unit (120). An area with a saturation level of, for example, 70% or more may be set as the first priority, but may be adjusted within the range of 60% to 80% according to the management standards of the livestock barn. A wet state area may be set as the second priority, and a dry state area as the third priority, and different stirring intensities and patterns may be applied for each priority.

[0080] If abnormal sections are detected, the corresponding area may be classified as a special management target and assigned the highest priority. In areas suspected of leakage, intensive stirring can facilitate moisture removal, while in unstirred areas, deep stirring can ensure complete mixing of the bedding. For areas with structural issues, the manager may be notified that separate inspection or repair work may be required in addition to stirring.

[0081] The pattern generation module can generate optimal stirring patterns by considering priorities and characteristics of each area. Basic stirring patterns may include zigzag, spiral, and grid patterns, each of which may have different advantages and disadvantages. Zigzag patterns can efficiently cover large areas but may result in insufficient processing of corners. Spiral patterns allow for gradual stirring from the center to the periphery or from the periphery to the center, but the processing time may be relatively long. Grid patterns enable uniform stirring but may require a longer travel distance.

[0082] The optimal pattern can be selected according to the characteristics of each area. A spiral pattern can be applied to high-saturation intensive treatment areas to treat severe contamination starting from the center, while a zigzag pattern can be applied to general wet areas for efficient treatment. A grid pattern can be applied to dry areas to enable uniform maintenance.

[0083] The path optimization module can optimize the movement path of the actual stirring equipment based on the determined stirring pattern. This can be approached as an optimization problem similar to the Traveling Salesman Problem (TSP) or Vehicle Routing Problem (VRP), and metaheuristic algorithms such as genetic algorithms, simulated annealing, and ant colony optimization can be utilized. Optimization goals may include minimizing total travel distance, working time, and energy consumption, and the weights among these can be adjusted according to operational policies.

[0084] Various constraints can be considered during path optimization. By considering the location and movement patterns of livestock, the path can be adjusted to avoid areas with high livestock concentration during specific time periods. Taking into account the battery capacity and charging time of the stirring equipment, a path to a charging station can be included if charging is required during operation. Additionally, a path can be generated to avoid fixed obstacles (posts, water dispensers, feed troughs, etc.) within the livestock barn.

[0085] The scheduling module can optimize the temporal arrangement of stirring tasks. By considering the activity patterns of livestock, stirring can be performed intensively during periods when the animals are resting. Since cattle are generally most active during the early morning and evening hours and relatively quiet during the day, scheduling that takes these patterns into account is possible.

[0086] Environmental conditions can also affect scheduling. On days with high humidity, the stirring effect may be limited, so the stirring intensity or frequency may be increased. Conversely, on dry days, the stirring intensity may be lowered to minimize dust generation. Additionally, since outdoor work may be restricted during rainfall, focus can be placed on indoor stirring tasks.

[0087] The processing unit (130) may also include a predictive management function. By analyzing time-series data, it can learn the pattern of change in the bottom state and predict future states to plan preemptive stirring operations. For example, if a pattern is observed where a specific area periodically reaches a saturated state, stirring operations can be performed in advance before reaching saturated state to prevent problems.

[0088] Referring to FIG. 5, another embodiment of the present invention may include a stirring unit (180) that moves along a predetermined path and stirs the bedding based on the bedding stirring pattern. This is a physical work system that actually executes the stirring pattern determined by the processing unit (130), as described in claim 9.

[0089] The stirring unit (180) can be implemented based on an autonomous driving robot platform and may consist of a moving unit, a stirring device unit, a sensor unit, and a control unit. The moving unit is a drive system that enables stable and precise movement on the floor of a livestock barn, and, for example, a four-wheel drive system or an endless track system may be applied. The four-wheel drive system enables efficient movement on a generally flat floor, but may lack traction in areas where bedding is piled thickly or in wet areas. The endless track system enables stable movement due to a large contact area, but the turning radius may be large.

[0090] The movement speed can be variably adjusted according to the characteristics of the work area. For example, in general movement sections, it can move at a speed of 1.0 m / s, but in intensive processing sections requiring precise stirring, the speed can be reduced to 0.3 m / s. The maximum movement speed can be, for example, 2.0 m / s, but can be adjusted considering safety and stirring quality.

