Intelligent wearable inspection equipment for livestock farm

Abstract

The invention relates to the technical field of wearable monitoring, in particular to farm intelligent wearable inspection equipment, which comprises a track thermal difference recognition module, an abnormal target labeling module, a high fever behavior screening module, an image information extraction module, a structure label generation module and an individual inventory recognition module. According to the invention, by extracting the infrared temperature measurement data and calculating the thermal fluctuation difference value, the positioning of a temperature abnormal section is realized, the identification capability of tiny thermal environment variation is enhanced, the high-heat target area is positioned, the image filing is completed, and the target screening accuracy and the image utilization efficiency are improved. Static high-heat individuals are identified based on the calorific value state and the spatial position stability, the discovery probability of potential abnormal objects is improved, through image boundary contour extraction and structural feature comparison, the individual identification efficiency is improved, and image information and dynamic path data are synchronously analyzed, so that the monitoring result is more continuous and complete, and the detection accuracy is improved. And the intelligent response capability and the information processing depth are improved.

Description

Technical Field

[0001] This invention relates to the field of wearable monitoring technology, and in particular to an intelligent wearable inspection device for livestock farms. Background Technology

[0002] Wearable monitoring technology falls under the cross-application direction of intelligent equipment and information perception. It primarily involves the real-time monitoring and recording of the status, location, and behavior of target objects through wearable terminals. Its core aspects include data acquisition, information transmission, and visualization using portable sensors, and the use of multi-sensor fusion, wireless communication, and edge computing to dynamically perceive and manage specific objects. In this field, wearable monitoring devices are widely used in industrial inspections, medical monitoring, security management, and livestock management. Traditional smart wearable inspection devices for livestock farms refer to wearable monitoring devices with inspection functions applied in livestock and poultry farming environments. The technical challenge they address is the incomplete information collection and difficulty in recording inspection trajectories during manual inspections of the breeding environment and livestock and poultry status within the farming area. Traditional farm inspection methods rely on manual wearing of terminals with single positioning functions combined with paper records to complete inspection tasks. These terminals typically use a GPS receiver module to collect location information and manually record the environment and livestock and poultry status.

[0003] Traditional inspection methods rely on personnel wearing terminal devices with single positioning functions, combined with paper records to register environmental information and livestock status. In practice, this approach suffers from fragmented information and limited content. Because location information and status records are disconnected, it's difficult to establish continuous data trajectories, making it hard to reconstruct the correlation between environmental changes and target behavior. This information isolation severely impacts the value of post-inspection data analysis. Furthermore, paper-based methods are slow to react to abnormal thermal environments and cannot visually identify details of surface temperature changes, easily overlooking individuals with early potential lesions. During inspections, relying solely on location modules to generate location information records lacks image-level structural feature references, hindering effective classification and comparative analysis of individual appearance and status. This results in limited information hierarchy, difficulty in individual identification, and high repetition rates in statistical records, severely restricting the efficiency and accuracy of inspection management. Summary of the Invention

[0004] To address the technical problems existing in the prior art, this invention provides a smart wearable inspection device for livestock farms. The technical solution is as follows: On the one hand, a smart wearable inspection device for farms is provided, the system comprising: The trajectory thermal difference recognition module acquires infrared temperature measurement data of the wearable device along the inspection path, organizes the temperature sequence, constructs the difference set, calculates the thermal fluctuation reference value, segments according to the time index, identifies the temperature abnormal area in the path, and generates abnormal path location data. The abnormal target annotation module calls the time and location in the abnormal path location data, extracts heat value distribution features from the image, identifies areas with concentrated heat values, archives associated image numbers, and generates a list of high-heat image frame numbers. The high-temperature behavior screening module extracts the spatial displacement trajectory of the target area based on the list of high-temperature image frame numbers, and combines it with the heat value status to identify objects with stable positions and abnormal heat values ​​within a specified time period, and generates static high-temperature target labels. The image information extraction module calls the static high-temperature target label image number, extracts the boundary contour of the pig image, identifies the length, height and regional structure information, and forms individual structural feature data; The structural label generation module calculates the structural proportion and area ratio of the target contour based on the image boundary parameters in the individual structural feature data, judges the state in conjunction with the reference interval, assigns differential structural labels, and generates an inspection structural state label set.

[0005] As a further embodiment of the present invention, the abnormal path location data includes path points, time index, and temperature abnormality sections; the high-temperature image frame number list includes image number, heat value distribution characteristics, and heat area identification information; the static high-temperature target label includes target number, spatial location trajectory, and heat value status information; the individual structural feature data includes image boundary contour, horizontal length, vertical height, and image area range; and the inspection structure status label set includes structural ratio value, image area ratio, and differential structure label identifier.

[0006] As a further aspect of the present invention, the trajectory thermal difference recognition module includes: The temperature sequence processing submodule acquires the infrared temperature measurement data recorded by the wearable device on the inspection path, processes it in chronological order according to the device location information, constructs the temperature sequence of the path points, extracts the temperature value pairs between adjacent path points, processes them to obtain the set of temperature differences between adjacent points, and obtains the temperature difference value between adjacent points. The thermal fluctuation calculation submodule, based on the temperature difference value of the adjacent points, calls the time information of the path points to segment the temperature difference value, extracts the temperature difference set with continuous time, calculates the average temperature difference of each set, filters out abnormal temperature difference values, regenerates the average temperature difference, and obtains the thermal fluctuation reference value. The abnormal path extraction submodule calls the thermal fluctuation reference value, segments the temperature difference value based on the time period of the inspection path, determines whether the temperature difference value of consecutive path points in each segment exceeds the reference value, filters the path segments that continuously exceed the reference value, and marks the equipment location and time index to obtain abnormal path location data.

