Low power long endurance hunting camera method and system

Abstract

The present application relates to the field of computer vision, and discloses a low-power long-endurance hunting camera method and system, comprising: constructing a multi-level sensing layer of the hunting camera, collecting a thermal signal sequence and a grayscale contour image of an application scene to extract a multi-dimensional biological feature vector in the application scene; calculating a matching degree of the multi-dimensional biological feature vector and a non-target feature; when the matching degree is lower than a preset threshold, generating a first-level wake-up signal of the hunting camera; collecting a preliminary infrared image of a potential target, extracting an image feature of the preliminary infrared image to determine a target prey confidence of the potential target; when the target prey confidence is not higher than a preset confidence threshold, returning to a low-power sensing layer, and when the target prey confidence is higher than the preset confidence threshold, generating a second-level wake-up signal of the hunting camera to switch to a high-resolution sensing layer of the multi-level sensing layer to shoot a high-definition video of a high-value target. The present application can improve the endurance time and shooting effect of the hunting camera.

Description

Technical Field

[0001] This invention relates to a low-power, long-battery-life hunting camera method and system, belonging to the field of computer vision. Background Technology

[0002] Hunting cameras are highly automated digital cameras designed specifically for outdoor environments. Their core function is to automatically detect and photograph animal activity passing through their sensing area without human intervention. The low power consumption and long battery life of hunting cameras ensure that the device can operate continuously for extended periods in unattended outdoor environments, providing complete image data. This significantly reduces the number of times users need to visit deployment sites to change batteries, saving time, manpower, and transportation costs.

[0003] Traditional hunting cameras are triggered by passive infrared sensors. By detecting temperature changes in infrared radiation within their field of view, the main processor, which is in a dormant state, is immediately awakened when the temperature change reaches a fixed threshold, and the camera is instructed to perform the shooting task. This method cannot distinguish the source of heat, and any factor that causes rapid changes in infrared radiation may trigger the camera, resulting in a large number of false alarms. This causes the battery power to be consumed rapidly, which greatly shortens the actual battery life of the camera. Summary of the Invention

[0004] This invention provides a low-power, long-battery-life hunting camera method and system, the main purpose of which is to improve the battery life and shooting effect of the hunting camera.

[0005] To achieve the above objectives, the present invention provides a low-power, long-battery-life hunting camera method, comprising: A multi-level perception layer for a hunting camera is constructed. Through the low-power perception layer in the multi-level perception layer, thermal signal sequences and grayscale contour images of the hunting camera corresponding to the application scenario are collected to extract multi-dimensional biological feature vectors in the application scenario. Calculate the matching degree between the multidimensional biological feature vector and the corresponding non-target features in the non-target feature library of the multi-level perception layer; When the matching degree is lower than a preset threshold, potential targets in the application scenario are identified, and a first-level wake-up signal for the hunting camera is generated. Based on the first-level wake-up signal, the low-resolution perception layer of the multi-level perception layer is activated to acquire a preliminary infrared image of the potential target, and the image features of the preliminary infrared image are extracted to determine the target prey confidence of the potential target. When the confidence level of the target prey is not higher than the preset confidence threshold, the system returns to the low-power perception layer. When the confidence level of the target prey is higher than the preset confidence threshold, the system identifies high-value targets of the potential targets and generates a secondary wake-up signal for the hunting camera to switch to the high-resolution perception layer of the multi-level perception layer. Based on the high-resolution perception layer, the system captures high-definition video of the high-value targets.

[0006] Optionally, the construction of the multi-level perception layer for the hunting camera includes: A multi-level perception framework for the hunting camera is constructed, wherein the multi-level perception framework includes: a low-power perception layer, a low-resolution perception layer, and a high-resolution perception layer. Configure the multi-level sensing framework with multi-level sensors and multi-level processors; Construct a non-target feature library for the multi-level perception framework; Define the multi-level power management states corresponding to the multi-level perception framework, wherein the multi-level power management states include: sleep state, visual verification state, and high-definition recording state; Determine the wake-up rules of the multi-level sensing framework and the state switching logic of the multi-level power management state; The hunting camera's multi-level perception layer is integrated based on the wake-up rules, the state switching logic, the non-target feature library, the multi-level sensors, and the multi-level processor.

[0007] Optionally, the step of determining the wake-up rules of the multi-level sensing framework and the state switching logic of the multi-level power management state includes: The reinforcement learning elements of the hunting camera corresponding to the multi-level perception framework are defined, wherein the reinforcement learning elements include: agent, application environment, discrete state, action space and learning reward. Based on the aforementioned reinforcement learning elements, the reinforcement learning algorithm for the hunting camera is determined. Based on the reinforcement learning algorithm, the multi-level wake-up threshold and multi-level state switching parameters of the hunting camera are output. Based on the multi-level wake-up threshold and the multi-level state switching parameters, the wake-up rules of the multi-level perception framework and the state switching logic of the multi-level power management state are determined respectively.

[0008] Optionally, the extraction of multidimensional biometric feature vectors in the application scenario includes: The thermal sensing signal sequence corresponding to the application scenario is filtered and denoised to obtain a denoised thermal sensing signal sequence. Based on the denoised thermal sensing signal sequence, the movement continuity characteristics of organisms and the rate of change of hot spot area in the application scenario are extracted. The grayscale contour image corresponding to the application scenario is binarized to obtain a black and white binary image. Based on the black-and-white binary image, calculate the aspect ratio of the target outline and the complexity of the main outline of the organism in the application scenario; The features of the movement continuity, the hot spot area change rate, the target contour aspect ratio, and the main body contour complexity are fused to obtain a multidimensional biological feature vector.

[0009] Optionally, the step of extracting the movement continuity features and hot spot area change rate of organisms in the application scenario based on the denoised thermal sensing signal sequence includes: Identify the dominant probe sequence of the denoised thermal sensing signal sequence; Calculate the number of discontinuous transitions in the dominant probe sequence; Based on the number of discontinuous jumps, calculate the mobility continuity score of the organisms in the application scenario to determine the mobility continuity characteristics of the organisms in the application scenario; Based on the denoised thermal sensing signal sequence, the instantaneous hot spot area sequence of the organism in the application scenario is calculated; Calculate the mean and standard deviation of the instantaneous hot spot area of ​​the instantaneous hot spot area sequence; The rate of change of hot spot area of ​​organisms in the application scenario is calculated based on the mean instantaneous hot spot area and the standard deviation of instantaneous hot spot area.

[0010] Optionally, calculating the matching degree between the multidimensional biometric feature vector and the corresponding non-target features in the non-target feature library of the multi-level perception layer includes: Calculate the prototype weights of the non-target features; Calculate the cosine similarity between the multidimensional biological feature vector and the non-target feature; The matching degree between the multidimensional biometric vector and the non-target feature is calculated based on the prototype weight and the cosine similarity.

