Environmental perception system and method for a farm

Abstract

The application relates to the technical field of image analysis, and particularly discloses an environment sensing system for a farm, which comprises a thermal imaging acquisition module, an image recognition module and an analysis module. The thermal imaging acquisition module is used for acquiring a thermal imaging image outside the farm within a first preset time. The image recognition module is used for analyzing the thermal imaging image, identifying and marking first preset animals and second preset animals. The analysis module is used for extracting activity characteristics of the second preset animals when the first preset animals and the second preset animals are simultaneously identified, constructing a first preset animal identifiable score model according to the activity characteristics of the second preset animals, and extracting activity characteristics of the current second preset animals when only the second preset animals are identified, inputting the activity characteristics of the current second preset animals into the first preset animal identifiable score model, and obtaining an output of a first preset animal existence probability. The technical scheme of the application can realize active identification and early warning of dangerous animals such as snakes.

Description

Technical Field

[0001] This invention relates to the field of image analysis technology, and in particular to an environmental sensing system and method for livestock farms. Background Technology

[0002] As the ecological environment gradually improves, the external environment of pig farms is becoming increasingly complex. Feed and excrement within the farms easily attract rodents, which in turn attract snakes. Although pigs possess a certain instinct to hunt snakes, once a snake enters the farm area or moves along its patrol routes, a significant safety hazard remains. This is especially true at night, when snakes are highly concealed and the complex environment outside the farm makes them difficult to detect using conventional methods.

[0003] Currently, while snake farms are generally equipped with video surveillance systems for daily theft prevention, protection, and production management, effectively identifying and issuing early warnings of snake activity remains technically challenging. As poikilothermic animals, snakes' body temperature changes with the ambient temperature. When their body temperature approaches the surrounding environment, traditional thermal imaging monitoring equipment struggles to effectively capture their thermal signature. This is especially true at night, when the air temperature drops and the snake's temperature approaches the ambient temperature, or when snakes hide in complex terrain such as grass or rock piles. Neither visible light cameras nor thermal imaging devices can directly and accurately detect their tracks, rendering security measures ineffective.

[0004] Therefore, how to effectively utilize and upgrade the existing monitoring system and develop a technical solution that can overcome adverse conditions such as nighttime and obstruction to achieve intelligent and proactive identification and early warning of dangerous animals such as snakes has become an urgent technical problem to be solved in the field of farm safety management. Summary of the Invention

[0005] One of the objectives of this invention is to provide an environmental sensing system for breeding farms, which can actively identify and warn of dangerous animals such as snakes.

[0006] To solve the above-mentioned technical problems, this application provides the following technical solution:

[0007] Environmental sensing systems for farms include:

[0008] The thermal imaging acquisition module is used to acquire thermal images of the area outside the farm within a first preset time.

[0009] The image recognition module is used to analyze thermal imaging images, identify and mark a first preset animal and a second preset animal;

[0010] The analysis module is used to extract the activity characteristics of the second preset animal when the first preset animal and the second preset animal are identified simultaneously, and to construct an identification scoring model for the first preset animal based on the activity characteristics of the second preset animal.

[0011] It is also used to extract the activity features of the current second preset animal when only the second preset animal is identified, input the activity features of the current second preset animal into the first preset animal identifiability scoring model, and obtain the output of the probability of the existence of the first preset animal;

[0012] The first preset animal is a snake, and the second preset animal is a mouse.

[0013] Furthermore, it also includes a visible light acquisition module for acquiring visible light images outside the farm within a second preset time period; and an image recognition module for analyzing the visible light images, identifying and marking the first preset animal and the second preset animal.

[0014] It is also used to identify the species of a first-preset animal;

[0015] It is also used to identify terrain features within an image range.

[0016] Furthermore, the activity characteristics include time, ambient temperature data, number of mice, average thermal signal intensity of mice, average background thermal signal intensity, duration of mouse activity, activity coverage of mice, and terrain features corresponding to the activity range of mice.

[0017] Furthermore, the analysis module is also used to construct an activity feature training dataset through activity features; and to train a first preset animal identifiability scoring model based on the second preset animal activity feature training dataset.

