Method for identifying physiological feature of animal, imaging device and medium

Abstract

The invention provides a method for identifying physiological features of an animal, imaging equipment and a medium. The method comprises the following steps: acquiring image data; identifying an animal target in the image data and a first physiological feature of the animal target; acquiring three-dimensional size data of the animal target; and according to the three-dimensional size data and the first physiological feature, obtaining the physiological feature of the animal target. According to the method, the three-dimensional size data of the animal target is obtained by using the distance measurement tool, a reliable physical scale basis can be provided, the inherent limitation of pure visual analysis is broken through by combining the first physiological feature based on image recognition and the three-dimensional size data obtained by distance measurement, and information complementation is realized. The three-dimensional size data obtained through distance measurement can effectively assist and optimize the physiological features obtained according to the image data, and the accuracy of animal target physiological feature recognition is remarkably improved.

Description

Technical Field

[0001] This application relates to the fields of imaging equipment and image processing technology, and in particular to a method, imaging equipment and medium for identifying physiological characteristics of animals. Background Technology

[0002] Imaging equipment utilizes specific light sources or sensors to non-invasively acquire visual images of observed targets, making it suitable for long-distance behavioral research and ecological monitoring of wild animals. Existing imaging equipment is also equipped with displays to show image data and animal information. This animal information is typically derived from image data. Therefore, the quality of the image data directly impacts the accuracy of the animal information.

[0003] Common types of imaging equipment include visible light imaging devices and infrared thermal imaging devices. Visible light imaging relies on ambient light and is easily affected by vegetation obstruction, weather conditions, and animal camouflage. Infrared thermal imaging devices, on the other hand, image based on differences in thermal radiation on the surface of objects and are capable of operating in all weather conditions. However, they are not sensitive to texture details, have limited dynamic range, and suffer from blurred edges of anatomical structures due to thermal diffusion effects. Therefore, it is difficult to obtain accurate animal information solely based on image data. Summary of the Invention

[0004] To address the existing technical problems, this application provides a method, imaging device, and medium for identifying the physiological characteristics of animals, which can improve the accuracy of animal information identification.

[0005] Firstly, a method for identifying physiological characteristics of animals is provided, applied to an imaging device; the method includes: Acquire image data; Identify the animal target in the image data, and the first physiological characteristics of the animal target; Obtain the three-dimensional size data of the animal target; Based on the three-dimensional size data and the first physiological characteristic, the physiological characteristics of the animal target are obtained.

[0006] In a second aspect, an imaging device is provided, including a memory, a processor, and an image data acquisition module and a ranging module connected to the processor; the image data acquisition module is used to acquire image data; the ranging module is used to measure the three-dimensional size data of the animal target; the memory stores a computer program; when the computer program is executed by the processor, it implements the method for identifying the physiological characteristics of animals as described in the above embodiments.

[0007] Thirdly, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the method for identifying the physiological characteristics of animals as described in the above embodiments.

[0008] Compared to schemes that rely solely on image information for identification, the method for identifying animal physiological characteristics provided in the above embodiments, when using an imaging device, further actively utilizes a ranging tool to measure the animal target and obtain its three-dimensional size data, in addition to identifying the animal target in the image data and its first physiological characteristics. Combined with the image-recognition-based first physiological characteristics and the three-dimensional size data, a more accurate physiological characteristic of the animal target is obtained. This method, by using a ranging tool to obtain the animal target's three-dimensional size data, provides a reliable physical scale basis. Furthermore, by combining the image-recognition-based first physiological characteristics with the ranging-obtained three-dimensional size data, it overcomes the inherent limitations of pure visual analysis and achieves information complementarity. The ranging-obtained three-dimensional size data can effectively assist and optimize the physiological characteristics obtained from the image data, significantly improving the accuracy of animal target physiological characteristic identification.

[0009] The imaging device and medium provided in the above embodiments belong to the same concept as the corresponding method embodiments for identifying the physiological characteristics of animals, and thus have the same technical effects as the corresponding method embodiments for identifying the physiological characteristics of animals, which will not be repeated here. Attached Figure Description

[0010] Figure 1 This is a structural block diagram of an imaging device in one embodiment.

[0011] Figure 2 This is a flowchart of a method for identifying the physiological characteristics of an animal in one embodiment.

[0012] Figure 3 This is a diagram illustrating the display effect of an imaging device displaying image data and physiological characteristics in one embodiment.

[0013] Figure 4 This is a diagram illustrating the display effect of an imaging device showing image data and physiological characteristics in another embodiment.

[0014] Figure 5 This is a schematic diagram of the structure of the recognition model in one embodiment.

[0015] Figure 6 This is a schematic diagram of the recognition process of the recognition model in one embodiment.

[0016] Figure 7 This is a flowchart of the steps for training a recognition model in one embodiment.

[0017] Figure 8This is a schematic diagram of key point detection and identification in one embodiment. Detailed Implementation

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] In the following description, the phrase "some embodiments" refers to a subset of all possible embodiments. It should be noted that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict.

[0021] In the following description, the terms "first, second, and third" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, and third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0022] In this embodiment, an imaging device is provided, such as... Figure 1 As shown, the imaging device includes a memory 101, a processor 102, an image data acquisition module 103 and a ranging module 104 connected to the processor.

[0023] The image data acquisition module 103 is used to acquire image data. The ranging module 104 is used to measure the actual distance to the measurement point of the animal target in order to obtain the three-dimensional size data of the animal target.

[0024] The image data acquisition module 103 sends the acquired image data to the processor 102, and the ranging module 104 sends the measured actual distance to the processor 102. The processor 102 implements a method for identifying the physiological characteristics of animals.

[0025] Specifically, the processor 102 acquires image data; identifies animal targets in the image data and the first physiological characteristics of the animal targets; acquires three-dimensional size data of the animal targets; and obtains the physiological characteristics of the animal targets based on the three-dimensional size data and the first physiological characteristics.

[0026] When in use, this imaging device, based on identifying animal targets in image data and their primary physiological characteristics, actively uses a ranging tool to measure the animal target, obtaining its three-dimensional dimensions. Combining this with the image-recognized primary physiological characteristics and the three-dimensional dimensions, a more accurate physiological profile of the animal target is obtained. By utilizing the ranging tool to obtain the three-dimensional dimensions of the animal target, this imaging device provides a reliable physical scale basis. Furthermore, by combining the image-recognized primary physiological characteristics with the ranging-obtained three-dimensional dimensions, it overcomes the inherent limitations of pure visual analysis, achieving information complementarity. The ranging-obtained three-dimensional dimensions effectively assist and optimize the physiological characteristics obtained from the image data, significantly improving the accuracy of animal target physiological characteristic recognition.

[0027] In one embodiment, the image data acquisition module of the imaging device is a visible light image acquisition module, so the image data is visible light image data. In another example, the image data acquisition module of the imaging device is an infrared image acquisition module, so the image data is a visible infrared image. In yet another example, the imaging device is equipped with a binocular camera (visible light image acquisition module and infrared image acquisition module), so the image data can be a fused image of infrared and visible light images, or infrared images can be used at night and visible light images can be used during the day.