[0007] As a further aspect of the present invention, the abnormal target annotation module includes: The image extraction submodule calls the time period and location recorded in the abnormal path location data to extract the original video image data collected by the corresponding device, filters the image frame sequence in time order, and extracts the heat value distribution information of each frame in the image frame sequence to obtain the heat value characteristics of the image frame. Based on the heat value features of the image frame, the heat zone identification submodule determines the continuity and displacement stability of the high heat value region in the image frame, selects heat value regions with continuous distribution features and stable position changes as candidate heat target regions, and marks the corresponding region number to obtain the heat value concentration region identifier. The numbering and archiving submodule extracts the associated image frame numbers based on the heat value concentration area identifier, sorts and classifies all the numbers, summarizes the filtered image numbers into a unified sequence, and generates a list of high-heat image frame numbers.

[0008] As a further aspect of the present invention, the high-temperature behavior screening module includes: The trajectory extraction submodule, based on the list of high-temperature image frame numbers, calls the target region position parameters in the image frame sequence, arranges the center position of the target region in the image frames in chronological order, calculates the position change characteristics between adjacent frames, and generates regional displacement trajectory data. The state recognition submodule extracts the heat value sequence corresponding to the target area in the image frame based on the regional displacement trajectory data, calculates the regional position change amplitude and heat value change trend, and filters the frame numbers whose position change is lower than the spatial displacement stability threshold and whose heat value is higher than the heat value abnormal threshold within a specified time period to obtain the static heat value abnormal sequence. The target labeling submodule calls the target region coordinates in the image frames corresponding to the static calorific value anomaly sequence, classifies and marks the regions based on the consistency of inter-frame positions, numbers the target region set in the recognition results, and establishes static high-heat target labels.

[0009] As a further aspect of the present invention, the image information extraction module includes: The image localization submodule calls the image number in the static high-heat target label, extracts the corresponding frame image from the image sequence, locates the spatial boundary position of the labeled area in each frame image, marks the image coordinate range of the area corresponding to the pig, and establishes target area index data; The contour recognition submodule extracts the set of pixels within the marked area in the image based on the target area index data, identifies the difference in edge pixel distribution density, performs threshold judgment on the gray-scale difference between pixels inside and outside the edge line, obtains the set of contour boundary coordinates, and generates individual boundary contour data. The structure extraction submodule calls the horizontal length and vertical height values ​​between boundary coordinate points in the individual boundary contour data to estimate the area of ​​the boundary-enclosed region, merge and organize the structural parameters, and establish individual structural feature data.

[0010] As a further aspect of the present invention, the structural label generation module includes: The proportion calculation submodule calculates the structural proportion value within the boundary range and the area proportion parameter in the image region based on the image boundary parameters in the individual structural feature data, and uniformly classifies them into the same data group to generate a structural proportion parameter group. The status determination submodule calls the structural proportion value and area proportion parameter in the structural proportion parameter group, performs status determination according to the upper and lower boundaries of the set reference interval, identifies parameter items that exceed the interval range, and obtains structural offset label data. The tag generation submodule establishes a mapping relationship between the identifiers in the structural offset tag data and the corresponding image numbers, combines the structural status identifier content contained in the image, and obtains the inspection structural status tag set.

[0011] As a further aspect of the present invention, the system further includes: The individual inventory identification module, based on the image tag information in the structural state tag set, calls the image frame sequence recorded by the helmet device in the same inspection cycle, analyzes the structural features, body shape outline and image boundary position of the target object in the frame, determines whether there are image records with the same structural parameters and overlapping outlines, filters out the unique target object, completes automatic number matching based on the image number, and generates a statistical list of pig numbers inspected by the wearable device. The wearable device's inspection list of pig numbers includes target structural features, body shape outline, image boundary position, and image number matching relationship; The image frame sequence refers to the set of static frames extracted from video images continuously collected by the wearable device within the same inspection cycle. The image number refers to the numbering information used to uniquely identify an image frame. The structural state label set refers to the set of classification labels generated based on image structural information analysis.

[0012] As a further aspect of the present invention, the individual inventory identification module includes: The image extraction submodule, based on the image tag information in the structural state tag set, calls the image frame sequence recorded by the wearable helmet device to extract the structural features, body contours and image boundary position information of the target object in the image, and combines them to form object image description data under continuous frames to obtain the inspection image structural parameter group; The structure matching submodule calls the structural features and contour position parameters in the inspection image structure parameter group to determine whether there are image records with consistent structural parameters and overlapping contour boundaries in adjacent frames. It then filters object images that are repeatedly recorded in the same inspection cycle, establishes a unique matching item, and obtains the target object matching result. The number generation submodule performs number comparison and information integration operations based on the image number sequence corresponding to the unique object in the target object matching result, generates a statistical number list of the target object, and obtains the wearable device number list.