[0018] Furthermore, the first preset animal identifiability scoring model adopts a logistic regression model. During construction, the activity feature training dataset is concentrated, and the activity feature samples that simultaneously identify the first preset animal and the second preset animal are marked as positive samples, while the samples that only identify the second preset animal are marked as negative samples. The activity features of each sample are used as the input feature vector of the model, and the weight parameters of the logistic regression model are trained through supervised learning.

[0019] Furthermore, it also includes an information notification module, which generates animal warning information when the number of mice exceeds a number threshold, a snake is detected, or the probability of a snake being present is greater than a probability threshold.

[0020] Furthermore, the thermal imaging acquisition module is used to scan a preset area outside the farm at a preset rate within a first preset time and acquire thermal imaging images.

[0021] It also includes a sound acquisition module and a sound recognition module. The sound acquisition module is used to collect sound data in the farm; the sound recognition module is used to analyze whether the sound data is the sound of pigs. If so, it determines whether the decibel level exceeds the decibel threshold. If it does, it increases the scanning rate of the thermal imaging acquisition module.

[0022] Furthermore, the information notification module is also used to record the time when a user views animal warning information;

[0023] The analysis module is also used to record the location of the mouse when a snake appears, and when the probability of the snake's presence is greater than a probability threshold, and mark these locations as dangerous locations.

[0024] The image recognition module is also used to analyze thermal imaging images to identify whether a person is present. If a person is present, it determines the distance between the person and the dangerous location, as well as the speed at which the person approaches the dangerous location.

[0025] The information notification module is also used to play an audio alert when the current time and the viewing time are less than 1 hour apart, and when the distance between a person and the dangerous location is less than a distance threshold.

[0026] It is also used to continuously play sound alerts when the current time and the viewing time are greater than or equal to 1 hour, and the distance between a person and the dangerous location is less than a distance threshold. The closer the person is to the dangerous location, the louder the sound.

[0027] A second objective of this invention is to provide an environmental sensing method for aquaculture farms, comprising the following steps:

[0028] S1. Acquire thermal imaging images of the area outside the farm within a first preset time period;

[0029] S2. Acquire visible light images outside the farm within the second preset time period;

[0030] S3. Analyze the thermal imaging image to identify and mark the first and second preset animals; and analyze the visible light image to identify and mark the first and second preset animals.

[0031] S4. Identify the terrain features within the image area;

[0032] S5. Extract the activity features of the second preset animal when both the first and second preset animals are identified simultaneously, and also extract the activity features when the second preset animal appears alone, and construct an activity feature training dataset; train the first preset animal identifiability scoring model based on the activity feature training dataset of the second preset animal.

[0033] S6. When only the second preset animal is identified, extract the activity features of the current second preset animal, input the activity features of the current second preset animal into the first preset animal identifiability scoring model, and obtain the output of the probability of the existence of the first preset animal.

[0034] S7. Generate animal warning information when the number of mice exceeds the number threshold, a snake is detected, or the probability of a snake's presence is greater than the probability threshold.

[0035] Furthermore, the first preset animal is a snake, and the second preset animal is a mouse;

[0036] Landform features include mud, piles of rocks, shrubs, waterlogged areas, crop fields, sandy soil, garbage dumps, brick and stone roads, and cement ground;

[0037] Activity characteristics include time, ambient temperature data, number of mice, average thermal signal intensity of mice, average thermal signal intensity of background, duration of mouse activity, coverage area of ​​mouse activity, and geomorphological features corresponding to the coverage area of ​​mouse activity.

[0038] This solution overcomes the technical bottleneck of traditional monitoring methods, which struggle to directly detect cold-blooded dangerous animals like snakes in complex farm environments. By innovatively establishing a correlation model between snake and rodent activity characteristics, indirect prediction is possible. Even under unfavorable conditions such as nighttime, obstruction by vegetation, or snake body temperature similar to the ambient temperature, the probability of snake presence can be accurately assessed by analyzing specific rodent activity patterns and leveraging the obvious and easily detectable infrared signatures of rodents. This enables effective perception of concealed threats, significantly improving the accuracy and reliability of early warning systems.