[0028] In a preferred embodiment, the image data acquisition module of the imaging device is an infrared image acquisition module. The corresponding imaging device is an infrared imaging device. The infrared imaging device detects the infrared radiation (heat) naturally emitted by the surface of an object and converts it into a visualized infrared image representing the temperature distribution. The infrared imaging device enables non-contact measurement and does not rely on ambient light; it detects the heat emitted by the object itself, thus providing clear imaging in complete darkness, dense smoke, smog, or underground environments. This makes it an important tool for wildlife observation.

[0029] In one embodiment, the imaging device can be a handheld infrared thermal imager. Handheld infrared thermal imagers offer advantages in portability and concealment, making them uniquely advantageous for observing wildlife.

[0030] In one embodiment, the ranging module is a laser ranging module; the laser ranging module is parallel or coaxial with the optical axis of the image data acquisition module. During measurement, it is only necessary to align the aiming point of the imaging device (such as a thermal imager) with a specific part of the animal (such as the trunk or horn base). In this way, the laser will also hit the same position, and the measured distance is the actual distance from the laser ranging module to the corresponding position of the animal target, which is an accurate distance parameter for measuring the key morphology of the animal target.

[0031] In one embodiment, such as Figure 1As shown, the imaging device also includes a display screen 105 connected to the processor, the display screen 105 being configured to display image data and physiological characteristics of the animal target in the image data.

[0032] Therefore, in addition to displaying image data in real time, the imaging device's screen also simultaneously overlays physiological characteristic information identified and analyzed by the algorithm. This information, presented as a combination of graphical labels (such as anchor points and bounding boxes) and text information, is precisely associated with the animal target identified in the image. This enhances the user's observation experience, allowing them to intuitively obtain physiological characteristic information such as animal category, sex, and age while observing the animal target, greatly expanding the functional boundaries of traditional observation equipment.

[0033] Another aspect of this application provides a method for identifying the physiological characteristics of animals, applicable to, for example... Figure 1 Imaging devices, such as Figure 2 As shown, the method for identifying the physiological characteristics of an animal includes: Step 202: Obtain image data.

[0034] The type of image data in this embodiment is determined by the image data acquisition module configured in the imaging device. In one example, if the image data acquisition module is a visible light image acquisition module, the image data is visible light image data. In another example, if the image data acquisition module is an infrared image acquisition module, the image data is an infrared image. In yet another example, if the imaging device is equipped with a binocular camera (visible light image acquisition module and infrared image acquisition module), the image data can be a fused image of infrared and visible light images, or an image using infrared at night and visible light during the day.

[0035] Step 204: Identify the animal target in the image data and the first physiological characteristics of the animal target.

[0036] The first physiological characteristic refers to the inherent biological attributes of an animal target exhibited in its structure, function, and development during image recognition, which can be observed or measured. The physiological characteristics of an animal target include at least one or more of the following: animal category, age, sex, growth stage, and abnormal thermal signal information. In practical applications, the specific types of physiological characteristics of the animal target to be identified can be selected according to actual needs. For example, in one example, the physiological characteristics of the animal target include the animal target's sex and age.

[0037] In one embodiment, a pre-trained recognition model can be used to identify animal targets in image data. The pre-trained recognition model refers to a neural network model mounted on the imaging device that has been trained based on a training dataset. The training dataset includes a massive amount of labeled image data of different animal categories. The annotations include not only key points of the animal targets, such as head contours, trunk joints, limb joints, and secondary sexual characteristics, but also physiological feature labels of the animal targets, such as animal category labels, age labels, sex labels, growth stage labels, and thermal signal abnormality labels.

[0038] Therefore, the recognition model identifies animal targets in image data, including identifying the location, category, and key points of the animal targets in the image data, and then determines the first physiological characteristics of the animal targets based on the identified key points and extracted image features.

[0039] Understandably, the server can utilize newly collected infrared data of animals, such as new animal categories, animal target images from different angles, and animal target images in different environments, to iteratively train the original recognition model, resulting in a new version of the recognition model with stronger performance and better generalization ability. The imaging device connects via Wi-Fi, 4G / 5G, or other networks to retrieve the latest model file from the server and replace the old model stored locally, thus updating the recognition model. In this way, the imaging device's recognition model can be continuously optimized and iterated.

[0040] Step 206: Obtain the three-dimensional size data of the animal target.

[0041] Specifically, the imaging device is also equipped with a ranging module, which can combine the first physiological characteristics of the animal target output by the recognition model to measure the animal target and obtain the three-dimensional size data of the animal target, such as the body length, body height and other three-dimensional size data of the animal target.

[0042] Furthermore, the key morphology of the animal target is measured to obtain three-dimensional dimensional data of the key morphology.

[0043] In this context, key morphology refers to crucial physical parameters and / or locations that reflect the physiological characteristics of an animal target. In one example, key morphology includes, but is not limited to, the animal target's body size and sex dimorphism.

[0044] The three-dimensional dimensional data of an animal target can include parameters that reflect its size, such as limb length and trunk length. Among individuals of the same species, a larger body size usually indicates older age or greater maturity.

[0045] Sexual dimorphism in animals refers to the differences between different sexes of the same species. Sexual dimorphism can include body proportions, secondary sexual characteristics, etc. Generally, cubs typically have a specific "large head, small body" proportion, which changes as they grow. Furthermore, males and females may differ in the proportions of different body parts across different animal species.

[0046] Different animal species possess different body proportions that reflect age differences and / or sexual dimorphism. For example, some animals can be sexed by the ratio of shoulder width to hip width. In some birds, the sex can be determined by the ratio of tail feather length to body length. For instance, male pheasants have extremely long tail feathers, possibly equal to or even longer than their bodies, while females have very short tail feathers. A tail length / body length ratio close to or greater than 1 indicates a male pheasant.

[0047] Secondary sexual characteristics refer to morphological features related to sex, which can be used to distinguish the sex of animals or assess their sexual development level. Different animals have different secondary sexual characteristics. For example, the secondary sexual characteristic of deer is antlers (male deer have antlers, female deer do not). The secondary sexual characteristic of peacocks is their tail feathers (the male peacock's tail feathers are fan-shaped when spread, and the feathers have eye-like markings; the female peacock's tail feathers are shorter). The secondary sexual characteristic of lions is their mane (male lions have manes, female lions do not).

[0048] Therefore, the key morphologies that need to be measured differ for different animals. For example, the key morphologies for deer include body length, body height, and antler height. For peacocks, key morphologies include body length, wing length, and tail feather length. For lions, key morphologies include body length, body height, body proportions, and mane length.

[0049] In one embodiment, based on the animal target, the measurement points corresponding to the key morphology to be measured are determined, and then the ranging module of the imaging device is controlled to measure the actual distance to the measurement point of the key morphology. Combining the field of view of the imaging device itself and the pixel coordinates of the measurement points, the actual distance between the measurement points is estimated by the principle of triangulation.

[0050] Step 208: Based on the three-dimensional size data and the first physiological characteristics, obtain the physiological characteristics of the animal target.

[0051] The three-dimensional size data in this embodiment can provide a reliable physical scale basis. By combining the three-dimensional size data with the first physiological characteristics, a more accurate physiological characteristic of the animal target can be obtained.