[0011] Optionally, the extraction of image features from the preliminary infrared image includes: The preliminary infrared image is denoised to obtain a denoised infrared image; The denoised infrared image is binarized to obtain a binarized infrared image; Morphological processing is performed on the binarized infrared image to obtain a morphological infrared image; The morphological infrared image is segmented for interest to obtain the region of interest; Extract the boundary contours of the region of interest in the image; Based on the boundary contour, calculate the geometric features, statistical features, and texture features of the region of interest; The geometric features, statistical features, and texture features are integrated to obtain image features.

[0012] Optionally, determining the target prey confidence of the potential target includes: Define the prey prototype vector and the interference source feature vector of the potential target; Calculate the feature similarity between the prey prototype vector and the image features corresponding to the potential target; Analyze the distribution of interference sources in the feature vectors of the interference sources; Calculate the deviation between the image features and the distribution of the interference sources; The target prey confidence of the potential target is calculated based on the feature similarity and the deviation.

[0013] Optionally, capturing high-definition video of the high-value target based on the high-resolution perception layer includes: Based on the high-resolution perception layer, continuous frame images of the high-value target are acquired; Based on the consecutive frame images, the target size and movement speed of the high-value target are determined; Based on the target's size and movement speed, the shooting parameters and shooting mode of the main image sensor corresponding to the high-resolution perception layer are determined; Based on the shooting parameters and shooting mode, the high-value target is photographed to obtain high-definition video.

[0014] To address the above problems, the present invention also provides a low-power, long-battery-life hunting camera system, the system comprising: A multi-level perception layer construction module is used to construct a multi-level perception layer for a hunting camera. Through the low-power perception layer in the multi-level perception layer, thermal signal sequences and grayscale contour images of the hunting camera corresponding to the application scenario are collected to extract multi-dimensional biological feature vectors in the application scenario. The matching degree calculation module is used to calculate the matching degree between the multidimensional biological feature vector and the corresponding non-target feature in the non-target feature library of the multi-level perception layer; A first-level wake-up signal generation module is used to determine potential targets in the application scenario and generate a first-level wake-up signal for the hunting camera when the matching degree is lower than a preset threshold. The confidence calculation module is used to activate the low-resolution perception layer of the multi-level perception layer based on the first-level wake-up signal, so as to acquire the preliminary infrared image of the potential target, extract the image features of the preliminary infrared image, and determine the target prey confidence of the potential target. The high-resolution shooting module is used to return to the low-power perception layer when the confidence level of the target prey is not higher than a preset confidence threshold, and to identify high-value targets of the potential targets when the confidence level of the target prey is higher than the preset confidence threshold, and to generate a secondary wake-up signal for the hunting camera to switch to the high-resolution perception layer of the multi-level perception layer, and to shoot high-definition video of the high-value targets based on the high-resolution perception layer.

[0015] Compared to the problems described in the background technology, this invention acquires thermal signal sequences and grayscale contour images through a low-power sensing layer to extract multi-dimensional biological feature vectors and calculate their matching degree with a non-target feature library. This enables effective filtering under extremely low power consumption. When the matching degree is higher than a preset threshold, a first-level wake-up signal is generated only for potential targets, avoiding frequent false triggers caused by non-target factors, thus minimizing the number of times the device wakes up from deep sleep. This feature-matching-based pre-screening mechanism is the core of the long-lasting battery design, ensuring that the camera maintains a microamplitude standby current for most of the time. This invention activates a low-resolution sensing layer based on the first-level wake-up signal, acquires preliminary infrared images, and extracts image features to determine the target prey confidence level, constituting a second-level intelligent filter. This layer can distinguish between real animal targets and non-living heat sources that cause thermal signals. When the confidence level is not higher than the threshold, it quickly returns to a low-power standby state, avoiding energy waste caused by activating high-power modules. Only when the confidence level exceeds the threshold and it is confirmed as a high-value target is a second-level wake-up signal generated. The hierarchical decision-making mechanism precisely allocates computing resources and energy consumption to valuable shooting events, greatly improving battery efficiency. Upon the generation of the secondary wake-up signal, the camera switches to the high-resolution perception layer for high-definition video recording, ensuring high-quality recording of high-value targets. The entire process, from eliminating non-target features to initial confirmation of potential targets and precise identification of high-value targets, progresses step-by-step, avoiding ineffective shooting and wasted storage space. The storage medium is no longer filled with meaningless empty shots or accidental captures, but rather with high-value, analyzable data, improving data management efficiency and ease of subsequent retrieval. Finally, this multi-level perception architecture, through intelligent hierarchical wake-up and decision-making, achieves an optimal balance between power consumption, performance, and data value, fundamentally solving the core pain points of traditional hunting cameras: short battery life, high false alarm rate, and poor data quality. This enables the device to conduct autonomous, efficient, and accurate monitoring in the wild for months or even longer, providing reliable technical support for wildlife research, security monitoring, and outdoor activities. Therefore, the low-power, long-battery-life hunting camera method provided in this embodiment of the invention can improve the battery life and shooting effect of the hunting camera. Attached Figure Description

[0016] Figure 1This is a flowchart illustrating a low-power, long-battery-life hunting camera method according to an embodiment of the present invention.

[0017] Figure 2 An exploded view of a hunting camera according to an embodiment of the low-power, long-battery-life hunting camera method provided by the present invention.

[0018] Figure 3 The diagram shows a hunting camera circuit for a low-power, long-battery-life hunting camera method provided in an embodiment of the present invention.

[0019] Figure 4 This is a schematic diagram of a low-power, long-battery-life hunting camera system provided in an embodiment of the present invention.

[0020] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0022] This application provides a low-power, long-battery-life hunting camera method. The executing entity of the low-power, long-battery-life hunting camera method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application embodiment: a server, a terminal, etc. In other words, the low-power, long-battery-life hunting camera method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0023] Reference Figure 1 The diagram shown is a flowchart illustrating a low-power, long-battery-life hunting camera method according to an embodiment of the present invention. In this embodiment, the low-power, long-battery-life hunting camera method includes: S1. Construct a multi-level perception layer for the hunting camera. Through the low-power perception layer in the multi-level perception layer, collect the thermal signal sequence and grayscale contour image of the application scenario corresponding to the hunting camera to extract the multi-dimensional biological feature vector in the application scenario.

[0024] This invention, through constructing a multi-level perception layer for hunting cameras, can significantly extend device battery life by avoiding frequent startups of the high-power main system through tiered wake-up, allowing it to operate only when necessary. The multi-level perception layer refers to a hierarchical intelligent perception system architecture divided by power consumption and function.

[0025] As an embodiment of the present invention, the construction of the multi-level perception layer of the hunting camera includes: A multi-level perception framework for the hunting camera is constructed, wherein the multi-level perception framework includes: a low-power perception layer, a low-resolution perception layer, and a high-resolution perception layer. Configure the multi-level sensing framework with multi-level sensors and multi-level processors; Construct a non-target feature library for the multi-level perception framework; Define the multi-level power management states corresponding to the multi-level perception framework, wherein the multi-level power management states include: sleep state, visual verification state, and high-definition recording state; Determine the wake-up rules of the multi-level sensing framework and the state switching logic of the multi-level power management state; The hunting camera's multi-level perception layer is integrated based on the wake-up rules, the state switching logic, the non-target feature library, the multi-level sensors, and the multi-level processor.