[0013] As a further aspect of the present invention, the wearable device refers to an intelligent terminal device worn by a person and integrating image acquisition, infrared temperature measurement, and positioning recording functions, preferably an intelligent helmet or a head-mounted sensor system; The inspection path refers to the sequence of points formed by the wearable device walking along a set route within the breeding farm area, which has timestamps and spatial coordinates; The infrared temperature measurement data refers to the sequence of animal body surface temperature values ​​collected by wearable devices through infrared sensors during the inspection process; The thermal fluctuation reference value refers to the change ratio obtained by constructing a set of differences from adjacent infrared temperature measurement data; The heat value distribution feature refers to the heat distribution structure formed at the pixel level in the region of concentrated heat energy in an infrared image frame; The image boundary contour refers to the closed contour structure extracted after edge detection of the pig object during image processing; The structural ratio refers to the proportional relationship between the horizontal pixel length and the vertical pixel height of the target object in the image, reflecting the structural characteristics of the pig's body shape; The area ratio refers to the percentage of the pixel area occupied by the pig object in the image relative to the entire image.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by extracting infrared thermometry data and calculating thermal fluctuation differences, the location of temperature anomaly zones is achieved, enhancing the ability to identify minute thermal environmental variations. High-temperature target areas are located and images are archived, improving the accuracy of target selection and image utilization efficiency. Based on calorific value status and spatial location stability, stationary high-temperature individuals are identified, increasing the probability of discovering potential abnormal objects. Individual identification efficiency is improved through image boundary contour extraction and structural feature comparison. Image information and dynamic path data are analyzed synchronously, making the monitoring results more continuous and complete, and enhancing intelligent response capabilities and information processing depth. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a system flowchart of the present invention; Figure 2 This is a system block diagram of the present invention; Figure 3 This is a flowchart of the trajectory thermal difference recognition module of the present invention; Figure 4 This is a flowchart of the abnormal target annotation module of the present invention; Figure 5 This is a flowchart of the high-temperature behavior screening module of the present invention; Figure 6 This is a flowchart of the image information extraction module of the present invention; Figure 7 This is a flowchart of the structural label generation module of the present invention; Figure 8 This is a flowchart of the individual inventory identification module of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0019] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0020] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0021] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0022] This invention provides an intelligent wearable inspection device for livestock farms, such as... Figure 1-2 The diagram shows a smart wearable inspection device for a farm. The system includes: The trajectory thermal difference recognition module acquires infrared temperature measurement data recorded by the wearable device on the inspection path, organizes the temperature sequence of points along the path, constructs a difference set, calculates the thermal fluctuation reference value, and performs segmentation processing according to the inspection time index to identify temperature abnormal areas in continuous path segments. Combined with the device location and time, it generates abnormal path location data. Wearable devices refer to intelligent terminal devices worn by personnel that integrate image acquisition, infrared temperature measurement, and positioning recording functions. They are preferably intelligent helmets or head-mounted sensor systems used for on-site dynamic inspection tasks. The inspection path refers to the sequence of points formed by wearable devices walking along a set route within the breeding farm area. It has timestamps and spatial coordinates and is used to construct a data collection time sequence framework. Infrared thermometry data refers to the sequence of animal body surface temperature values ​​collected by wearable devices through infrared sensors during inspections, which is used to identify local thermal anomalies. Thermal fluctuation reference value refers to the change ratio obtained by constructing a set of differences from adjacent infrared temperature measurement data. It is used to identify local temperature anomalies in the path and to make judgments in combination with time series points. The abnormal target labeling module calls the time period and location recorded in the abnormal path location data, extracts the heat value distribution features from the video images collected by the device, identifies the heat value concentration area by the continuity and displacement stability of the heat area in the image frame, archives the image number corresponding to the area, and generates a list of high heat image frame numbers. The thermal distribution feature refers to the thermal distribution structure formed at the pixel level in the area of ​​concentrated thermal energy in an infrared image frame, which is used to help determine the thermal anomaly of the target object; The high-temperature behavior screening module extracts the spatial displacement trajectory of the target area in continuous images based on the high-temperature image frame number list, and combines the heat value status of the area to identify objects with stable spatial location and abnormal heat value within a specified time period, completes the target labeling process, and generates static high-temperature target labels. The image information extraction module calls the image number in the static high-heat target label, extracts the boundary contour of the pig in the corresponding image, identifies the horizontal length, vertical height and image area range structural information, completes the collection and organization of structural information, and generates individual structural feature data. Image boundary contours refer to the closed contour structures extracted after edge detection of pig objects during image processing, and are used to identify the object's structural shape and image localization; The structural label generation module calculates the structural proportion value and image area of ​​the target contour based on the image boundary parameters in the individual structural feature data, combines the reference interval to make a state judgment, and assigns differential structural label identifiers to generate an inspection structure state label set. The structural ratio value refers to the proportional relationship between the horizontal pixel length and the vertical pixel height of a target object in an image, reflecting the structural characteristics of a pig's body shape. Image proportion area refers to the percentage of pixel area occupied by a pig object in an image relative to the entire image, and is used to assess the size of an individual. The individual inventory identification module uses image tag information from the structural state tag set to call the image frame sequence recorded by the helmet device in the same inspection cycle, analyzes the structural features, body shape outline and image boundary position of the target object in the frame, determines whether there are image records with the same structural parameters and overlapping outlines, filters out the unique target object, completes automatic number matching based on the image number, and generates a statistical list of pig numbers inspected by the wearable device. An image frame sequence refers to a set of static frames extracted from video images continuously captured by a wearable device within the same inspection cycle, sorted by shooting time and used for target recognition processing; Image number refers to the numbering information used to uniquely identify an image frame, and it is used to establish a numbering index in image recording, target recognition, and statistical matching; The structural state label set refers to the set of classification labels generated based on image structural information analysis. It is used to represent the structural state (including normal and abnormal) of pig targets with differences and for identification, numbering and classification.

[0023] The abnormal path location data includes path locations, time indexes, and temperature anomaly sections; the high-temperature image frame number list includes image number, calorific value distribution characteristics, and hot area identification information; the static high-temperature target label includes target number, spatial location trajectory, and calorific value status information; the individual structural feature data includes image boundary contours, horizontal length, vertical height, and image area range; the inspection structure status label set includes structural ratio values, image area proportions, and differential structure label identifiers; and the wearable device inspection pig number statistics list includes target structural features, body shape contours, image boundary positions, and image number matching relationships.

[0024] Specifically, such as Figure 2 , 3 As shown, the trajectory thermal difference recognition module includes: The temperature sequence processing submodule acquires the infrared temperature measurement data recorded by the wearable device on the inspection path, processes it in chronological order according to the device location information, constructs the temperature sequence of the path points, extracts the temperature value pairs between adjacent path points, processes them to obtain the set of temperature differences between adjacent points, and obtains the temperature difference value between adjacent points. Infrared thermometry records data including temperature, time, and location coordinates during inspections. First, the data is sorted chronologically to form a temperature measurement sequence. Each record includes the temperature value and corresponding latitude and longitude. If 10 data points are collected during a single inspection, these will form 10 sequentially arranged data records. The temperature values ​​of any two adjacent measurement points are extracted to form temperature pairs. The difference between each pair is then calculated, with the absolute value of the difference between the two being the temperature difference. For example, if the temperature at the first point is 35.6°C and the second is 36.1°C, the temperature difference is 0.5°C. This process is repeated for each point, from the first to the ninth, to all subsequent points, forming nine consecutive temperature differences. All these temperature differences constitute a set of temperature difference values, providing a basis for subsequent thermal change analysis.