[0052] Compared to schemes that rely solely on image information for identification, this application, when using imaging equipment, not only identifies the animal target in the image data and its primary physiological characteristics, but also actively uses a ranging tool to measure the animal target, obtaining its three-dimensional size data. Combining this with the image-recognition-based primary physiological characteristics and the three-dimensional size data yields a more accurate physiological characteristic of the animal target. This method, by using a ranging tool to obtain the animal target's three-dimensional size data, provides a reliable physical scale basis. Furthermore, by combining the image-recognition-based primary physiological characteristics with the ranging-obtained three-dimensional size data, it overcomes the inherent limitations of pure visual analysis, achieving information complementarity. The ranging-obtained three-dimensional size data effectively assists and optimizes the physiological characteristics obtained from the image data, significantly improving the accuracy of animal target physiological characteristic identification.

[0053] In one embodiment, obtaining the physiological characteristics of an animal target based on three-dimensional size data and a first physiological characteristic includes: obtaining a second physiological characteristic of the animal target based on the three-dimensional size data; and modifying the first physiological characteristic of the animal target based on the second physiological characteristic to obtain the physiological characteristics of the animal target.

[0054] In one embodiment, three-dimensional size data provides a reliable physical scale basis, effectively assisting and optimizing preliminary physiological characteristics obtained solely from image appearance. For example, based on three-dimensional size data, if an animal target is identified as male by obvious secondary sexual characteristics (e.g., the animal target is a deer, and antlers are identified), but the first physiological characteristic based on image recognition predicts the animal target as female, then the actual measured three-dimensional size data should be used to identify the animal target as male. As another example, if the animal target is identified as a juvenile deer approximately 3 years old based on body size data from the three-dimensional size data of a key morphology, but the first physiological characteristic based on image recognition predicts the deer as a sub-adult 2 years old with low confidence, then the three-dimensional size data of the key morphology and the recognition results of the recognition model can be combined for fusion correction.

[0055] In one embodiment, key morphologies of different categories can be used to distinguish different physiological characteristics. For example, the sexual dimorphism of an animal can distinguish its sex. As another example, different categories of animals have different body proportions that can reflect age differences and / or sex differences.

[0056] This method improves the accuracy of physiological feature recognition of animal targets by correcting and optimizing the first physiological features based on the initial image identification using three-dimensional size data.

[0057] In one embodiment, the physiological characteristics include at least one or more of the following: category, age, sex, growth stage, and abnormal thermal signal.

[0058] Specifically, the specific types of physiological features that the recognition model needs to identify depend on the labeled content of the model's training data. The labeled content includes not only key points of the animal target, such as head outline, trunk joints, limb joints, and secondary sexual characteristics, but also physiological feature labels of the animal target, such as animal category labels, age labels, sex labels, growth stage labels, and abnormal thermal signal labels.

[0059] Accordingly, the trained recognition model can perform multi-task analysis on animal targets in images, including target localization, species classification, and keypoint detection. Finally, the model integrates morphological indicators derived from keypoints and other image features to predict and output one or more of the aforementioned physiological characteristics.

[0060] That is, the physiological characteristics of the animal target corresponding to the label, including at least one or more of the following: animal target category, age, sex, growth stage, and abnormal thermal signal information.

[0061] The category labels in the training data of the identification model can be broad categories at the family level (such as elephant family) or specific to species or subspecies (such as Asian elephant, African elephant). Thus, the category of the animal target predicted by the identification model can be a broad category at the family level or a species or subspecies.

[0062] Thermal signal analysis is a unique advantage of infrared imaging equipment. Infrared imaging devices detect the infrared energy radiated from the surface of an animal, convert it into electrical signals, and generate thermal distribution images, enabling visual analysis of the target's temperature field. By analyzing the temperature of the animal target in the infrared image, abnormally high temperatures (inflammation, infection) or abnormally low temperatures (poor blood circulation, necrosis) in localized areas of the animal target can be detected. Thermal signal anomalies in the training data of the identification model can include both abnormally high and low temperatures. Therefore, the identification model can predict whether a moving target has temperature anomalies and, if so, the type of anomaly, based on image data.

[0063] In this embodiment, various physiological characteristics of animal targets can be identified, providing users with rich observation information about animals when using imaging equipment to observe animal targets.

[0064] In one embodiment, the imaging device further includes a display screen, and the method for identifying the physiological characteristics of an animal further includes: displaying image data and the physiological characteristics of the animal target in the image data on the display screen of the imaging device.

[0065] Specifically, if an animal target is detected in the image data, a rectangular detection box is drawn on the original image based on the bounding box coordinates of the animal target output by the recognition model to mark the position of the animal target in the original image. Further, such as... Figure 3As shown, the physiological characteristics of the animal target are displayed above the detection box. Alternatively, as shown... Figure 4 As shown, the display screen is divided into two areas: a larger area for displaying image data and a smaller area for displaying the physiological characteristics of the animal target.

[0066] Therefore, in addition to displaying image data in real time, the imaging device's screen also simultaneously overlays physiological characteristic information identified and analyzed by the algorithm. This information, presented as a combination of graphical labels (such as anchor points and bounding boxes) and text information, is precisely associated with the identified animal target in the image. This enhances the user's observation experience, allowing them to intuitively obtain physiological characteristic information such as animal category, sex, and age while observing the animal target, greatly expanding the functional boundaries of traditional observation equipment.

[0067] In one embodiment, obtaining the second physiological characteristics of an animal target based on three-dimensional size data includes: obtaining a mapping table between the physiological characteristics of the animal target and the three-dimensional size data; and querying the mapping table based on the three-dimensional size data to obtain the second physiological characteristics of the animal target.

[0068] Specifically, the imaging device's memory is also used to store mapping tables between the physiological characteristics and three-dimensional dimensional data of different categories of animal targets. The mapping table records the physiological characteristics corresponding to the three-dimensional dimensional data of the animal target of that category, such as the physiological characteristics corresponding to the body length, body width, and other dimensional data of that category.

[0069] In one embodiment, the imaging device's memory is further used to store a mapping table between the physiological characteristics and key morphological three-dimensional dimensions of different categories of animal targets. The mapping table records the physiological characteristics corresponding to the key morphological three-dimensional dimensions of the animal target of that category.

[0070] In one embodiment, the physiological characteristics recorded in the mapping table are partial and strongly correlated with the three-dimensional dimensional data of key morphologies, and do not need to include all physiological characteristics. For example, physiological characteristics such as abnormal thermal signals are not strongly correlated with key body shape measurement data, while age, sex, and growth stage can be reflected in the three-dimensional dimensional data of body shape and the three-dimensional dimensional data of sex dimorphism. Therefore, the mapping table can record the mapping relationship between the specific three-dimensional dimensional data of key morphologies (such as body length, body height, shoulder width to hip width ratio, etc.) and age, sex, and growth stage, so as to verify and correct the physiological characteristics initially identified based on image data.

[0071] Depending on the characteristics of different animal targets, the key morphologies recorded in the relational mapping table also vary. For example, the key morphologies of a deer include body length, body height, and antler height. As another example, the key morphologies of a peacock include body length, wing length, and tail feather length.