[0026] The multi-level perception framework refers to a hierarchical, event-driven intelligent perception system architecture. The low-power perception layer, the first level of the multi-level perception framework, is composed of ultra-low-power components and is responsible for the entire operation. It quickly filters non-target events by acquiring basic signals and only triggers the wake-up of the next higher level when a potential target is detected. The low-resolution perception layer, the second level of the multi-level perception framework, is in sleep mode by default and is activated after being woken up by the low-power layer. It uses low-resolution images to perform preliminary visual verification of potential targets and assess their confidence level as target prey. The resolution perception layer, the third level of the multi-level perception framework, is in deep sleep mode by default and is activated after being identified as a high-value target by the low-resolution layer. It is responsible for mobilizing all resources for high-definition image capture. The multi-level sensor refers to a heterogeneous sensor combination customized for different perception levels, including a basic sensor for the low-power layer, a low-resolution image sensor for the low-resolution layer, and a high-resolution image sensor for the high-resolution layer. The multi-level processor refers to a collaborative computing unit composed of processors with different performance and power consumption levels, including a coprocessor responsible for low-power layer algorithms and wake-up management, and a main processor responsible for complex image processing and system control. The non-target feature library refers to a lightweight feature database stored in the coprocessor, containing multi-dimensional feature vectors of typical non-target events. The multi-level power management states refer to a set of refined system power operation modes coupled with the perception framework. The sleep state refers to the default lowest power consumption state, where only the coprocessor and low-power layer sensors remain running, while power to other high-power components is cut off. The visual verification state refers to a medium power consumption state, where the main processor is awakened, the low-resolution sensor is activated, and preliminary image analysis and target confidence assessment are performed. The high-definition recording state refers to the highest power consumption state, where the main processor runs at full speed, and high-resolution sensors and supplementary lighting modules are activated to capture and store high-definition images. The wake-up rule refers to a preset condition that defines the conditions that trigger the system's transition from a low-power state to a high-power state. The state switching logic refers to the decision-making process that controls the transitions between the multi-level power management states.

[0027] For further information about the hunting camera, please refer to [link / reference]. Figure 2 and Figure 3 , Figure 2An exploded view of a hunting camera according to an embodiment of the present invention shows all the disassembled parts of the hunting camera and their assembly relationship. The following numbered parts are included in the figure: (1) Antenna: used for wireless communication (such as 4G / 5G); (2) Front shell: the front protective shell of the camera; (3) Rear shell: the rear protective shell of the camera; (4) Main board: the core control circuit board of the camera; (5) Lens module: including lens, lens bracket, lens waterproof ring, etc., used for optical imaging; (6) Main body waterproof rubber ring: ensures the waterproof seal between the camera body and the shell; (7) Light board: integrates infrared fill light and visible light fill light; (8) Lens cover: protects the lens, usually with Fresnel lens; (9) Infrared lens: allows infrared light to pass through, used for infrared imaging; (10) Charging port plug: protects the charging interface; (11) Power board: manages battery power supply and charging; (12) Backing adhesive: used for fixing or sealing; (13) Microphone: used for sound acquisition; (14) Battery charging board: manages battery charging. Process; (15) Lens bracket: to fix the lens assembly; (16) Lens waterproof ring: to ensure the waterproofness of the lens module; (17) Waterproof rubber plug: to seal other holes; (18) Back copper nut: to fix the back shell; (19) Shaft: to connect the front and rear shells and realize opening and closing; (20) Fresnel lens: to focus the infrared thermal signal and improve the detection range and sensitivity of the PIR sensor; (21) Light-blocking sponge: to prevent stray light from interfering with the sensor; (22) Buckle: to fix the housing components; (23) (24) Battery holder: to hold the battery in place; (25) Battery: to provide power to the camera; (26) Bottom copper nut: to hold the bottom of the camera in place; (27) Button board: the circuit board that holds the buttons; (28) Front cover: the protective cover at the front of the camera; (29) Front cover waterproof rubber ring: to ensure the seal between the front cover and the housing; (30) Battery compartment waterproof rubber ring: to ensure the waterproofness of the battery compartment; (31) Shaft: same as above, connecting the front and rear housings; (32) Battery compartment bottom cover: the bottom cover of the battery compartment.

[0028] Figure 3 The diagram shows a hunting camera circuit based on a low-power, long-battery-life hunting camera method according to an embodiment of the present invention. The diagram illustrates the main control circuit of the hunting camera, with MCU: L170FAUA and HOST: T32Z as its core. Clear lines connect the various functional modules and their interface relationships. Key modules and their functions are described below: The main control unit includes an MCU (L170FAUA): a low-power microcontroller unit responsible for wake-up management, basic signal processing, and power control of the low-power sensing layer; and a HOST (T32Z): a main processor responsible for complex image processing, high-resolution shooting control, system decision-making, and communication.

[0029] The sensors and peripherals include: PIR (Passive Infrared Sensor): a passive infrared sensor used to detect heat source movement and a key component of the low-power sensing layer; STA-LED: a status indicator used to display the camera's operating status; RST_KEY: a reset button; XTAL-32.768Hz: a real-time clock crystal oscillator providing a precise time base for the system; XTAL-24MHz: a main processor's operating clock crystal oscillator; Hygro / Temp Meter: a temperature and humidity sensor used for environmental monitoring; SENSOR: a main image sensor used for high-resolution imaging; IR-LED: an infrared fill light used for infrared imaging in nighttime or low-light environments; T-FLASH: a memory chip used to store programs and data; ID-EEPCM: an electrically erasable programmable read-only memory used to store key information such as the device ID; MIC: a microphone used for audio acquisition; and 4G EC800 Module: a 4G communication module used for remote data transmission.

[0030] Power management includes VIN, Samp, BAT: power input, sampling, and battery interface; CHANGE, CouComB, LDG: charging management related circuits; ELECTRICITYV: power management unit, responsible for voltage regulation and distribution of the entire system.

[0031] Storage and interfaces include FLASH: main memory used to store high-definition video and images; MIF1, SPI, GPIO, ADC, SDIO, I2C: various communication interfaces for connecting sensors, memory and other peripherals; DEBUG, ES-ANT EC800: debug interface and antenna interface.

[0032] The entire circuit diagram clearly shows the connections and signal interfaces between each module. Figure 2 The internal structure of the hunting camera and the assembly sequence of its components are clearly presented, facilitating understanding of the product design, repair and disassembly, and bill of materials verification. Figure 3 This effectively demonstrates the hardware architecture of the hunting camera and the connection relationships between its modules, thus providing structural data support for the subsequent low-power, long-battery-life control of the hunting camera.

[0033] Optionally, the multi-level perception framework of the hunting camera can be constructed using machine learning techniques, such as supervised learning and unsupervised learning.