[0025] The thermal fluctuation calculation submodule is based on the temperature difference value of adjacent points. It calls the time information of the path points to segment the temperature difference value, extracts the temperature difference set with continuous time, calculates the average temperature difference of each set, filters out abnormal temperature difference values, regenerates the average temperature difference, and obtains the thermal fluctuation reference value. The temperature difference set combines the time information recorded at each measurement point, dividing the path into multiple consecutive time periods based on time. For example, each period can be set to 10 minutes. Consecutive points falling within the same time period are grouped together. All temperature difference values ​​in this group are averaged to obtain the mean temperature difference within that time period. For example, if the temperature differences recorded in a certain time period are 0.4, 0.7, and 0.3, the corresponding mean is 0.467°C. To eliminate abnormal temperature differences that deviate from the norm, the median and standard deviation of the average temperature difference set for all time periods are calculated. Mean values ​​that differ from the median by more than twice the standard deviation are removed from the set, leaving only the remaining values. Their average is then calculated again as a benchmark value for determining whether temperature fluctuations are abnormal. For example, if the remaining average temperature differences in the effective set are 0.41, 0.46, and 0.45, averaging them yields a benchmark temperature difference of 0.44°C.

[0026] The abnormal path extraction submodule calls the thermal fluctuation reference value, segments the temperature difference value based on the time period of the inspection path, determines whether the temperature difference value of consecutive path points in each segment exceeds the reference value, filters the path segments that continuously exceed the reference value, and marks the equipment location and time index to obtain abnormal path location data. The baseline temperature difference value is applied to each time period after the inspection path is divided. The temperature difference of continuous points within each segment is traversed and judged. Whenever a temperature difference value is higher than the reference value, the point is marked as an abnormal point. If two or more consecutive temperature differences exceed the reference value, the path corresponding to that time period is determined to be an abnormal segment. For example, if the temperature difference values ​​in a segment are 0.3, 0.47 and 0.5, the latter two are higher than the reference value of 0.44°C. This segment is identified as an abnormal path segment. The location information and time records contained in this segment are further extracted and marked as abnormal path points. The location and time data of all path segments that meet the continuous out-of-tolerance condition are summarized into an abnormal path data set.

[0027] Specifically, such as Figure 2 , 4 As shown, the anomaly target annotation module includes: The image extraction submodule calls the time period and location recorded in the abnormal path location data to extract the original video image data collected by the corresponding device, filters the image frame sequence in time order, and extracts the heat value distribution information of each frame in the image frame sequence to obtain the heat value characteristics of the image frame. Based on the time period and location information of the abnormal path, the original infrared video data corresponding to the time and location is first retrieved. The video is then split into frames according to the time sequence to generate an image frame sequence. Each image frame contains a unique timestamp and image number. Subsequently, the image frame sequence is traversed to extract the thermal distribution information of each image frame. Thermal value is obtained by reading the grayscale value in the image and mapping the grayscale of each pixel to a temperature value according to the infrared imaging calibration parameters of the device. For example, a pixel with a grayscale of 128 has a temperature of 32 degrees Celsius under the condition that the mapping coefficient is 0.1 degrees Celsius per grayscale level and the initial temperature is 20 degrees Celsius. And so on. Temperature mapping is performed on all pixels in the entire image frame to form a complete thermal distribution matrix. The thermal value matrix is ​​a two-dimensional structure, corresponding to the rows and columns of pixels in the image. After traversal, each image frame is bound to its thermal value matrix. By processing all image frames, an image frame and thermal value feature set is constructed. This set has the characteristics of temporal continuity and spatial distribution consistency and can be used for subsequent region recognition processes.

[0028] The hot zone identification submodule judges the continuity and displacement stability of high heat value areas in the image frame based on the heat value features of the image frame. It selects heat value areas with continuous distribution characteristics and stable position changes as candidate hot target areas, and marks the corresponding area number to obtain the heat value concentration area identifier. By combining image frames and thermal value feature sets, the continuity and positional stability of high-heat areas in the image are determined frame by frame. First, based on the temperature range of the infrared device, a threshold range for determining high heat is set. For example, when the thermal value is generally distributed between 20 and 50 degrees Celsius, 40 degrees Celsius can be set as the recognition threshold. Areas exceeding this value in the thermal value matrix of each frame are marked to form an initial distribution map of hot spots. The spatial position and shape changes of each hot spot area are tracked in multiple consecutive image frames. If there are hot spots in consecutive frames whose area remains basically consistent and whose displacement does not exceed two pixels, they are considered to have thermal value continuity and displacement stability. This area is further listed as a candidate hot target area and assigned a number. For example, if the path segment number is 3, the second hot spot can be numbered 3_2. The numbering structure includes path segment information and area sequence number. Each stable hot spot has an independent number and forms a corresponding mapping with its position in the image frame.

[0029] The numbering and archiving submodule extracts the associated image frame numbers based on the heat value concentration area identifier, sorts and classifies all numbers, summarizes the filtered image numbers into a unified sequence, and generates a list of high heat image frame numbers. Based on the numbering information of stable hot zones, the frame numbers in which they appear in the image frame sequence are retrieved in reverse. After summarizing all the numbers, they are sorted and classified. For example, if a hot zone number appears in frames 21 to 24, numbers 21, 22, 23, and 24 are extracted. If another hot zone appears in frames 31, 32, 34, and 35, they are extracted in the same way. After extracting all the frame numbers associated with the hot zones, the number set is deduplicated and sorted, the belonging path segments are divided, and the classified image frame numbers are merged to form a unified sequence. The output is a list of high-temperature image frame numbers. This list provides the time and location index information of the hot target area contained in each frame and has a complete path segment grouping and image frame number structure.