[0072] Specifically, based on the category of the animal target in the image identified from the image data, a mapping table between the physiological characteristics and the three-dimensional dimensions of the key morphology of that animal category is retrieved from memory. Based on the measured three-dimensional dimensions of the key morphology, the mapping table is consulted to obtain the second physiological characteristic of the animal target. The second physiological characteristic is compared with the first physiological characteristic; if they differ, the first physiological characteristic is corrected using the second physiological characteristic.

[0073] The second physiological feature is derived from the three-dimensional dimensional data of the key morphology of the animal target, providing a reliable physical scale basis. Therefore, utilizing the second physiological feature can optimize the first physiological feature obtained from image data, significantly improving the accuracy of animal target physiological feature recognition.

[0074] In one embodiment, different categories of key morphologies can be used to distinguish different physiological characteristics. For example, the sexual dimorphism of an animal can distinguish its sex.

[0075] Sexual dimorphism in animals refers to the differences between different sexes of the same species. This includes body proportions, secondary sexual characteristics, and body size. Generally, young animals typically have a "large head, small body" proportion, which changes as they grow. Furthermore, the proportions of different body parts can vary between males and females in different animal species.

[0076] Different animal species possess different body proportions that reflect age differences and / or sexual dimorphism. For example, some animals can be sexed by the ratio of shoulder width to hip width. In some birds, the sex can be determined by the ratio of tail feather length to body length. For instance, male pheasants have extremely long tail feathers, possibly equal to or even longer than their bodies, while females have very short tail feathers. A tail length / body length ratio close to or greater than 1 indicates a male pheasant.

[0077] Secondary sexual characteristics refer to morphological features related to sex, which can be used to distinguish the sex of animals or assess their sexual development level. Different animals have different secondary sexual characteristics. For example, the secondary sexual characteristic of deer is antlers (male deer have antlers, female deer do not). The secondary sexual characteristic of peacocks is their tail feathers (the male peacock's tail feathers are fan-shaped when spread, and the feathers have eye-like markings; the female peacock's tail feathers are shorter). The secondary sexual characteristic of lions is their mane (male lions have manes, female lions do not).

[0078] Therefore, the mapping table of animal targets can include the mapping relationship between the sex of the animal target and the three-dimensional size data of the sex dimorphism.

[0079] Different animal targets exhibit different sexual dimorphism characteristics. In general, the mapping table for animal targets includes the mapping relationship between the animal target's sex and at least one of the following three-dimensional dimensional data: body proportion, secondary sexual characteristics, and body size. Specifically, the mapping table may include the mapping relationship between the animal target's sex and at least one of the following: measurement data of secondary sexual characteristics, measurement data of body proportion of specific body parts, and body size measurement data.

[0080] Taking deer as an example, male and female deer can be distinguished by measuring secondary sexual characteristics and body proportions of specific parts (shoulder width and hip width).

[0081] In one embodiment, the key features include: sex dimorphism; the mapping table of animal targets includes the sex of the animal target and the mapping relationship between the three-dimensional size data of sex dimorphism.

[0082] Based on the three-dimensional size data of key morphology, the mapping table is queried to obtain the second physiological characteristics of the animal target, including: based on the three-dimensional size data of sex dimorphism, the mapping table is queried to obtain the second sex of the animal target.

[0083] Correspondingly, based on the second physiological characteristic, the first physiological characteristic of the animal target is modified to obtain the physiological characteristics of the animal target, including: in the case where the first sex and the second sex are inconsistent, the sex of the animal target is determined based on the second sex.

[0084] The secondary physiological characteristic is derived from the three-dimensional dimensional data of the key morphology of the animal target, which has a reliable physical scale basis and therefore a higher confidence level. Therefore, the sex of the animal target is determined based on the secondary sex identified by the three-dimensional dimensional data of the key morphology.

[0085] In one embodiment, a mapping table records the mapping relationship between secondary sexual characteristics (such as horns, mane, crest, etc.), body proportions, and sex. Therefore, by querying the mapping table based on the measurement data of secondary sexual characteristics and body proportions, the secondary sex of the animal target can be determined. Thus, by using secondary sexual characteristics in combination with body proportions to infer the sex of this type of animal target, the accuracy of animal target sex identification can be improved.

[0086] Among the characteristics of sexual dimorphism, the confidence level of secondary sexual characteristics is higher than that of body proportion. In one embodiment, the determination logic for the secondary sex of the animal target is as follows: based on the measurement data of secondary sexual characteristics and the measurement data of body proportion, a mapping table is consulted. Based on the measured three-dimensional dimensional data of sexual dimorphism, a pre-defined mapping relationship between secondary sexual characteristics, body proportions, and sex is queried to determine the secondary sex of the animal target. Specific determination strategies include: First, when the mapping table clearly defines that secondary sexual characteristics can uniquely determine sex, the sex result is output directly based on the measurement data of that characteristic. For example, if an adult male deer has typical forked antlers while a female deer has no antlers, then detecting complete antlers indicates that it is male.

[0087] Secondly, when a single secondary sexual characteristic measurement corresponds to multiple possible sexes (especially in individuals of different ages or physiological stages), it is necessary to further combine body proportion data for comprehensive judgment. For example, if the mapping table indicates that "having antlers" usually indicates a male, but juvenile male deer also do not have antlers, then it is necessary to further analyze body shape parameters: if the target simultaneously meets the conditions of "no antlers" and "body proportion exceeding the adult female threshold," then it is determined to be an adult female; conversely, if "no antlers" and "body proportion less than the adult threshold," then it is determined to be a juvenile individual.

[0088] This strategy significantly improves the accuracy and robustness of gender identification through a hierarchical decision-making mechanism, and is particularly suitable for complex scenarios with confounding factors such as age and developmental stage.

[0089] In one embodiment, age is used as an example of a physiological characteristic, and an animal's age is usually related to its body size.

[0090] The three-dimensional dimensional data of the animal target's body shape can include parameters that reflect its size, such as limb length and trunk length. Among individuals of the same type, a larger body shape usually indicates older age or greater maturity.

[0091] The mapping table for animal targets includes the mapping relationship between the age and three-dimensional size data of the animal targets. Taking the deer as an example, the mapping table is shown in Table 1.

[0092] Table 1 Mapping Relationships for Deer In the table above, body length and body height are the three-dimensional dimensions of the deer's body.

[0093] In one embodiment, the second physiological characteristic of the animal target is obtained by querying a mapping table based on the three-dimensional size data of the key morphology, including: obtaining the second age of the animal target by querying the mapping table based on the three-dimensional size data of the body shape. For example, if the animal target in the measurement image is a deer, and its body length is 60 cm and its body height is 50 cm, then the age identified based on the three-dimensional size data of the key morphology is 0-1 years old.

[0094] Correspondingly, based on the second physiological characteristic, the first physiological characteristic of the animal target is modified to obtain the physiological characteristics of the animal target, including: when the first age and the second age are inconsistent, the second age and the first age are merged to obtain the age of the animal target.

[0095] The specific fusion strategy can be weighted or averaged.