[0034] Optionally, the non-target feature library of the multi-level perception framework can be constructed using unsupervised algorithms, such as DBSCAN, Isolation Forest, One-Class SVM, etc.

[0035] Optionally, the multi-level perception layer of the hunting camera can be integrated using a virtual integration method of digital twins. For example, based on the wake-up rules, the state switching logic, the non-target feature library, the multi-level sensors, and the multi-level processor, a multi-level perception layer digital twin of the hunting camera can be constructed. The multi-level perception layer digital twin can be simulated and tested to obtain simulation test results, so as to analyze the response accuracy and power efficiency of the multi-level perception layer digital twin. Based on the response accuracy and the power efficiency, the multi-level perception layer of the hunting camera can be determined.

[0036] Optionally, the step of determining the wake-up rules of the multi-level sensing framework and the state switching logic of the multi-level power management state includes: The reinforcement learning elements of the hunting camera corresponding to the multi-level perception framework are defined, wherein the reinforcement learning elements include: agent, application environment, discrete state, action space and learning reward. Based on the aforementioned reinforcement learning elements, the reinforcement learning algorithm for the hunting camera is determined. Based on the reinforcement learning algorithm, the multi-level wake-up threshold and multi-level state switching parameters of the hunting camera are output. Based on the multi-level wake-up threshold and the multi-level state switching parameters, the wake-up rules of the multi-level perception framework and the state switching logic of the multi-level power management state are determined respectively.

[0037] The reinforcement learning elements refer to a set of core concepts used to build and train an agent. The agent, within the reinforcement learning framework, is a computational entity capable of autonomously perceiving its environment, making decisions, and executing actions. The application environment refers to the real physical world where the hunting camera is deployed and operates, and the set of all its related variables, such as geography, climate, lighting, and battery level. The discrete states refer to a finite set of non-overlapping environmental snapshots formed after quantifying and classifying the application environment. The action space refers to the set of all possible actions that the agent can perform in any given state. The learning reward is a scalar signal fed back to the agent by the environment, used to evaluate the "goodness" of the agent performing a certain action in a specific state. The reinforcement learning algorithm refers to a specific mathematical model and computational process used to guide the agent in continuously optimizing its decision-making strategy based on experience (states, actions, reward sequences) generated from interactions with the environment. The multi-level wake-up threshold refers to the dynamic boundary conditions output by the reinforcement learning algorithm that trigger transitions from low-power states to high-power states. The multi-level state switching parameters refer to a set of control variables output by a reinforcement learning algorithm, used for smooth and intelligent switching between power management states.

[0038] Optionally, the reinforcement learning elements of the hunting camera corresponding to the multi-level perception framework can be defined through simulation and iterative definition of digital twins, such as MATLAB / Simulink, Unity, etc.

[0039] Optionally, the reinforcement learning algorithm of the hunting camera can be determined by a fast adaptive algorithm of meta-learning, such as a meta-learning model.

[0040] This invention, through a low-power sensing layer within the multi-level sensing layer, acquires thermal signal sequences and grayscale contour images of the hunting camera's corresponding application scenario. This enables continuous monitoring of environmental changes with extremely low power consumption, effectively filtering out non-target events caused by environmental factors such as wind, rain, and lighting. The thermal signal sequence refers to a low-dimensional thermal signal data stream, varying over time and containing spatial information, output by a one-dimensional pyroelectric infrared (PIR) array sensor. The grayscale contour image refers to a multi-frame image captured by an ultra-low-power, low-resolution grayscale vision sensor, containing only scene brightness information and exhibiting extremely low detail.

[0041] This invention enables accurate species identification and individual re-identification of targets by extracting multidimensional biometric vectors from the application scenario, and automatically analyzes and classifies their behavior. The multidimensional biometric vector refers to a feature vector composed of multiple numerical features extracted from thermal signal sequences and grayscale contour images, which collectively encode the essential attributes of the triggering target in dynamic thermal behavior and static morphological contours.

[0042] As an embodiment of the present invention, the extraction of multidimensional biological feature vectors in the application scenario includes: The thermal sensing signal sequence corresponding to the application scenario is filtered and denoised to obtain a denoised thermal sensing signal sequence. Based on the denoised thermal sensing signal sequence, the movement continuity characteristics of organisms and the rate of change of hot spot area in the application scenario are extracted. The grayscale contour image corresponding to the application scenario is binarized to obtain a black and white binary image. Based on the black-and-white binary image, calculate the aspect ratio of the target outline and the complexity of the main outline of the organism in the application scenario; The features of the movement continuity, the hot spot area change rate, the target contour aspect ratio, and the main body contour complexity are fused to obtain a multidimensional biological feature vector.

[0043] The denoised thermal sensing signal sequence refers to a smooth and stable signal sequence obtained by filtering the signal output from the original one-dimensional pyroelectric infrared array sensor. The movement continuity feature is a scalar feature quantifying the smoothness and directional consistency of the biological movement path. The hot spot area change rate is a scalar feature describing the degree of fluctuation in the effective thermal radiation area of ​​the biological heat source during movement. The binary image refers to an image containing only two pixel values ​​obtained by segmenting the original low-resolution grayscale image using a fixed threshold method. The target contour aspect ratio is a scalar feature describing the overall shape proportion of the biological target in the image, calculated using its minimum bounding rectangle. The main body contour complexity is a scalar feature measuring the irregularity and fragmentation of the biological target contour.

[0044] Optionally, the denoised thermal sensing signal sequence can be obtained by a filtering algorithm, such as moving average filtering or low-pass digital filtering.

[0045] Optionally, the black-and-white binary image can be obtained using a global thresholding method or Otsu's method.

[0046] Optionally, the step of extracting the movement continuity features and hot spot area change rate of organisms in the application scenario based on the denoised thermal sensing signal sequence includes: Identify the dominant probe sequence of the denoised thermal sensing signal sequence; Calculate the number of discontinuous transitions in the dominant probe sequence; Based on the number of discontinuous jumps, calculate the mobility continuity score of the organisms in the application scenario to determine the mobility continuity characteristics of the organisms in the application scenario; Based on the denoised thermal sensing signal sequence, the instantaneous hot spot area sequence of the organism in the application scenario is calculated; Calculate the mean and standard deviation of the instantaneous hot spot area of ​​the instantaneous hot spot area sequence; The rate of change of hot spot area of ​​organisms in the application scenario is calculated based on the mean instantaneous hot spot area and the standard deviation of instantaneous hot spot area.

[0047] The dominant detector sequence refers to the ordered sequence of detector numbers extracted chronologically from the denoised thermal signal sequence, representing the detectors with the strongest signal amplitude at each sampling moment. The number of discontinuous jumps refers to the number of times the absolute value of the change in the dominant detector number exceeds 1 between two adjacent moments in the dominant detector sequence. The motion continuity score is a scalar index used to quantify motion continuity, calculated by normalizing the number of discontinuous jumps. The instantaneous hotspot area sequence refers to the ordered sequence of effective hotspot area values ​​calculated chronologically from the denoised thermal signal sequence at each sampling moment. The mean instantaneous hotspot area refers to the arithmetic mean of all values ​​in the instantaneous hotspot area sequence. The standard deviation of the instantaneous hotspot area refers to the statistical measure of the dispersion of all values ​​in the instantaneous hotspot area sequence relative to its mean.