[0030] Specifically, such as Figure 2 , 5 As shown, the high fever behavior screening module includes: The trajectory extraction submodule is based on the list of high-temperature image frame numbers. It calls the target region position parameters in the image frame sequence, arranges the center position of the target region in the image frames in chronological order, calculates the position change characteristics between adjacent frames, and generates regional displacement trajectory data. Based on the list of high-temperature image frame numbers, the positional parameter information of the target region in the corresponding frame image is retrieved one by one. The positional parameters include the bounding box coordinates and center coordinates of the target region. The center coordinates can be obtained by reading the coordinates of the upper left and lower right corners of the region boundary and taking the midpoint of the horizontal and vertical directions. After performing the center coordinate extraction operation on each frame image, all center points are arranged in chronological order to form a region position sequence. Then, the positional changes between consecutive frames are compared in sequence according to the frame number. The movement amplitude of the target region is reflected by calculating the change distance of the horizontal and vertical coordinates. For example, if the center of the target region in a certain frame is 100 on the horizontal coordinate and 150 on the vertical coordinate, and the next frame is 104 and 153, the change between the two frames is an increase of 4 pixels in the horizontal direction and an increase of 3 pixels in the vertical direction. The above process is repeated for the entire image frame sequence to form continuous movement data of the target region. By accumulating and statistically analyzing the displacement values ​​between each pair of adjacent frames, the dynamic trajectory information of the target region in the image frame can be established to comprehensively describe the displacement characteristics of the thermal target region in the high-temperature image frame over time.

[0031] The state recognition submodule extracts the heat value sequence corresponding to the target area in the image frame based on the regional displacement trajectory data, calculates the magnitude of regional position change and heat value change trend, and filters the frame numbers whose position change is lower than the spatial displacement stability threshold and whose heat value is higher than the heat value abnormal threshold within a specified time period to obtain the static heat value abnormal sequence. Based on regional displacement trajectory data, the average heat value of the target region in the corresponding frame image is extracted one by one. The average heat value can be obtained by averaging the temperature values ​​of all pixels in the target region in the frame image. After arranging them in the time sequence of the image frames, a heat value sequence is formed. Then, the displacement amplitude and heat value of each frame are jointly analyzed. When setting the spatial displacement stability judgment threshold, the image resolution and the shooting angle of the device should be comprehensively considered. For example, if the image is 1280 x 720 resolution and the target is shot by a fixed device, then when the center change of the target region between consecutive frames is less than 3 pixels, it can be regarded as a stationary state. The heat value abnormality threshold is set according to the scene temperature benchmark value. For example, in the normal background temperature distribution range of 30 to 40 degrees Celsius, the abnormal high heat state can be defined as above 45 degrees Celsius. The displacement amplitude and heat value are compared simultaneously in each frame of data. All image frame numbers that are both stationary and have abnormally high heat values ​​in the time continuous region are selected and constructed into a stationary heat value abnormality sequence, which is used to further identify thermal target areas that may have abnormal activity.

[0032] The target labeling submodule calls the target region coordinates in the image frames corresponding to the static calorific value anomaly sequence, classifies the region markings based on the consistency of inter-frame positions, numbers the target region set in the recognition results, and establishes static high-heat target labels. Based on the image frame numbers in the static heat value anomaly sequence, the contour boundary coordinate information of the target region is extracted frame by frame, and the region center position parameter is read. The target region positions in different frame images are compared with each other. If the target region positions in multiple frame images remain highly consistent, that is, the center coordinate changes by no more than 3 pixels, it can be regarded as the continuous occurrence of the same target region in the time series. According to this rule, each region is clustered into the same group and classified into the same target region category. A unique number is assigned to each type of target region. The number can be represented by a combination of type identifier and number. For example, "static high heat_01" represents the first identified static high heat target region. Each group of numbers represents a heat anomaly target region, forming a static high heat target label dataset.

[0033] Specifically, such as Figure 2 , 6 As shown, the image information extraction module includes: The image localization submodule calls the image number in the static high-heat target label, extracts the corresponding frame image from the image sequence, locates the spatial boundary of the labeled area in each frame image, marks the image coordinate range of the area corresponding to the pig, and establishes target area index data; The system retrieves the corresponding frame images one by one by calling the image number provided by the static high-heat target label. Combining the coordinate range information given in the label, the target region in the image is extracted. The spatial range of the region is determined by the upper left and lower right corners. For example, in a certain image frame, the upper left corner coordinates are 230 pixels horizontally and 120 pixels vertically, and the lower right corner coordinates are 340 pixels horizontally and 260 pixels vertically. Then the region width is 110 pixels and the height is 140 pixels. The region is marked with a rectangle on the image to identify the spatial range of the target. A correspondence is established between the region and the image frame number, and a unique target number is assigned, for example, T_05. The same process is performed on all numbered image frames in sequence to record the spatial location, number, and frame index of each target region. A target region location index data structure is established by field aggregation. Each record includes the image number, target number, horizontal and vertical coordinate range, etc., forming a structured spatial location information set.

[0034] The contour recognition submodule extracts the set of pixels within the marked area in the image based on the target area index data, identifies the difference in edge pixel distribution density, performs threshold judgment on the gray-scale difference between pixels inside and outside the edge line, obtains the set of contour boundary coordinates, and generates individual boundary contour data. Based on the target region location index information, the marked region image is cropped from the corresponding frame image and converted into a grayscale image for processing. Grayscale readings are performed on each pixel, and its grayscale distribution characteristics are analyzed. The abrupt change characteristics of grayscale values ​​are detected at the outer and inner boundaries of the marked region. Pixels with significant grayscale changes at the edges are extracted as contour boundary points. If the grayscale inside the region is concentrated between 180 and 200, while the grayscale of the adjacent outer region decreases to between 90 and 110, and the grayscale difference exceeds 50, the point can be identified as a boundary pixel. The entire region is scanned pixel by pixel, and contour recording operations are performed on pixels that meet the grayscale difference threshold. The coordinate information of these points is saved sequentially to form a continuous set of edge points. These coordinate points are represented by two-dimensional image pixel positions, such as horizontal 234 vertical 122, horizontal 235 vertical 122, etc., forming the boundary coordinate set of the target region. This set covers the entire contour structure of the extracted target and reflects its complete edge shape in the image.