[0096] In one example, the age of the animal target is obtained by fusing the second age and the first age, including: determining the weights of the first age and the second age based on the confidence level of the first age; calculating the weighted sum of the first age and the second age based on the weights; and determining the age of the animal target based on the weighted sum of the ages.

[0097] Specifically, Age = α Age1 + (1 - α) Age 2.

[0098] Where Age represents the predicted final age of the animal target, Age1 represents the first age, Age2 represents the second age, α is the weight of the first age, and 1-α is the weight of the second age.

[0099] The weights of the first and second ages are dynamically determined based on the confidence level of the first age. Specifically, the confidence level of the first age is positively correlated with its weight, while the confidence level of the first age is negatively correlated with its weight. For example, if the confidence level of the first age is high, its weight α can be set to a larger value; conversely, if the confidence level of the first age is low, its weight α can be lowered, and correspondingly, the weight 1-α of the second age can be increased.

[0100] When the first age predicted based on image data has a high confidence level, it indicates that the image quality is high and the features are obvious (such as standard animal posture and good angle), and the prediction result itself is very reliable. At this time, assigning higher weight to the first age predicted by the model can give full play to the advantages of deep learning models in capturing complex patterns and obtain a more refined age estimate.

[0101] When the model's predicted first age has low confidence, it usually means that the image is challenging (e.g., occlusion, blurring, unusual pose, rare individuals). In this case, the model's prediction is unreliable. By reducing its weight and relying more on the relatively stable and reliable second age derived from a lookup table based on the key morphological 3D size data, the model can be effectively prevented from making incorrect predictions, and the age prediction results can be brought back to a reasonable range.

[0102] In one example, if the confidence level of the first age output by the recognition model is >0.8, then the weight α of the first age can be 0.7, and the weight of the second age can be 0.3. However, if the confidence level of the first age output by the recognition model is ≤0.8, then the weight α of the first age can be 0.4, and the weight of the second age can be 0.6.

[0103] In this embodiment, the weights of the first and second ages are dynamically determined based on the confidence level of the first age output by the recognition model, and then the weighted sum of the ages is calculated to determine the age of the animal target. This allows the entire system to maintain the powerful capabilities of the deep learning model while relying on the stability and reliability of the three-dimensional size data of the key morphology, thus significantly reducing the error rate of the age output results.

[0104] In one embodiment, acquiring the three-dimensional size data of the animal target includes: acquiring the three-dimensional size data of the animal target measured by a laser ranging module.

[0105] Specifically, the laser ranging module is controlled to measure the animal target and obtain its three-dimensional dimensions. More specifically, the laser ranging module is controlled to measure the key features of the animal target and obtain its three-dimensional dimensions.

[0106] The laser ranging module and the image data acquisition module have parallel or coaxial optical axes. During measurement, it is only necessary to align the aiming point of the imaging device (such as a thermal imager) with a specific part of the animal (such as the trunk or base of the horns). In this way, the laser will also hit the same position, and the measured distance is the actual distance from the laser ranging module to the corresponding position of the animal target, providing accurate distance parameters for measuring the key morphology of the animal target.

[0107] In one embodiment, acquiring the three-dimensional size data of an animal target measured by a laser ranging module includes: determining two measurement points of the animal target based on the animal target and its key points; calculating the pixel coordinate difference between the two measurement points on the image, and determining the angle difference between the two measurement points based on the pixel coordinate difference; acquiring the actual distance from the laser ranging module to any one of the measurement points; and determining the three-dimensional size data based on the angle difference and the actual distance.

[0108] Specifically, based on the animal target and its key points, two measurement points are determined for each key morphology of the animal target; for each key morphology, the pixel coordinate difference between the two measurement points on the image is calculated, and the angle difference between the two measurement points is determined based on the pixel coordinate difference; the ranging module is controlled to measure the actual distance to any measurement point of the key morphology; and the key morphology measurement parameters are determined based on the angle difference and the actual distance.

[0109] The imaging device pre-stores categories of key morphological features for each animal target. Based on the identified animal target, its outline, and key points, it determines two measurement points for each key morphology. For example, the measurement points for a deer's body length include the tip of the nose and the base of the tail, specifically the length from the tip of the nose to the base of the tail. The measurement points for a deer's body height include the distance from the shoulder to the foot, specifically the vertical height from the shoulder to the foot. The measurement points for a deer's antler height include the base of the antler and the tip, specifically the height from the base of the antler to the tip.

[0110] Once the key morphological measurement points of the animal target are identified, the pixel coordinate difference (Δy_pixels) between the two measurement points on the infrared image can be determined. The aiming point of the ranging module is moved to any measurement point of the key morphology (e.g., the base of the antlers), and ranging is triggered to obtain the precise distance (D) from the ranging module to the measurement point of the key morphology. The infrared thermal imager itself has a fixed field of view (FOV) and resolution. This means that the actual angle corresponding to each pixel is known (e.g., 0.1 milliradians / pixel).

[0111] Therefore, the distance between two measurement points = distance (D) × tan(angle difference (Δθ)). Where, angle difference Δθ = Δy_pixels × the angular resolution of a single pixel. For example, antler height (H) = distance (D) × tan(angle difference (Δθ)).

[0112] In this embodiment, the actual distance to the measurement points of the key shape is measured using a laser ranging module. The actual distance between the two points of the key shape can be calculated using the angle difference between the two measurement points and trigonometric functions. This method is extremely fast and can output results in real time.

[0113] In one embodiment, an infrared imaging device detects the infrared radiation (heat) naturally emitted from an object's surface and converts it into a visualized infrared image representing the temperature distribution. Infrared imaging devices enable non-contact measurement and do not rely on ambient light; they detect the heat emitted by the object itself, thus providing clear imaging in complete darkness, dense smoke, smog, or underground environments. This makes it an important tool for wildlife detection. However, its imaging mechanism, based on surface thermal radiation, results in inherent defects in infrared images, such as severe loss of texture information, dynamic range compression, and blurred anatomical structures. This leads to a very low pixel percentage for key features for identifying wildlife physiological characteristics (such as antler buds in male deer and body proportions in juveniles), far lower than the representational capabilities of visible light images. Consequently, the accuracy of identifying the physiological characteristics of animal targets is low.

[0114] To address this issue, this application further optimizes the infrared image recognition model to effectively overcome the problem of low image recognition accuracy caused by missing textures and blurred anatomical structures in infrared images.

[0115] Specifically, such as Figure 5 As shown, the recognition model includes a backbone network 501, a pose normalization network 502, a convolutional network 503, and a regression network 504. like Figure 6As shown, identifying animal targets in image data and their first physiological characteristics includes: inputting image data into a pre-trained recognition model; identifying preliminary detection key points in the infrared image through a backbone network; cropping a key point region image using a pose normalization network with the preliminary detection key points as a reference; obtaining precisely identified key points from the key point region image and the infrared image through a convolutional network; and identifying the pose information constituted by the precisely identified key points through a regression network to obtain the animal target in the infrared image and its first physiological characteristics.

[0116] The backbone network 501 is typically a CNN (Convolutional Neural Network) for feature extraction. Its input is the entire infrared image. The backbone network 501 extracts features from the global infrared image and generates preliminary, low-resolution keypoint heatmaps. The peak position of each heatmap corresponds to the coarse coordinates of the corresponding keypoint, which is the initial detection of keypoints.