[0091] The stirring device may be composed of a mechanical device capable of effectively mixing and turning over the bedding. In one embodiment of the present invention, a variable depth stirring device based on a linear actuator may be used. The stirring depth can be automatically adjusted according to the bottom condition, for example, within a range of 5 cm to 15 cm. In dry areas, surface preparation can be performed through shallow stirring (5-8 cm), and in saturated areas, complete mixing with the dry bedding underneath can be achieved through deep stirring (12-15 cm).

[0092] Mixing devices can take various forms, and rotary tillers, vibrating agitators, and screw-type agitators may be applied. Rotary tillers are similar to methods used in soil cultivation and can effectively turn over and mix the bedding. Vibrating agitators can promote air circulation by loosening the bedding through vibration. Screw-type agitators can mix the bedding while moving it using spiral blades.

[0093] The sensor unit may be composed of various sensors for safe and accurate operation of the stirring unit (180). A LiDAR sensor can scan the surrounding environment in three dimensions to detect obstacles and generate an avoidance path. The measurement range may be, for example, a radius of 10 meters, but can be adjusted to a range of 5 to 20 meters depending on the size of the barn and the density of obstacles. The measurement precision may be, for example, ±2 cm, but can be selected from a range of ±1 cm to ±5 cm depending on the operational precision requirements.

[0094] Ultrasonic sensors can be used for detecting nearby obstacles and measuring the distance from the floor. The measurement range can be, for example, 0.1 meters to 3 meters, which can be utilized for controlling stirring depth and maintaining a safe distance. Inertial Measurement Unit (IMU) sensors can monitor the robot's posture and orientation in real time to enable accurate path following.

[0095] While GPS sensors can be used to determine absolute position in large livestock barns, their accuracy may be limited in indoor environments; therefore, RTK-GPS or an Indoor Positioning System (IPS) may be used complementarily. Positional precision can be, for example, ±10 cm, but it can be adjusted within a range of ±5 cm to ±20 cm depending on operational precision requirements.

[0096] The control unit is a central processing unit that integrally controls all operations of the stirring unit (180) and can be implemented based on a high-performance embedded computer. For example, a platform with performance greater than or equal to that of NVIDIA Jetson AGX Xavier can be used, but other platforms may be selected depending on processing requirements. The control unit can be based on a real-time operating system (RTOS) to enable precise timing control.

[0097] The path-following algorithm converts the stirring pattern received from the processing unit (130) into an actual movement path and can modify the path in real time based on sensor information. Path-following algorithms such as Pure Pursuit, Stanley Controller, and Model Predictive Control (MPC) can be applied, and parameters can be adjusted according to the characteristics of the livestock barn environment.

[0098] Obstacle avoidance algorithms can detect unexpected obstacles (livestock, people, temporary structures, etc.) in real time and generate a safe avoidance path. Algorithms such as the Dynamic Window Approach (DWA), Rapidly-exploring Random Tree (RRT), and Artificial Potential Field can be used and optimized for real-time processing.

[0099] The power management system can perform efficient energy management for continuous operation of the stirring unit (180). For example, a lithium iron phosphate (LiFePO4) battery may be used, which has excellent safety and lifespan. The battery capacity may be designed to allow for continuous operation for, for example, 8 hours, but may be adjusted within the range of 4 to 12 hours depending on the work intensity and the size of the barn.

[0100] The charging system may include an automatic charging function, and can automatically move to a charging station to perform charging when the battery level drops, for example, below 20%. The charging time may be, for example, around 2 hours, but may be adjusted depending on the battery capacity and charger performance. It may also be designed to enable 50% charging within 30 minutes through a fast charging function.

[0101] The overall operation process of the real-time livestock barn floor analysis system (100) of the present invention may consist of the following steps. First, the shooting unit (110) may capture a high-resolution image of the livestock barn floor in real time and record metadata at the time of shooting. At the same time, the collection unit (140) may collect environmental data inside the livestock barn and external weather information.

[0102] The collected images can be converted into a form that allows for consistent analysis through a standardization process in the conversion unit (150). In this process, variables such as illumination, time, and weather conditions are considered to convert the images into standard conditions. The standardized images are transmitted to the analysis unit (120) so that the condition of the bedding can be analyzed through AI-based computer vision technology.

[0103] The analysis unit (120) can accurately distinguish objects within the image and classify the impregnation state of the bedding as dry, wet, or saturated. In this process, shadows or unidentified areas caused by livestock or other objects can be taken into account to perform accurate analysis. The analysis results can be generated as distribution information of the impregnation state by area and visualized in the form of a heatmap.