[0048] Optionally, the rate of change of the hot spot area of ​​the organism in the application scenario can be obtained by calculating the ratio of the standard deviation of the instantaneous hot spot area to the mean of the instantaneous hot spot area.

[0049] As another implementation, the number of discontinuous jumps is calculated using the following formula: ; in, Indicates the number of discontinuous transitions. Indicates the time index of the dominant probe sequence. Indicates the sequence length of the dominant probe sequence. Indicates an indicator function, Indicating the dominant probe sequence The dominant detector element number at that time Indicating the dominant probe sequence The dominant detection element number at that time.

[0050] It needs to be explained that in this application, the formula... Indicates an indicator function, when When the condition is met, the value is 1; otherwise, the value is 0.

[0051] In another implementation, the mobility continuity score is calculated using the following formula: ; in, This represents the continuity score, where K represents the number of discontinuous jumps. This indicates the sequence length of the dominant probe sequence.

[0052] As another implementation, the mean instantaneous hot spot area and the standard deviation of the instantaneous hot spot area are calculated using the following formula: ; ; in, This represents the average instantaneous hot spot area. This represents the standard deviation of the instantaneous hot spot area. The sequence length represents the instantaneous hot spot area sequence. The time index represents the instantaneous hotspot area sequence. Represents the time in the instantaneous hotspot area sequence The instantaneous area of ​​the hot spot.

[0053] S2. Calculate the matching degree between the multidimensional biological feature vector and the corresponding non-target feature in the non-target feature library of the multi-level perception layer.

[0054] This invention, through calculating the matching degree between the multidimensional biometric vector and the corresponding non-target features in the multi-level perception layer's non-target feature library, achieves precise filtering of non-target events (such as rustling grass or changes in light and shadow). This significantly reduces the system's false alarm rate, effectively preventing the main processor and high-power modules from being frequently woken up by invalid events. This not only improves the accuracy and reliability of system decision-making but also saves energy to extend device battery life, ultimately increasing the proportion of valid data collected. The matching degree refers to a quantified value of the similarity between the currently extracted multidimensional biometric vector and the vector templates pre-stored in the non-target feature library.

[0055] As an embodiment of the present invention, calculating the matching degree between the multidimensional biometric feature vector and the corresponding non-target features in the non-target feature library of the multi-level perception layer includes: Calculate the prototype weights of the non-target features; Calculate the cosine similarity between the multidimensional biological feature vector and the non-target feature; The matching degree between the multidimensional biometric vector and the non-target feature is calculated based on the prototype weight and the cosine similarity.

[0056] Here, the prototype weight refers to the weight corresponding to each non-target prototype (cluster center) generated in the non-target feature library. The cosine similarity is an index that measures the degree of similarity between two vectors in a direction.

[0057] Optionally, the prototype weights of the non-target features can be calculated using clustering algorithms such as K-Means, K-Medoids, etc.

[0058] As another implementation, the matching degree can be calculated using the following formula: ; in, Indicates the degree of matching. This indicates the number of prototypes in the non-target feature library. Represents multidimensional biological feature vectors and the first The cosine similarity between each prototype and a non-target feature. Represents a multidimensional biological feature vector. Indicates the first non-target feature in the feature library Each prototype corresponds to a non-target feature. Represents prototype weights. Indicated by An exponential function with base 0. This represents the focusing coefficient.

[0059] It needs to be explained that in this application, This represents the focusing coefficient, ranging from 1 to 10, used to adjust the weighting of cosine similarity. When it is large (e.g.) =10), making the largest one Dominating the final result, highly rewarding specific matching, when When smaller (e.g.) =1), the adjustment effect will be smoother. The prototype weights refer to the importance coefficients of each cluster center in the non-target feature library, calculated using the K-Means clustering algorithm.

[0060] S3. When the matching degree is lower than a preset threshold, identify potential targets in the application scenario and generate a first-level wake-up signal for the hunting camera.

[0061] This invention, by identifying potential targets in the application scenario when the matching degree is below a preset threshold, can achieve intelligent filtering of invalid events, reduce the system's false alarm rate, avoid frequent false triggering of high-power modules, thereby saving device power to extend battery life, improving system resource utilization efficiency, and ensuring that subsequent processing can focus on high-value targets. The potential targets refer to all unknown organisms that may attract the system's further attention after initial judgment by the low-power perception layer, excluding explicit non-target events.

[0062] This invention, through the generation of a first-level wake-up signal for the hunting camera, can activate the sensing modules following the low-power sensing layer as needed, achieving stepped control of system power consumption. This avoids the direct activation of high-power modules such as the high-resolution sensing layer, ensuring monitoring sensitivity while achieving preliminary energy management and constructing key decision nodes in the multi-level sensing process. The first-level wake-up signal is an internal instruction generated by the low-power sensing layer, whose core function is to notify the system's control unit to wake up from deep sleep and activate the low-resolution sensing layer.

[0063] S4. Based on the first-level wake-up signal, activate the low-resolution perception layer of the multi-level perception layer to acquire a preliminary infrared image of the potential target, extract the image features of the preliminary infrared image, and determine the target prey confidence of the potential target.

[0064] In this embodiment of the invention, by activating the low-resolution perception layer of the multi-level perception layer based on the first-level wake-up signal, the potential target can be preliminarily imaged and confirmed. Image features of the preliminary infrared image are extracted to calculate the target prey confidence level, thereby distinguishing the value of potential targets and achieving secondary filtering of non-target animals. This avoids unnecessarily activating the high-power high-definition shooting function, ultimately further reducing system energy consumption and extending device battery life.

[0065] This invention enables the acquisition of visual morphological information of potential targets by collecting preliminary infrared images, extracting image features suitable for target identification, and providing a data foundation for calculating target prey confidence. The preliminary infrared images refer to low-resolution thermal imaging images acquired by a low-resolution sensing layer for preliminary confirmation and value judgment of potential targets.

[0066] Optionally, the preliminary infrared image can be acquired using a low-power near-infrared supplemental LED in a low-resolution sensing layer and a low-resolution main image sensor.

[0067] This invention, through the extraction of image features from the preliminary infrared image, can transform image information into quantifiable data, providing a basis for target classification and identification, thereby achieving automated value assessment of potential targets. The image features refer to quantifiable mathematical indicators extracted from the preliminary infrared image that can be used to describe and distinguish potential targets.

[0068] As an embodiment of the present invention, the extraction of image features from the preliminary infrared image includes: The preliminary infrared image is denoised to obtain a denoised infrared image; The denoised infrared image is binarized to obtain a binarized infrared image; Morphological processing is performed on the binarized infrared image to obtain a morphological infrared image; The morphological infrared image is segmented for interest to obtain the region of interest; Extract the boundary contours of the region of interest in the image; Based on the boundary contour, calculate the geometric features, statistical features, and texture features of the region of interest; The geometric features, statistical features, and texture features are integrated to obtain image features.