[0117] After the backbone network 501 outputs the preliminary detected keypoints, the pose normalization network 502 calculates the affine transformation matrix for each keypoint based on its coarse coordinates. It then performs cropping and alignment operations on regions including the animal's head and torso, which contain the preliminary detected keypoints, from the original image or intermediate feature maps of the backbone network, generating standardized regions of interest (ROIs) to facilitate stable extraction of thermodynamic features from the infrared image. The output of the pose normalization network 502 is a fixed-size image patch centered on the keypoint.

[0118] The convolutional network 503, specifically a small CNN network with shared weights, inputs the key point region images and infrared images into the convolutional network 503 respectively. Based on the features of the key point region images and the features of the infrared images, the key points are accurately identified, thus achieving fine-grained classification of the key points. Its output is a small, high-resolution heatmap, and its peak position corresponds to the precise position of the key point within the current image block.

[0119] Regression Network 504 identifies the species and primary physiological characteristics of animal targets based on the pose information composed of the coordinates of all precise key points.

[0120] The aforementioned recognition model combines high-precision keypoint detection technology with infrared thermal imaging analysis. By locating anatomical key points such as the animal's head, trunk joints, and sex-specific regions (e.g., antler buds in male deer), it constructs a "region of interest focusing" mechanism. This mechanism effectively addresses the inherent defects of infrared images, such as missing textures and blurred anatomical structures. Compared to traditional target detection models, it improves the recognition accuracy of sex characteristics in juveniles by more than 40%. A pose normalization module forms a unified feature representation space, ensuring the stability of feature extraction and laying the foundation for fine-grained classification.

[0121] like Figure 5 As shown, the backbone network 501 includes a first feature extraction branch 5011, a second feature extraction branch 5012, a third feature extraction branch 5013, a feature fusion network 5014, and a key point detection network 5015. The backbone network 501 identifies preliminary key points in an infrared image, including: inputting the infrared image into the first feature extraction branch 5011, the second feature extraction branch 5012, and the third feature extraction branch 5013 respectively; extracting low-frequency thermal distribution features from the infrared image using the first feature extraction branch 5011; extracting high-frequency edge structure features from the infrared image using the second feature extraction branch 5012; extracting features from the infrared image using the third feature extraction branch 5013 under spatial attention and channel attention to obtain attention features of the infrared image; fusing the low-frequency thermal distribution features, high-frequency edge structure features, and attention features in the feature fusion network 5014 to obtain fused features; and identifying the fused features using the key point detection network 5015 to obtain preliminary key points in the infrared image.

[0122] Specifically, the first feature extraction branch 5011 and the second feature extraction branch 5012 perform wavelet transform processing on the infrared image. The first feature extraction branch 5011 extracts low-frequency thermal distribution features from the infrared image through wavelet transform, which helps in locating the overall area of ​​the animal. The second feature extraction branch 5012 extracts high-frequency edge features from the infrared image through wavelet transform, which helps in locating keypoint contours in the infrared image. Through the third feature extraction branch 5013, an attention module combining spatial and channel attention is introduced to adaptively weight the infrared image features, thereby highlighting the animal target area, suppressing background interference, and extracting attention features focused on the animal itself. The fusion network 5014 concatenates or adds the low-frequency thermal distribution features, high-frequency edge structure features, and attention features to obtain complementary fusion features. The keypoint detection network 5015 includes convolutional layers and upsampling layers, processes the fusion features, and outputs a preliminary keypoint heatmap. The peak positions in the heatmap are the initially detected keypoints.

[0123] The aforementioned backbone network structure fully considers the characteristics of infrared images. It captures features of different properties through a multi-branch structure, and then optimizes the model using attention mechanisms and fusion strategies to achieve high-precision preliminary prediction of key points. Specifically, guided by an attention mechanism, spatial and channel attention are added during the feature extraction stage, enabling the model to focus on key point regions even in complex backgrounds. Wavelet transforms are used within the network to separate the low-frequency thermal distribution from the high-frequency edge structure of the infrared image, which are then fused back into the backbone network. High-frequency information enhances boundary detection, while low-frequency information preserves the overall thermal distribution, allowing key points to be stably located even in blurred scenes.

[0124] By using a recognition model that includes the aforementioned backbone network, and by introducing attention mechanisms, frequency domain enhancement, and pose normalization strategies, the model can still accurately extract information even when gender features occupy only a very small number of pixels in infrared images.

[0125] In one embodiment, such as Figure 7 As shown, the steps for training the recognition model include: Step 702: Train a first model based on the first set of labeled infrared images. The first model is used to identify animal targets in the infrared image data and the physiological characteristics of the animal targets. The labeled data of the infrared images in the infrared image training set includes the categories, key points and physiological feature labels of the animal targets.

[0126] This process involves using infrared imaging equipment (such as thermal imagers) to acquire infrared images of animals, covering various natural scenes to ensure data diversity. After data acquisition, manual annotation is performed. For example... Figure 8 As shown, key points such as the animal's head outline, trunk joints, limb joints, and antler buds were marked, and physiological characteristics (such as age and sex labels) were annotated for each infrared image. Next, the infrared images were normalized to uniform size, with pixel values ​​normalized to the [0,1] or [-1,1] range. Finally, the infrared image data was divided into training, validation, and test sets in an 8:1:1 ratio, providing a high-quality and reasonably allocated foundation of infrared image data for subsequent model training.

[0127] The first model is relatively large. Preferably, the first model can be a pre-trained model. Then, the pre-trained weights of the first model are loaded, and the first model is fine-tuned based on the labeled first infrared image training set. During training, the initial learning rate is set to 0.001, and the learning rate decreases by 0.1 every 10 epochs. Training continues until the benchmark accuracy on the validation set reaches more than 90%, making it a "knowledge source" for knowledge distillation.

[0128] Step 704: Use the first model as the teacher model and the second model as the student model; the second model is smaller in size than the first model.

[0129] The first model is typically a complex, large model that can achieve high-precision discrimination through training. The second model is a lightweight model. Although the first model has superior performance, it has high hardware requirements and consumes a lot of resources, making it difficult to deploy on resource-constrained imaging equipment to complete complex tasks. In this embodiment, the first model is used as the teacher model, and the lightweight second model is used as the student model. Through knowledge distillation, a teacher-student network transfer learning framework is used to learn multimodal fusion features from the teacher network and transfer the knowledge to the second model.

[0130] In one embodiment, the second model can be a pre-trained model. Then, the pre-trained weights of the second model are loaded, and the second model is fine-tuned through knowledge distillation, which can improve training efficiency.

[0131] Step 706: Freeze the parameters of the teacher model, and input the second infrared image training set labeled in the same batch into the teacher model and the student model respectively to obtain the soft label recognized by the teacher model, and the hard label and soft label recognized by the student model.

[0132] During training, the weights of the teacher model are frozen, and the student model is forward-propagated on the same input data. The student model makes two predictions, resulting in two outputs. One is a hard prediction, which yields hard labels. Hard predictions are the output of softmaxt at normal temperatures; hard labels provide explicit category information but lack the correlation between categories. The other is a soft prediction, which uses the softmaxt output at the same high temperature as the teacher model, yielding soft labels. Soft labels are probability distributions reflecting the confidence level of each category.