[0104] At the same time, the recognition unit (160) can analyze the behavioral patterns of livestock to detect defecation postures and record location information at the time of defecation. As this data accumulates, a distribution of defecation preference areas can be generated. The comparison unit (170) can determine abnormal sections by comparing the floor condition distribution with the defecation preference area distribution.

[0105] The processing unit (130) can determine an optimal stirring pattern by combining the distribution information from the analysis unit (120) and the abnormal section information from the comparison unit (170). In this process, steps such as determining priority, generating a pattern, optimizing the path, and scheduling can be performed sequentially. The determined stirring pattern is transmitted to the stirring unit (180) so that actual stirring operations can be performed.

[0106] The stirring unit (180) can perform the rolling of bedding while moving along a designated path via autonomous driving according to the received stirring pattern. During the operation, it can monitor the surrounding environment in real time through sensors and respond to unexpected obstacles or changes in the situation. When the stirring operation is completed, the result is fed back to the system and can be reflected in the next work plan.

[0107] The real-time livestock barn floor analysis system (100) of the present invention can provide more advanced functions through various extended embodiments. By establishing a cloud-based integrated management system, multiple livestock barns can be managed simultaneously, thereby enabling efficient operation of large-scale livestock farms. The cloud platform may utilize commercial cloud services such as AWS, Azure, and Google Cloud, but a private cloud or hybrid cloud configuration may also be possible considering security and cost.

[0108] By enhancing data analysis and prediction capabilities, advanced analytical functions such as time series analysis, pattern recognition, and anomaly detection can be provided. Through machine learning-based prediction models, changes in bottom conditions can be predicted in advance, and preemptive management tasks can be planned. Prediction accuracy can aim for over 85%, for example, in predicting the state 24 hours in advance, but this may be adjusted depending on data quality and model performance.

[0109] The user interface can consist of a web-based dashboard and a mobile application, and can provide real-time monitoring, remote control, and notification functions. The dashboard utilizes responsive web design to provide an optimized user experience across various devices. Through 3D heatmap visualization, the condition of the barn floor can be intuitively understood, and changes over time can be displayed via animation.

[0110] The notification system can immediately convey important situations to the manager through various channels such as SMS, email, and push notifications. Notification criteria can be set by the user, and for example, notifications may be sent in situations such as the occurrence of areas with a saturation level of 80% or higher, detection of three or more abnormal sections, or failure of stirring equipment.

[0111] This invention can be applied not only to cattle barns but also to various other livestock species. Floor management based on similar principles can be performed in barns for pigs, chickens, sheep, goats, and others, and analysis algorithms and stirring patterns can be adjusted to suit the characteristics of each species. For example, since the defecation patterns of pig barns differ from those of cattle, a behavioral analysis model tailored to this can be applied.

[0112] Furthermore, the technology of the present invention can be applied to fields beyond livestock barn management. It can be utilized for soil management in greenhouses or vinyl greenhouses, management of fermentation status in composting facilities, and monitoring of soil conditions at construction sites. By adjusting the sensor configuration and analysis algorithms to suit the characteristics of each application field, an optimized solution can be provided.

[0113] The real-time livestock barn floor analysis system (100) of the present invention can achieve the following comprehensive technical effects. First, the accuracy of livestock barn floor condition analysis can be significantly improved through AI-based computer vision technology. It can achieve a high accuracy of over 95% compared to existing visual inspection or simple sensor-based methods, which can contribute to the scientific and precise management of livestock barns.

[0114] Second, through environment-adaptive image standardization technology, consistent analysis performance can be maintained even under various shooting conditions. This enables stable system operation regardless of seasonal, weather, or time of day changes. Third, the convergence of livestock behavior pattern analysis and floor condition analysis allows for cause analysis and predictive management, moving beyond simple condition diagnosis.

[0115] Fourth, the efficiency of mixing operations can be improved by more than 40% through an autonomous driving-based precision mixing system. This is a significant effect that enables simultaneous reduction of labor costs and improvement of work quality. Fifth, 24-hour unmanned management is possible through real-time monitoring and remote control functions, which can accelerate the digital transformation of the livestock industry.