[0069] The denoised infrared image refers to an image obtained by applying filtering algorithms (such as Gaussian filtering or median filtering) to a preliminary infrared image to reduce or eliminate random noise. The binarized infrared image refers to a black-and-white image obtained by applying threshold segmentation to the denoised infrared image, where pixel values ​​contain only 0 (representing background) and 255 (representing the target). The morphological infrared image refers to an image obtained by performing morphological operations such as dilation, erosion, opening, and closing operations on the binarized infrared image to correct the target shape, fill internal holes, and remove isolated noise points. The region of interest (ROI) refers to a set of interconnected pixels in the morphological infrared image, identified through connected component analysis, representing an independent potential target. The boundary contour refers to an ordered sequence of pixel coordinates that constitutes the outer boundary of the ROI. The geometric features refer to quantitative indicators calculated based on the boundary contour, describing the target's shape and size, including area, perimeter, aspect ratio, and compactness. The statistical features refer to quantitative indicators calculated based on the pixel values ​​of the ROI in the original grayscale image, describing the target's temperature distribution, including average temperature, maximum / minimum temperature, and temperature variance. The texture features refer to quantitative indicators that describe the spatial distribution pattern and arrangement of pixel grayscale values ​​within a region of interest, such as energy and contrast extracted through the grayscale co-occurrence matrix.

[0070] Optionally, the region of interest can be obtained by connected component analysis, such as the Two-Pass algorithm or the seed filling algorithm.

[0071] Optionally, the boundary contour of the region of interest can be extracted using a contour-finding algorithm, such as the Suzuki algorithm.

[0072] This invention, through determining the target prey confidence level of the potential target, enables automated value judgment of the potential target, providing a decision-making basis for whether the system should initiate high-resolution shooting. This accurately distinguishes high-value prey from non-target animals, avoids ineffective high-definition shooting of low-value targets, maximizes the effective utilization of high-power mode, and ultimately greatly reduces overall system energy consumption and extends device battery life. The target prey confidence level refers to the degree of credibility with which the system judges a potential target as high-value prey (such as deer, wild boar, etc.).

[0073] As an embodiment of the present invention, determining the target prey confidence of the potential target includes: Define the prey prototype vector and the interference source feature vector of the potential target; Calculate the feature similarity between the prey prototype vector and the image features corresponding to the potential target; Analyze the distribution of interference sources in the feature vectors of the interference sources; Calculate the deviation between the image features and the distribution of the interference sources; The target prey confidence of the potential target is calculated based on the feature similarity and the deviation.

[0074] Here, the prey prototype vector refers to a baseline feature vector in the feature space that represents the ideal prey. The interference source feature vector refers to the set of sample feature vectors representing various known interference sources in the feature space. The feature similarity refers to the degree of similarity in direction and pattern between the image feature vector of the potential target and the prey prototype vector. The interference source distribution refers to the overall distribution characteristics of all interference source feature vectors in the feature space. The deviation refers to the degree to which the image feature vector of the potential target deviates from the interference source distribution.

[0075] Optionally, the prey prototype vector and the interference source feature vector of the potential target can be defined by a clustering algorithm, such as K-Means.

[0076] Optionally, the interference source distribution of the interference source feature vector can be determined by calculating the covariance matrix and mean vector of the interference source feature vector to fit the probability density function of the interference source feature vector, thereby determining the interference source distribution of the interference source feature vector. If the interference source feature vector follows a multidimensional Gaussian distribution, the interference source distribution is represented by the following formula: ; in, The distribution of interference sources represents the feature vectors of interference sources. Represents the feature vector of the interference source. Represents pi (π). This represents the dimension of the feature vector of the interference source. Represents the covariance matrix. Indicated by An exponential function with base 0. Represents the mean vector. Representing vectors transpose, This represents the inverse of the covariance matrix.

[0077] Optionally, the deviation between the image features and the distribution of the interference sources can be calculated using the Mahalanobis distance formula.

[0078] As another implementation, the target prey confidence level is calculated using the following formula: ; in, Indicates the confidence level of the target prey. Indicated by An exponential function with base 0. Indicates feature similarity. Indicates the degree of deviation. Indicated by Logarithmic function with base 0. Weights representing feature similarity The weight representing the degree of deviation.

[0079] It needs to be explained that in this application, the formula... The weights representing feature similarity take values ​​of (0,1). The weight representing the deviation, taking values ​​of (0,1), satisfies... .

[0080] S5. When the confidence level of the target prey is not higher than the preset confidence threshold, return to the low-power perception layer. When the confidence level of the target prey is higher than the preset confidence threshold, identify the high-value target of the potential target and generate a secondary wake-up signal for the hunting camera to switch to the high-resolution perception layer of the multi-level perception layer. Based on the high-resolution perception layer, capture high-definition video of the high-value target.

[0081] In this embodiment of the invention, when the confidence level of the target prey is not higher than a preset confidence threshold, returning to the low-power perception layer can avoid triggering the high-resolution imaging unit, prevent the system from entering a high-power operating mode, thereby reducing the storage and transmission of invalid images, greatly reducing the average power consumption of the device, significantly extending battery life, and improving the effective utilization of storage space.

[0082] This invention, in its embodiments, triggers a high-resolution imaging unit to capture high-definition images or videos of potential prey when the confidence level of the target prey exceeds a preset confidence threshold. This captures clear details and biological characteristics of the target, providing high-quality data for subsequent species identification and behavioral analysis. It ensures that only high-value targets are recorded in high definition, ultimately improving the effectiveness of monitoring data. The high-value targets refer to specific prey species, such as deer, wild boar, and bears, that are pre-selected by the user or system as having the highest monitoring and recording value in the application scenario of the hunting camera.

[0083] This invention enables tiered energy management by generating a secondary wake-up signal for the hunting camera. This allows for precise control of the startup timing of high-power modules, preventing false environmental alarms from triggering the entire system. Consequently, it significantly reduces the system's average standby power consumption, extends battery life, and improves the efficiency of capturing effective targets. The secondary wake-up signal is generated by the low-power sensing layer within the hunting camera system and serves to wake up and activate the higher-level, more power-consuming processing unit using internal electronic control signals.

[0084] This invention, by switching to the high-resolution perception layer of the multi-level perception layer, can acquire clear details and biological characteristics of the target, providing high-quality data for species identification and behavior analysis, thereby enhancing the scientific and commercial value of monitoring data, meeting users' needs for high-quality image recording, and ultimately achieving accurate and effective recording of high-value targets.

[0085] This invention, by capturing high-definition video of the high-value target based on the aforementioned high-resolution perception layer, can record the target's continuous dynamic behavior, capture complete activity sequences and interaction processes, thereby analyzing the target's movement patterns, predation strategies, or social behavior, and obtaining richer biological information than static images. The high-definition video refers to video data with specific technical specifications and purposes, specifically recorded by the high-resolution perception layer of the hunting camera after identifying the high-value target.