[0133] Step 708: Calculate the cross-entropy loss based on the difference between the hard labels of the student model and the true labels of the training set; calculate the KL divergence based on the difference between the soft labels of the student model and the soft labels of the teacher model; and obtain the total loss by weighting the cross-entropy loss and the KL divergence.

[0134] Cross-entropy loss is used to measure the difference between the student model's predictions and the actual labeled results.

[0135] KL divergence is used to measure the difference between the soft predictions of the student model and the soft predictions of the teacher model.

[0136] By fusing the cross-entropy loss and the KL divergence loss, the total loss is obtained, which is a weighted sum of the two and can be expressed as: in, Cross-entropy loss is used for supervised class prediction. The KL divergence loss is used to align the output distributions of the teacher and student models.

[0137] Step 710: Backpropagate and update the weights of the student model based on the total loss.

[0138] Step 712: Complete the training through multiple iterations and deploy the trained student model as the recognition model.

[0139] Specifically, the student network was trained on a GPU using an infrared image training set for 150 epochs. The optimizer was Adam, with a learning rate of 0.0001, decaying by 0.1 every 10 epochs, and a batch size of 16. During training, the validation set accuracy was monitored in real time. If the accuracy did not improve for 5 consecutive epochs (patience = 5), an early stopping mechanism was triggered. After training, the model's performance was validated using a test set, ensuring that the gender classification accuracy exceeded 95%, the misclassification rate for juveniles was below 5%, and the age estimation error was less than 3%, thus achieving the accuracy target.

[0140] In this embodiment, the teacher network possesses high-precision discrimination capabilities on large-scale data, but its computational overhead is high, making it unsuitable for real-time field monitoring equipment. This application utilizes knowledge distillation to transfer the knowledge of the teacher network to a lightweight student model, enabling low-power, high-speed inference on embedded hardware. In infrared thermal imaging, gender-related features account for a very small percentage of pixels in the image, and directly training the lightweight model easily leads to the loss of this fine-grained information. This invention introduces KL divergence constraints into the loss function, allowing the student network to acquire the "implicit knowledge" of the teacher model regarding weak gender features during the learning process, significantly improving the discrimination accuracy and robustness of the lightweight model. By fusing cross-entropy loss and KL divergence loss, the misclassification rate of the lightweight student network model in complex backgrounds is effectively reduced, and the model's stability in multiple scenarios is improved.

[0141] In one embodiment, the method for identifying the physiological characteristics of an animal includes a training phase and an application phase.

[0142] The training phase includes the following steps: Step 1: Data preprocessing.

[0143] Specifically, infrared images of animals were acquired using thermal imagers, covering various natural scenes to ensure data diversity. After data acquisition, manual annotation was performed, marking key points such as the animal's head outline, trunk joints, limb joints, and antler buds of male deer, and labeling each infrared image with age and sex. Next, the infrared images were normalized to standardize their size, and pixel values ​​were normalized to the [0,1] or [-1,1] range. Finally, the infrared image data was divided into training, validation, and test sets in an 8:1:1 ratio, providing a high-quality, reasonably allocated foundation of infrared image data for subsequent model training.

[0144] Step 2: Train the recognition model.

[0145] First, the teacher model is trained. A first model is trained based on the first set of labeled infrared images, and this first model is used as the teacher model.

[0146] Secondly, student models are trained based on knowledge distillation from teacher models.

[0147] Specifically, the second model is used as the student model; the second model is smaller than the first model. The parameters of the teacher model are frozen. The second infrared image training set labeled in the same batch is input into the teacher model and the student model respectively to obtain the soft labels recognized by the teacher model, and the hard labels and soft labels recognized by the student model. Based on the difference between the hard labels of the student model and the true labels of the training set, the cross-entropy loss is calculated. Based on the difference between the soft labels of the student model and the soft labels of the teacher model, the KL divergence is calculated. The total loss is obtained by weighted sum of the cross-entropy loss and the KL divergence. The weights of the student model are backpropagated and updated based on the total loss. The training is completed through multiple iterations, and the trained student model is deployed as the recognition model.

[0148] During training, the weights of the teacher model are frozen, and the student model is forward-propagated on the same input data. The student model makes two predictions, resulting in two outputs. One is a hard prediction, which yields hard labels. Hard predictions are the output of softmaxt at normal temperatures; hard labels provide explicit category information but lack the correlation between categories. The other is a soft prediction, which uses the softmaxt output at the same high temperature as the teacher model, yielding soft labels. Soft labels are probability distributions reflecting the confidence level of each category.

[0149] Cross-entropy loss is used to measure the difference between the student model's predictions and the actual labeled results.

[0150] KL divergence is used to measure the difference between the soft predictions of the student model and the soft predictions of the teacher model.

[0151] By fusing the cross-entropy loss and the KL divergence loss, the total loss is obtained, which is a weighted sum of the two and can be expressed as: in, Cross-entropy loss is used for supervised class prediction. The KL divergence loss is used to align the output distributions of the teacher and student models.

[0152] Teacher networks possess high-precision discrimination capabilities on large-scale data, but their high computational cost makes them unsuitable for real-time field monitoring equipment. This application utilizes knowledge distillation to transfer the knowledge of the teacher network to a lightweight student model, enabling low-power, high-speed inference on embedded hardware. In infrared thermal imaging, gender-related features account for a very small percentage of pixels in the image, and directly training the lightweight model easily leads to the loss of this fine-grained information. This invention introduces KL divergence constraints into the loss function, allowing the student network to acquire the "implicit knowledge" of the teacher model regarding weak gender features during the learning process, significantly improving the discrimination accuracy and robustness of the lightweight model. By fusing cross-entropy loss and KL divergence loss, the misclassification rate of the lightweight student network model in complex backgrounds is effectively reduced, and the model's stability in multiple scenarios is improved.

[0153] The application phase includes: Acquire image data; process the image data using a pre-trained recognition model to identify animal targets in the image data and their primary physiological characteristics; measure the key morphology of the animal targets to obtain their three-dimensional dimensions; and based on the three-dimensional dimensions of the key morphology, correct the primary physiological characteristics of the animal targets to obtain their physiological characteristics.

[0154] In order to make the method applicable to infrared imaging equipment, the structure of the recognition model has also been improved.

[0155] When processing image data using a pre-trained recognition model to identify animal targets and their primary physiological characteristics, the model first identifies preliminary key points in the infrared image through its backbone network. Then, based on these preliminary key points, the model's pose normalization network crops the key point region image. Finally, the model's convolutional network identifies both the key point region image and the infrared image to obtain precisely identified key points. Finally, the model's regression network identifies the pose information formed by the precisely identified key points to obtain the animal target and its primary physiological characteristics in the infrared image.