[0035] The structure extraction submodule calls the horizontal length and vertical height values ​​between boundary coordinate points in the individual boundary contour data, estimates the area of ​​the boundary-enclosed region, merges and organizes the structural parameters, and establishes individual structural feature data. Using the contour boundary coordinate set, the width of the target is determined by analyzing the minimum and maximum lateral position differences among all boundary points. Simultaneously, the height is estimated using the minimum and maximum vertical position differences. For example, if the minimum lateral coordinate of a boundary point is 225 and the maximum is 355, the width is 130 pixels; if the minimum vertical coordinate is 115 and the maximum is 275, the height is 160 pixels. Multiplying the width and height yields an approximate area of ​​20,800 square pixels. The aspect ratio parameter is then extracted (in this case, 0.81). Structures are classified according to aspect ratio ranges; values ​​between 0.75 and 0.90 are categorized as elliptical structures. By classifying and organizing these structural parameter values ​​(width, height, area, aspect ratio, etc.) and associating them with target numbers, a dataset describing the structural characteristics of static high-heat regions is constructed. Each target number corresponds to a set of structural data for subsequent comparison and structural analysis.

[0036] Specifically, such as Figure 2 , 7 As shown, the structure tag generation module includes: The proportion calculation submodule calculates the structural proportion value within the boundary range and the area proportion parameter in the image region based on the image boundary parameters in the individual structural feature data, and uniformly classifies them into the same data group to generate a structural proportion parameter group. Based on the image boundary parameters in the individual structural feature data, the set of contour coordinates within the boundary range of the target region is extracted. The difference between the maximum and minimum horizontal coordinate values ​​of the contour points is used to obtain the target structure width. The same operation is performed on the vertical coordinates to obtain the structure height. Then, the original width and height values ​​of the bounding box are proportionally converted. If the width of the bounding box of a target region is 150 pixels and the height is 180 pixels, and the width of the contour range is 120 pixels and the height is 160 pixels, then the horizontal ratio is 80% and the vertical ratio is 88.89%. Then, the area occupied by the contour structure within the boundary range is extracted. A closed region is formed by the contour points and its total number of pixels is calculated. The area ratio is compared with the area of ​​the bounding box to obtain the area ratio. For example, if the area of ​​the closed region is 16800 pixels² and the area of ​​the bounding box is 27000 pixels², then the area ratio is 62.22%. Other structural regions are processed in the same way. The three values ​​of horizontal structure ratio, vertical structure ratio and area ratio of each target are unified and grouped into a set of parameters, labeled with the image number and target number, and organized into a structure ratio parameter group. The parameter group is numbered and classified in a unified format to provide a basic data source for subsequent judgment and processing.

[0037] The status determination submodule calls the structural proportion value and area proportion parameter in the structural proportion parameter group, performs status determination according to the upper and lower boundaries of the set reference interval, identifies parameter items that exceed the interval range, and obtains structural offset label data. Each ratio value in the structural proportion parameter group is called, and a status judgment is performed against the set reference range. The reference range for the horizontal structural proportion is set to 75% to 90%, the vertical structural proportion to 80% to 95%, and the area proportion to 60% to 85%. When the horizontal proportion is 70%, the vertical proportion is 91%, and the area proportion is 58%, it is compared with the set range. If the horizontal proportion is below the lower limit and the area proportion is also below the lower limit, and the vertical proportion is within the reference range, the horizontal structural proportion and area proportion are determined to be in a downward bias state. The offset item is extracted and recorded as a structural status identifier. The offset direction and offset type are recorded using a unified description method, such as "horizontal proportion_downward bias" and "area proportion_downward bias". During the status judgment process, each parameter item is compared with the corresponding upper and lower limits individually. Conditional judgment statements are used to identify whether there is an over-limit situation. The structural status offset labels are organized and archived according to the image number and target number for easy subsequent summary and label generation operations.

[0038] The tag generation submodule establishes a mapping relationship between the identifiers in the structural offset tag data and the corresponding image numbers, combines the structural status identifiers contained in the images, and obtains the inspection structural status tag set. Based on the target number and offset identifier recorded in the structural offset label data, the image number information to which the target number belongs is retrieved. For each image number, all associated offset labels are merged. A mapping table is established to associate the structural status labels with the image numbers. If an image number is IMG_045 and the associated target number is T_07, and its structural offset labels include "lateral proportion_downward offset" and "area proportion_downward offset", then IMG_045 is combined with these two structural status labels to generate a complete image structural status label set. This process relies on the binding relationship between the target number and the image number to map the offset information from the target level to the image level, forming a set of structural status information corresponding to each image. All image numbers and corresponding structural label records are summarized to form a label set data, which serves as one of the structural status output results during the inspection process.

[0039] Specifically, such as Figure 2 , 8 As shown, the individual inventory identification module includes: The image extraction submodule uses the image tag information in the structural state tag set to call the image frame sequence recorded by the wearable helmet device to extract the structural features, body shape contour and image boundary position information of the target object in the image, and combines them to form object image description data in continuous frames to obtain the inspection image structural parameter group; Based on the image tag information in the structural state tag set, the image frame sequence acquired by the wearable helmet device is called frame by frame. Target recognition and structural information extraction are performed on each frame. First, the target region is located by analyzing the color differences, edge contours, and shape features of significant areas in the image. Then, an edge tracking method is used to obtain the contour boundary point set. The maximum and minimum horizontal and vertical coordinates are extracted to determine the boundary rectangular region where the target structure is located. For example, in image frame IMG_001, the boundary extends from column 28 to column 182 and row 43 to row 215, with a structural width of 154 pixels and a height of 172 pixels. Further analysis of the closed contour region... The area of ​​the region is estimated to be approximately 26488 pixels² by estimating the area of ​​the pixels within the region. The average grayscale value of the pixels within the region is extracted as the structural texture feature. The corner points of the boundary are counted. If the number is greater than 8, it can be recorded as a complex structural shape. The parameters such as the width and height of the structure, the number of contour feature points, the area value, the centroid coordinates, the boundary range, and the orientation angle are combined into a complete target structure description information, which is bound and stored with the current image frame number and the target number. The above process is repeated for consecutive frames from IMG_002 to IMG_010. The obtained structural description data is organized into a complete set of image structure parameters.