[0086] As an embodiment of the present invention, capturing high-definition video of the high-value target based on the high-resolution perception layer includes: Based on the high-resolution perception layer, continuous frame images of the high-value target are acquired; Based on the consecutive frame images, the target size and movement speed of the high-value target are determined; Based on the target's size and movement speed, the shooting parameters and shooting mode of the main image sensor corresponding to the high-resolution perception layer are determined; Based on the shooting parameters and shooting mode, the high-value target is photographed to obtain high-definition video.

[0087] The continuous frame images refer to a series of independent still images acquired at high speed and continuously in chronological order by the main image sensor of the high-resolution perception layer. The target size refers to the pixel area and outline size of the target within the frame, determined by the high-resolution perception layer after image analysis. The movement speed refers to the rate of change of the target's position per unit time, obtained after tracking and calculating the continuous frame images by the high-resolution perception layer. The shooting parameters refer to the camera hardware operating values ​​set to adapt to the target size and movement speed, mainly including shutter speed, aperture, ISO sensitivity, frame rate, and focus area. The shooting mode refers to a set of pre-set collaborative strategies for different scenarios, mainly including video tracking focus mode, exposure mode, image stabilizer activation, and whether high dynamic range imaging is used.

[0088] Optionally, the target size and movement speed of the high-value target can be determined by a deep learning model, such as Monodepth2 or MiDaS.

[0089] Compared to the problems described in the background technology, this invention acquires thermal signal sequences and grayscale contour images through a low-power sensing layer to extract multi-dimensional biological feature vectors and calculate their matching degree with a non-target feature library. This enables effective filtering under extremely low power consumption. When the matching degree is higher than a preset threshold, a first-level wake-up signal is generated only for potential targets, avoiding frequent false triggers caused by non-target factors, thus minimizing the number of times the device wakes up from deep sleep. This feature-matching-based pre-screening mechanism is the core of the long-lasting battery design, ensuring that the camera maintains a microamplitude standby current for most of the time. This invention activates a low-resolution sensing layer based on the first-level wake-up signal, acquires preliminary infrared images, and extracts image features to determine the target prey confidence level, constituting a second-level intelligent filter. This layer can distinguish between real animal targets and non-living heat sources that cause thermal signals. When the confidence level is not higher than the threshold, it quickly returns to a low-power standby state, avoiding energy waste caused by activating high-power modules. Only when the confidence level exceeds the threshold and it is confirmed as a high-value target is a second-level wake-up signal generated. The hierarchical decision-making mechanism precisely allocates computing resources and energy consumption to valuable shooting events, greatly improving battery efficiency. Upon the generation of the secondary wake-up signal, the camera switches to the high-resolution perception layer for high-definition video recording, ensuring high-quality recording of high-value targets. The entire process, from eliminating non-target features to initial confirmation of potential targets and precise identification of high-value targets, progresses step-by-step, avoiding ineffective shooting and wasted storage space. The storage medium is no longer filled with meaningless empty shots or accidental captures, but rather with high-value, analyzable data, improving data management efficiency and ease of subsequent retrieval. Finally, this multi-level perception architecture, through intelligent hierarchical wake-up and decision-making, achieves an optimal balance between power consumption, performance, and data value, fundamentally solving the core pain points of traditional hunting cameras: short battery life, high false alarm rate, and poor data quality. This enables the device to conduct autonomous, efficient, and accurate monitoring in the wild for months or even longer, providing reliable technical support for wildlife research, security monitoring, and outdoor activities. Therefore, the low-power, long-battery-life hunting camera method provided in this embodiment of the invention can improve the battery life and shooting effect of the hunting camera.

[0090] like Figure 4 The diagram shown is a functional block diagram of a low-power, long-battery-life hunting camera system according to the present invention.

[0091] The low-power, long-endurance hunting camera system 400 described in this invention can be installed in an electronic device. Depending on the functions implemented, the low-power, long-endurance hunting camera system includes a multi-level perception layer construction module 401, a matching degree calculation module 402, a first-level wake-up signal generation module 403, a confidence degree calculation module 404, and a high-resolution imaging module 405. The module described in this invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0092] In this embodiment of the invention, the functions of each module / unit are as follows: The multi-level perception layer construction module 401 is used to construct a multi-level perception layer for the hunting camera. Through the low-power perception layer in the multi-level perception layer, the thermal signal sequence and grayscale contour image of the hunting camera corresponding to the application scenario are collected to extract multi-dimensional biological feature vectors in the application scenario. The matching degree calculation module 402 is used to calculate the matching degree between the multidimensional biological feature vector and the corresponding non-target feature in the non-target feature library in the multi-level perception layer; The first-level wake-up signal generation module 403 is used to determine potential targets in the application scenario and generate a first-level wake-up signal for the hunting camera when the matching degree is lower than a preset threshold. The confidence calculation module 404 is used to activate the low-resolution perception layer of the multi-level perception layer based on the first-level wake-up signal, so as to acquire the preliminary infrared image of the potential target, extract the image features of the preliminary infrared image, and determine the target prey confidence of the potential target. The high-resolution shooting module 405 includes a continuous frame acquisition unit, a target parameter calculation unit, a shooting parameter configuration unit, and a high-definition shooting unit. It is used to return to the low-power perception layer when the target prey confidence level is not higher than a preset confidence threshold, and to identify high-value targets when the target prey confidence level is higher than the preset confidence threshold. It then generates a secondary wake-up signal for the hunting camera to switch to the high-resolution perception layer of the multi-level perception layer, and captures high-definition video of the high-value target based on the high-resolution perception layer.

[0093] In detail, the modules in the low-power, long-battery-life hunting camera system 400 described in this embodiment of the invention employ the same methods as described above during use. Figure 1 The method used is the same as that described in the article on a low-power, long-battery-life hunting camera, and can produce the same technical effect, so it will not be repeated here.

[0094] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0095] Finally, it should be noted that in the above embodiments, each embodiment can be combined with each other or independent. Deleting any one of them will not affect the technical implementation of other embodiments. The above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for using a low-power, long-battery-life hunting camera, characterized in that, The method includes: A multi-level perception layer for a hunting camera is constructed. Through the low-power perception layer in the multi-level perception layer, thermal signal sequences and grayscale contour images of the hunting camera corresponding to the application scenario are collected to extract multi-dimensional biological feature vectors in the application scenario. Calculate the matching degree between the multidimensional biological feature vector and the corresponding non-target features in the non-target feature library of the multi-level perception layer; When the matching degree is lower than a preset threshold, potential targets in the application scenario are identified, and a first-level wake-up signal for the hunting camera is generated. Based on the first-level wake-up signal, the low-resolution perception layer of the multi-level perception layer is activated to acquire a preliminary infrared image of the potential target, and the image features of the preliminary infrared image are extracted to determine the target prey confidence of the potential target. When the confidence level of the target prey is not higher than the preset confidence threshold, the system returns to the low-power perception layer. When the confidence level of the target prey is higher than the preset confidence threshold, the system identifies high-value targets of the potential targets and generates a secondary wake-up signal for the hunting camera to switch to the high-resolution perception layer of the multi-level perception layer. Based on the high-resolution perception layer, the system captures high-definition video of the high-value targets.