[0156] The aforementioned recognition model combines high-precision keypoint detection technology with infrared thermal imaging analysis. By locating anatomical key points such as the animal's head, trunk joints, and sex-specific regions (e.g., antler buds in male deer), it constructs a "region of interest focusing" mechanism. This mechanism effectively addresses the inherent defects of infrared images, such as missing textures and blurred anatomical structures. Compared to traditional target detection models, it improves the recognition accuracy of sex characteristics in juveniles by more than 40%. A pose normalization module forms a unified feature representation space, ensuring the stability of feature extraction and laying the foundation for fine-grained classification.

[0157] This application, when using imaging equipment, further utilizes ranging tools to actively measure key parts and morphological parameters of the animal target, in addition to identifying the animal target in the image data and its primary physiological characteristics, to obtain three-dimensional size data of the animal target's key morphology. Combined with the image-recognition-based primary physiological characteristics and the three-dimensional size data of the key morphology, a more accurate physiological characteristic of the animal target is obtained. On one hand, obtaining the three-dimensional size data of the animal target's key morphology using ranging tools provides a reliable physical scale basis. This, combined with the image-recognition-based primary physiological characteristics and the ranging-obtained three-dimensional size data of the key morphology, overcomes the inherent limitations of pure visual analysis, achieving information complementarity. The ranging-obtained three-dimensional size data of the key morphology effectively assists and optimizes the physiological characteristics obtained from the image data, significantly improving the accuracy of animal target physiological characteristic recognition. On the other hand, the three-dimensional size data of the animal target's key morphology serves as auxiliary input when using imaging equipment. The recognition model itself is still trained on the image dataset, without relying on difficult-to-obtain training samples containing ranging information. This ensures the powerful generalization ability of the recognition method in different deployment scenarios, resolving the contradiction between accuracy and universality inherent in traditional methods. This ensures the universality of the identification method in different scenarios.

[0158] The method for identifying the physiological characteristics of animals in this application utilizes high-precision keypoint detection technology to accurately locate the animal's head, trunk, limb joints, and sex-related feature regions, such as the head outline of a deer and the antler buds of a male deer, effectively overcoming the problems of missing texture and blurred anatomical structures in infrared images. The detected animal parts are standardized to enable the model to stably extract thermodynamic features. During model training, a knowledge distillation mechanism is used to transfer the implicit knowledge learned by a large-scale teacher model to the student model, which is then deployed on an imaging device. This enhances the model's ability to perceive subtle sex characteristics, significantly improving the accuracy of sex classification and age estimation, and strengthening the model's robustness in complex backgrounds and different day / night scenarios. In practical applications, laser ranging equipment is introduced to acquire three-dimensional dimensional data of key animal morphologies, assisting the model in more accurate physiological feature identification.

[0159] In another aspect, this application also provides a computer-readable storage medium storing a computer program. When executed by a processor, this computer program implements the various processes of the above-described method embodiment for identifying the physiological characteristics of animals, and achieves the same technical effects. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0160] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0161] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0162] The above are merely specific embodiments of this application, but the scope of protection of this application 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 this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method of identifying a physiological characteristic of an animal, characterized by, The method is applied to an imaging device, and comprises: acquiring image data; identifying an animal target in the image data and a first physiological feature of the animal target; acquiring three-dimensional size data of the animal target; obtaining a physiological feature of the animal target according to the three-dimensional size data and the first physiological feature.

2. The method of identifying a physiological characteristic of an animal of claim 1, wherein, The method further comprises: displaying the image data and the physiological feature of the animal target in the image data on a display screen of the imaging device.

3. The method of identifying a physiological characteristic of an animal according to claim 1 or 2, characterized in that, The physiological feature comprises at least one or more of a category, an age, a gender, a growth stage, and a thermal signal anomaly.

4. The method of identifying a physiological characteristic of an animal of claim 1, wherein, The obtaining of the physiological feature of the animal target according to the three-dimensional size data and the first physiological feature comprises: acquiring a second physiological feature of the animal target according to the three-dimensional size data; correcting the first physiological feature of the animal target according to the second physiological feature to obtain the physiological feature of the animal target.

5. The method of identifying a physiological characteristic of an animal of claim 4, wherein, The physiological feature comprises at least one of a gender and an age; the first physiological feature comprises at least one of a first gender and a first age; and the second physiological feature comprises at least one of a second gender and a second age. The correcting of the first physiological feature of the animal target according to the second physiological feature to obtain the physiological feature of the animal target comprises at least one of: in a case where the first gender is inconsistent with the second gender, determining the gender of the animal target according to the second gender; in a case where the first age is inconsistent with the second age, fusing the second age and the first age to obtain the age of the animal target.

6. The method of identifying a physiological characteristic of an animal according to claim 4 or 5, characterized in that, The acquiring of the second physiological feature of the animal target according to the three-dimensional size data comprises: acquiring a mapping relationship table between physiological features and three-dimensional size data of the animal target; querying the mapping relationship table according to the three-dimensional size data to acquire the second physiological feature of the animal target.

7. The method of identifying a physiological characteristic of an animal of claim 6, wherein, The three-dimensional size data adopts three-dimensional size data corresponding to a key morphology of the animal target. The physiological feature comprises a gender; the key morphology comprises a gender dimorphism feature; and the mapping relationship table of the animal target comprises a mapping relationship between the gender of the animal target and three-dimensional size data of the gender dimorphism feature. The physiological feature comprises an age; the key morphology comprises a body type; and the mapping relationship table of the animal target comprises a mapping relationship between the age of the animal target and three-dimensional size data of the body type. The fusing of the second age and the first age to obtain the age of the animal target comprises:

8. The method of identifying a physiological characteristic of an animal of claim 5, wherein, determining weights of the first age and the second age respectively according to a confidence of the first age; calculating an age weighted sum according to the weights, the first age, and the second age; determining the age of the animal target according to the age weighted sum. The acquiring of the three-dimensional size data of the animal target comprises:

9. The method of identifying a physiological characteristic of an animal of claim 1, wherein, acquiring three-dimensional size data of the animal target measured by a laser ranging module. The acquiring of the three-dimensional size data of the animal target measured by the laser ranging module comprises:

10. The method of identifying a physiological characteristic of an animal of claim 9, wherein, acquiring three-dimensional size data of the animal target measured by a laser ranging module. According to the animal target and the key points of the animal target, two measurement points of the animal target are determined; A pixel coordinate difference of the two measurement points on the image is calculated, and an angle difference between the two measurement points is determined according to the pixel coordinate difference; An actual distance from the laser ranging module to any one of the measurement points is obtained; According to the angle difference and the actual distance, the three-dimensional size data is determined.

11. An imaging device comprising a memory, a processor, an image data acquisition module and a ranging module connected to the processor; the image data acquisition module is configured to acquire image data; the ranging module is configured to measure three-dimensional size data of the animal target; the memory stores a computer program; the computer program is executed by the processor to implement the method for identifying the physiological characteristics of the animal according to any one of claims 1 to 10.

12. The imaging device of claim 11, wherein, The ranging module is a laser ranging module; the laser ranging module is parallel or coaxial with the optical axis of the image data acquisition module; And / or, The imaging device further comprises a display screen connected to the processor, and the display screen is configured to display the image data and the physiological characteristics of the animal target in the image data.

13. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the method for identifying the physiological characteristics of the animal according to any one of claims 1 to 10.

Images