[0040] The structure matching submodule calls the structural features and contour position parameters in the structural parameter group of the inspection image to determine whether there are image records with the same structural parameters and overlapping contour boundaries in adjacent frames. It filters the object images that are repeatedly recorded in the same inspection cycle, establishes a unique matching item, and obtains the target object matching result. Information such as structural width, height, boundary position, and number of turns is extracted from the image structural parameter set. The structural feature similarity of the target record in consecutive image frames is compared one by one. A structural matching tolerance of ±10% is set. If a target has a width of 150 pixels and a height of 170 pixels in IMG_003, and a width of 145 pixels and a height of 165 pixels in IMG_004, the errors are 3.3% and 2.9% respectively, both within the tolerance range. The next step is to determine the spatial overlap of the contour boundaries. By calculating the area ratio of the overlapping regions of the contours in adjacent frames, if the overlapping areas... If the area exceeds 80% of the structural region in the current frame, the records in the two images are determined to be the same target instance. For example, the overlapping area in IMG_003 and IMG_004 is 24,000 pixels², while the structural region in one frame is 27,000 pixels², with an overlap rate of 88.9%, which meets the matching requirements. Similarly, targets in all image frames are screened in groups to confirm the continuous occurrence of the same object in different frames. Target records that meet the conditions of consistent structural parameters and overlapping boundary heights are merged into a unique matching object, forming a target matching result and marking the corresponding image number.

[0041] The number generation submodule performs number comparison and information integration operations based on the image number sequence corresponding to the unique object in the target object matching results, generates a statistical number list of the target object, and obtains a list of wearable device numbers. Based on the unique objects identified in the target matching results and their number sequence in the image frames, a list of image numbers corresponding to each object is compiled and a number identifier is generated. The frame numbers in each matching record are extracted. For example, OBJ_05 corresponds to IMG_010 to IMG_016. When compiling the numbers, the continuity of the frame numbers and the consistency of the target numbers are first confirmed. If the conditions are met, the object number is generated as OBJ_05_7, indicating that the object is recorded in 7 frames. At the same time, a rule is set to regard targets with more than 5 frames as high-frequency objects, and an identifier A is added to represent them, such as OBJ_05_7_A. To avoid number conflicts, a combination of number and frame number is used to ensure uniqueness, such as OBJ_05_IMG010_016, which is used to distinguish different record groups. The numbers of all target objects and the corresponding image frame sequence are summarized to form a number list, which includes each target number, the number of times it appears and the range of frame numbers it contains, and serves as a unified output number record for wearable acquisition data.

[0042] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A smart wearable inspection device for livestock farms, characterized in that, The system includes: The trajectory thermal difference recognition module acquires infrared temperature measurement data of the wearable device along the inspection path, organizes the temperature sequence, constructs the difference set, calculates the thermal fluctuation reference value, segments according to the time index, identifies the temperature abnormal area in the path, and generates abnormal path location data. The abnormal target annotation module calls the time and location in the abnormal path location data, extracts heat value distribution features from the image, identifies areas with concentrated heat values, archives associated image numbers, and generates a list of high-heat image frame numbers. The high-temperature behavior screening module extracts the spatial displacement trajectory of the target area based on the list of high-temperature image frame numbers, and combines it with the heat value status to identify objects with stable positions and abnormal heat values ​​within a specified time period, and generates static high-temperature target labels. The image information extraction module calls the static high-temperature target label image number, extracts the boundary contour of the pig image, identifies the length, height and regional structure information, and forms individual structural feature data; The structural label generation module calculates the structural proportion and area ratio of the target contour based on the image boundary parameters in the individual structural feature data, judges the state in conjunction with the reference interval, assigns differential structural labels, and generates an inspection structural state label set.

2. The intelligent wearable inspection device for livestock farms according to claim 1, characterized in that: The abnormal path location data includes path points, time index, and temperature anomaly segments; the high-temperature image frame number list includes image number, calorific value distribution characteristics, and hot area identification information; the static high-temperature target label includes target number, spatial location trajectory, and calorific value status information; the individual structural feature data includes image boundary contour, horizontal length, vertical height, and image area range; and the inspection structure status label set includes structural ratio value, image area ratio, and differential structure label identifier.

3. The intelligent wearable inspection device for livestock farms according to claim 1, characterized in that: The trajectory thermal difference recognition module includes: The temperature sequence processing submodule acquires the infrared temperature measurement data recorded by the wearable device on the inspection path, processes it in chronological order according to the device location information, constructs the temperature sequence of the path points, extracts the temperature value pairs between adjacent path points, processes them to obtain the set of temperature differences between adjacent points, and obtains the temperature difference value between adjacent points. The thermal fluctuation calculation submodule, based on the temperature difference value of the adjacent points, calls the time information of the path points to segment the temperature difference value, extracts the temperature difference set with continuous time, calculates the average temperature difference of each set, filters out abnormal temperature difference values, regenerates the average temperature difference, and obtains the thermal fluctuation reference value. The abnormal path extraction submodule calls the thermal fluctuation reference value, segments the temperature difference value based on the time period of the inspection path, determines whether the temperature difference value of consecutive path points in each segment exceeds the reference value, filters the path segments that continuously exceed the reference value, and marks the equipment location and time index to obtain abnormal path location data.