2. The low-power, long-battery-life hunting camera method as described in claim 1, characterized in that, The construction of the multi-level perception layer for the hunting camera includes: A multi-level perception framework for the hunting camera is constructed, wherein the multi-level perception framework includes: a low-power perception layer, a low-resolution perception layer, and a high-resolution perception layer. Configure the multi-level sensing framework with multi-level sensors and multi-level processors; Construct a non-target feature library for the multi-level perception framework; Define the multi-level power management states corresponding to the multi-level perception framework, wherein the multi-level power management states include: sleep state, visual verification state, and high-definition recording state; Determine the wake-up rules of the multi-level sensing framework and the state switching logic of the multi-level power management state; The hunting camera's multi-level perception layer is integrated based on the wake-up rules, the state switching logic, the non-target feature library, the multi-level sensors, and the multi-level processor.

3. The low-power, long-battery-life hunting camera method as described in claim 2, characterized in that, The determination of the wake-up rules of the multi-level sensing framework and the state switching logic of the multi-level power management state includes: The reinforcement learning elements of the hunting camera corresponding to the multi-level perception framework are defined, wherein the reinforcement learning elements include: agent, application environment, discrete state, action space and learning reward. Based on the aforementioned reinforcement learning elements, the reinforcement learning algorithm for the hunting camera is determined. Based on the reinforcement learning algorithm, the multi-level wake-up threshold and multi-level state switching parameters of the hunting camera are output. Based on the multi-level wake-up threshold and the multi-level state switching parameters, the wake-up rules of the multi-level perception framework and the state switching logic of the multi-level power management state are determined respectively.

4. The low-power, long-battery-life hunting camera method as described in claim 1, characterized in that, The extraction of multidimensional biological feature vectors in the application scenario includes: The thermal sensing signal sequence corresponding to the application scenario is filtered and denoised to obtain a denoised thermal sensing signal sequence. Based on the denoised thermal sensing signal sequence, the movement continuity characteristics of organisms and the rate of change of hot spot area in the application scenario are extracted. The grayscale contour image corresponding to the application scenario is binarized to obtain a black and white binary image. Based on the black-and-white binary image, calculate the aspect ratio of the target outline and the complexity of the main outline of the organism in the application scenario; The features of the movement continuity, the hot spot area change rate, the target contour aspect ratio, and the main body contour complexity are fused to obtain a multidimensional biological feature vector.

5. The low-power, long-battery-life hunting camera method as described in claim 4, characterized in that, The step of extracting the continuous movement characteristics of organisms and the rate of change of hot spot area in the application scenario based on the denoised thermal sensing signal sequence includes: Identify the dominant probe sequence of the denoised thermal sensing signal sequence; Calculate the number of discontinuous transitions in the dominant probe sequence; Based on the number of discontinuous jumps, calculate the mobility continuity score of the organisms in the application scenario to determine the mobility continuity characteristics of the organisms in the application scenario; Based on the denoised thermal sensing signal sequence, the instantaneous hot spot area sequence of the organism in the application scenario is calculated; Calculate the mean and standard deviation of the instantaneous hot spot area of ​​the instantaneous hot spot area sequence; The rate of change of hot spot area of ​​organisms in the application scenario is calculated based on the mean instantaneous hot spot area and the standard deviation of instantaneous hot spot area.

6. The low-power, long-battery-life hunting camera method as described in claim 1, characterized in that, The calculation of the matching degree between the multidimensional biometric feature vector and the corresponding non-target feature in the non-target feature library of the multi-level perception layer includes: Calculate the prototype weights of the non-target features; Calculate the cosine similarity between the multidimensional biological feature vector and the non-target feature; The matching degree between the multidimensional biometric vector and the non-target feature is calculated based on the prototype weight and the cosine similarity.

7. The low-power, long-battery-life hunting camera method as described in claim 1, characterized in that, The extraction of image features from the preliminary infrared image includes: The preliminary infrared image is denoised to obtain a denoised infrared image; The denoised infrared image is binarized to obtain a binarized infrared image; Morphological processing is performed on the binarized infrared image to obtain a morphological infrared image; The morphological infrared image is segmented for interest to obtain the region of interest; Extract the boundary contours of the region of interest in the image; Based on the boundary contour, calculate the geometric features, statistical features, and texture features of the region of interest; The geometric features, statistical features, and texture features are integrated to obtain image features.

8. The low-power, long-battery-life hunting camera method as described in claim 1, characterized in that, The determination of the target prey confidence of the potential target includes: Define the prey prototype vector and the interference source feature vector of the potential target; Calculate the feature similarity between the prey prototype vector and the image features corresponding to the potential target; Analyze the distribution of interference sources in the feature vectors of the interference sources; Calculate the deviation between the image features and the distribution of the interference sources; The target prey confidence of the potential target is calculated based on the feature similarity and the deviation.

9. The low-power, long-battery-life hunting camera method as described in claim 1, characterized in that, The process of capturing high-definition video of the high-value target based on the high-resolution perception layer includes: Based on the high-resolution perception layer, continuous frame images of the high-value target are acquired; Based on the consecutive frame images, the target size and movement speed of the high-value target are determined; Based on the target's size and movement speed, the shooting parameters and shooting mode of the main image sensor corresponding to the high-resolution perception layer are determined; Based on the shooting parameters and shooting mode, the high-value target is photographed to obtain high-definition video.

10. A low-power, long-battery-life hunting camera system, characterized in that, The system includes: A multi-level perception layer construction module is used to construct a multi-level perception layer for a hunting camera. Through the low-power perception layer in the multi-level perception layer, thermal signal sequences and grayscale contour images of the hunting camera corresponding to the application scenario are collected to extract multi-dimensional biological feature vectors in the application scenario. The matching degree calculation module is used to calculate the matching degree between the multidimensional biological feature vector and the corresponding non-target feature in the non-target feature library of the multi-level perception layer; A first-level wake-up signal generation module is used to determine potential targets in the application scenario and generate a first-level wake-up signal for the hunting camera when the matching degree is lower than a preset threshold. The confidence calculation module is used to activate the low-resolution perception layer of the multi-level perception layer based on the first-level wake-up signal, so as to acquire the preliminary infrared image of the potential target, extract the image features of the preliminary infrared image, and determine the target prey confidence of the potential target. The high-resolution shooting module includes a continuous frame acquisition unit, a target parameter calculation unit, a shooting parameter configuration unit, and a high-definition shooting unit. When the confidence level of the target prey is not higher than a preset confidence threshold, it returns to the low-power perception layer. When the confidence level of the target prey is higher than the preset confidence threshold, it identifies high-value targets among the potential targets and generates a secondary wake-up signal for the hunting camera to switch to the high-resolution perception layer of the multi-level perception layer. Based on the high-resolution perception layer, it captures high-definition video of the high-value target.

Images