[0116] In terms of industrial applicability, the present invention can directly contribute to improving productivity and reducing costs in the livestock industry. It can reduce livestock barn management labor costs by more than 60% and save on material costs by extending the bedding replacement cycle. In addition, it can lower the incidence of disease and improve productivity by creating a healthy rearing environment for livestock.

[0117] From an environmental perspective, it can contribute to the realization of sustainable livestock farming through effects such as reduced ammonia emissions, odor mitigation, and prevention of water pollution. This can play a significant role in promoting ESG management and the expansion of eco-friendly livestock farming. In terms of animal welfare, providing a comfortable rearing environment can reduce stress on livestock and promote healthy growth.

[0118] Technological ripple effects include the widespread application of AI, IoT, and robotics technologies in the agricultural sector. The technology and know-how accumulated through this invention can spread throughout the smart farm ecosystem, accelerating the digital transformation of agriculture. Furthermore, it can contribute to the growth of related industries and job creation.

[0119] In terms of international competitiveness, this invention can be evaluated as a core technology capable of elevating Korea's smart farm technology level to the world's highest standard. This enables the export and transfer of related technologies, thereby contributing to the enhancement of national competitiveness.

[0120] In conclusion, the real-time livestock barn floor analysis system (100) of the present invention is an innovative system capable of realizing complete automation and intelligence in livestock barn floor management by integrating advanced technologies such as AI-based computer vision technology, environmental data fusion analysis, livestock behavior pattern recognition, and autonomous driving robot control. As a sustainable solution capable of simultaneously achieving increased productivity in the livestock industry, environmental protection, and enhanced animal welfare, this can present a new paradigm for the future livestock industry. Explanation of the symbols

[0121] 100: Real-time livestock barn floor analysis system 110: Filming Department 120: Analysis Department 130: Processing unit

Claims

Claim 1 A real-time livestock barn floor analysis system for livestock barn floor management, comprising: a shooting unit that generates an image containing status information of bedding distributed on the floor of the livestock barn; an analysis unit that analyzes the status information of the bedding by area of ​​the livestock barn and generates distribution information regarding the impregnation status by area; a comparison unit that determines an abnormal section by comparing the distribution information regarding the impregnation status by area with the distribution of a defecation preference area based on the location of the livestock at the time of defecation; and a processing unit that generates information for performing management tasks of the livestock barn by considering the abnormal section. Claim 2 A real-time livestock barn floor analysis system for livestock barn floor management, wherein, in claim 1, the image includes metadata regarding illumination or shooting date and time, and the analysis unit analyzes the condition information of the bedding material by considering the metadata. Claim 3 A real-time livestock barn floor analysis system for livestock barn floor management, wherein, in claim 1, a collection unit for collecting internal environment data of the livestock barn; and an analysis unit for analyzing the condition information of the bedding in consideration of the internal environment data. Claim 4 A real-time livestock barn floor analysis system for livestock barn floor management, further comprising: a display unit that displays a standardized image, which is corrected and converted to a color value at a reference point in time considering the date or time, in accordance with claim 1. Claim 5 A real-time livestock barn floor analysis system for livestock barn floor management, further comprising, in paragraph 4, a conversion unit that converts an input image into a standardized image using an artificial intelligence model based on a plurality of training images taken at different shooting times. Claim 6 A real-time livestock barn floor analysis system for livestock barn floor management according to claim 1, further comprising: a recognition unit that analyzes the behavioral patterns of livestock in the livestock barn to identify the defecation posture of the livestock and analyzes the distribution of the defecation preference area based on the position of the livestock at the time of defecation; wherein the analysis unit considers the distribution of the defecation preference area when determining the impregnation state for each area. Claim 7 In claim 6, the comparison unit compares the distribution information regarding the region-specific impregnation state generated by the analysis unit with the distribution of the defecation preference area to determine the non-overlapping area and determine the abnormal section, a real-time livestock barn floor analysis system for livestock barn floor management. Claim 8 In claim 7, the processing unit determines the bedding stirring pattern by considering the abnormal section, a real-time livestock barn floor analysis system for livestock barn floor management. Claim 9 A real-time livestock barn floor analysis system for livestock barn floor management, further comprising, in claim 8, a stirring unit that moves along a predetermined path and stirs the bedding based on the bedding stirring pattern. Claim 10 In claim 1, the analysis unit generates distribution information by considering shading information of the floor created as a small object within the image or an unidentified area of ​​the floor due to the small object, a real-time livestock barn floor analysis system for livestock barn floor management.

Images