4. The intelligent wearable inspection device for livestock farms according to claim 3, characterized in that: The abnormal target annotation module includes: The image extraction submodule calls the time period and location recorded in the abnormal path location data to extract the original video image data collected by the corresponding device, filters the image frame sequence in time order, and extracts the heat value distribution information of each frame in the image frame sequence to obtain the heat value characteristics of the image frame. Based on the heat value features of the image frame, the heat zone identification submodule determines the continuity and displacement stability of the high heat value region in the image frame, selects heat value regions with continuous distribution features and stable position changes as candidate heat target regions, and marks the corresponding region number to obtain the heat value concentration region identifier. The numbering and archiving submodule extracts the associated image frame numbers based on the heat value concentration area identifier, sorts and classifies all the numbers, summarizes the filtered image numbers into a unified sequence, and generates a list of high-heat image frame numbers.

5. The intelligent wearable inspection device for livestock farms according to claim 4, characterized in that: The high-temperature behavior screening module includes: The trajectory extraction submodule, based on the list of high-temperature image frame numbers, calls the target region position parameters in the image frame sequence, arranges the center position of the target region in the image frames in chronological order, calculates the position change characteristics between adjacent frames, and generates regional displacement trajectory data. The state recognition submodule extracts the heat value sequence corresponding to the target area in the image frame based on the regional displacement trajectory data, calculates the regional position change amplitude and heat value change trend, and filters the frame numbers whose position change is lower than the spatial displacement stability threshold and whose heat value is higher than the heat value abnormal threshold within a specified time period to obtain the static heat value abnormal sequence. The target labeling submodule calls the target region coordinates in the image frames corresponding to the static calorific value anomaly sequence, classifies and marks the regions based on the consistency of inter-frame positions, numbers the target region set in the recognition results, and establishes static high-heat target labels.

6. The intelligent wearable inspection device for a farm according to claim 5, characterized in that: The image information extraction module includes: The image localization submodule calls the image number in the static high-heat target label, extracts the corresponding frame image from the image sequence, locates the spatial boundary position of the labeled area in each frame image, marks the image coordinate range of the area corresponding to the pig, and establishes target area index data; The contour recognition submodule extracts the set of pixels within the marked area in the image based on the target area index data, identifies the difference in edge pixel distribution density, performs threshold judgment on the gray-scale difference between pixels inside and outside the edge line, obtains the set of contour boundary coordinates, and generates individual boundary contour data. The structure extraction submodule calls the horizontal length and vertical height values ​​between boundary coordinate points in the individual boundary contour data to estimate the area of ​​the boundary-enclosed region, merge and organize the structural parameters, and establish individual structural feature data.

7. The intelligent wearable inspection device for livestock farms according to claim 6, characterized in that: The structural tag generation module includes: The proportion calculation submodule calculates the structural proportion value within the boundary range and the area proportion parameter in the image region based on the image boundary parameters in the individual structural feature data, and uniformly classifies them into the same data group to generate a structural proportion parameter group. The status determination submodule calls the structural proportion value and area proportion parameter in the structural proportion parameter group, performs status determination according to the upper and lower boundaries of the set reference interval, identifies parameter items that exceed the interval range, and obtains structural offset label data. The tag generation submodule establishes a mapping relationship between the identifiers in the structural offset tag data and the corresponding image numbers, combines the structural status identifier content contained in the image, and obtains the inspection structural status tag set.

8. The intelligent wearable inspection device for livestock farms according to claim 1, characterized in that: The system also includes: The individual inventory identification module, based on the image tag information in the structural state tag set, calls the image frame sequence recorded by the helmet device in the same inspection cycle, analyzes the structural features, body shape outline and image boundary position of the target object in the frame, determines whether there are image records with the same structural parameters and overlapping outlines, filters out the unique target object, completes automatic number matching based on the image number, and generates a statistical list of pig numbers inspected by the wearable device. The wearable device's inspection list of pig numbers includes target structural features, body shape outline, image boundary position, and image number matching relationship; The image frame sequence refers to the set of static frames extracted from video images continuously collected by the wearable device within the same inspection cycle. The image number refers to the numbering information used to uniquely identify an image frame. The structural state label set refers to the set of classification labels generated based on image structural information analysis.

9. The intelligent wearable inspection device for a farm according to claim 8, characterized in that: The individual inventory identification module includes: The image extraction submodule, based on the image tag information in the structural state tag set, calls the image frame sequence recorded by the wearable helmet device to extract the structural features, body contours and image boundary position information of the target object in the image, and combines them to form object image description data under continuous frames to obtain the inspection image structural parameter group; The structure matching submodule calls the structural features and contour position parameters in the inspection image structure parameter group to determine whether there are image records with consistent structural parameters and overlapping contour boundaries in adjacent frames. It then filters object images that are repeatedly recorded in the same inspection cycle, establishes a unique matching item, and obtains the target object matching result. The number generation submodule performs number comparison and information integration operations based on the image number sequence corresponding to the unique object in the target object matching result, generates a statistical number list of the target object, and obtains the wearable device number list.

10. The intelligent wearable inspection device for livestock farms according to claim 1, characterized in that: The wearable device refers to an intelligent terminal device worn by a person and integrating image acquisition, infrared temperature measurement, and positioning recording functions, preferably an intelligent helmet or a head-mounted sensor system; The inspection path refers to the sequence of points formed by the wearable device walking along a set route within the breeding farm area, which has timestamps and spatial coordinates; The infrared temperature measurement data refers to the sequence of animal body surface temperature values ​​collected by wearable devices through infrared sensors during the inspection process; The thermal fluctuation reference value refers to the change ratio obtained by constructing a set of differences from adjacent infrared temperature measurement data; The heat value distribution feature refers to the heat distribution structure formed at the pixel level in the region of concentrated heat energy in an infrared image frame; The image boundary contour refers to the closed contour structure extracted after edge detection of the pig object during image processing; The structural ratio refers to the proportional relationship between the horizontal pixel length and the vertical pixel height of the target object in the image, reflecting the structural characteristics of the pig's body shape; The area ratio refers to the percentage of the pixel area occupied by the pig object in the image relative to the entire image.

Images