[0039] This solution can not only issue an alarm when the number of snakes or rats exceeds the limit, but also provide probabilistic risk warnings based on models, enabling farm managers to know potential danger areas in advance and avoid accidental contact with snakes during inspections or operations, thereby effectively protecting the lives of personnel.

[0040] This solution can also make efficient use of the existing monitoring equipment in the farm, and upgrade and transform the monitoring system at a low cost, realizing all-weather automated monitoring and early warning, and reducing the pressure and risk of manual patrols at night. Attached Figure Description

[0041] Figure 1 This is a logic block diagram of an embodiment of an environmental sensing system for a farm.

[0042] Figure 2 This is a flowchart of an embodiment of an environmental sensing method for a livestock farm. Detailed Implementation

[0043] The following detailed description illustrates the specific implementation method:

[0044] Example 1

[0045] like Figure 1 As shown, the environmental sensing system for aquaculture farms in this embodiment includes a thermal imaging acquisition module, an image recognition module, an analysis module, a visible light acquisition module, and an information notification module.

[0046] The thermal imaging acquisition module is used to acquire thermal images outside the farm within a first preset time period; in this embodiment, the thermal imaging acquisition module adopts a thermal imaging observation tube.

[0047] The visible light acquisition module is used to acquire visible light images outside the farm within a second preset time period. In this embodiment, the first preset time period is from sunset to sunrise, i.e., nighttime, and the remaining time period of the second preset time period is daytime. In other embodiments, thermal imaging images and visible light images can be switched according to the ambient brightness.

[0048] The image recognition module is used to analyze thermal imaging images, identify and mark the first and second preset animals;

[0049] It is also used to analyze visible light images, identify and label a first preset animal and a second preset animal;

[0050] It is also used to identify the species of the first preset animal; to know in advance whether the snakes appearing in the local area are venomous or non-venomous, and to help quickly identify the species and match the corresponding antivenom serum when a snake bite occurs.

[0051] It is also used to identify terrain features within the image range. In this embodiment, terrain features include mud, piles of stones, bushes, waterlogged areas, crop fields, sandy soil, garbage dumps, brick and stone roads, cement ground, etc. In other embodiments, there may be other terrain features that cover common ground and environmental elements around the farm.

[0052] The first preset animal is a snake, and the second preset animal is a mouse.

[0053] The analysis module is used to extract the activity features of the second preset animal when both the first and second preset animals are identified simultaneously, and also to extract the activity features of the second preset animal when it appears alone, and to construct an activity feature training dataset; the identification scoring model of the first preset animal is trained based on the activity feature training dataset of the second preset animal.

[0054] Activity characteristics include time, ambient temperature data, number of mice, average thermal signal intensity of mice, average thermal signal intensity of background, duration of mouse activity, coverage area of ​​mouse activity, and geomorphological features corresponding to the coverage area of ​​mouse activity.

[0055] In this embodiment, when extracting activity features, the number of mice is obtained by counting using a target detection algorithm; the average thermal signal intensity of the mice and the average thermal signal intensity of the background are obtained by calculating the average pixel values ​​of the target area and the background area in the thermal imaging image, and the difference between the two is calculated. This difference reflects the salience of the thermal signal when the mice are active; the activity coverage of the mice is quantified by calculating the bounding box area of ​​the mouse's movement trajectory in consecutive image frames; and the terrain features are obtained by identifying the range coordinates through visible light images and then applied to the thermal imaging image.

[0056] This embodiment employs a logistic regression algorithm to construct a first-preset animal identifiability scoring model. Specifically, all activity features are normalized to ensure comparability of features with different dimensions. A stepwise regression method is used to filter out feature items with significantly different values. Among these, time, ambient temperature data, number of mice, duration of mouse activity, and average background thermal signal intensity are the main influencing factors; the average thermal signal intensity of mice, the activity coverage area of ​​mice, and the terrain features corresponding to the activity coverage area of ​​mice are the secondary influencing factors. An activity feature training dataset is then constructed.

[0057] In the activity feature training dataset, samples that simultaneously identify both a first and a second preset animal are labeled as positive samples, while samples that only identify the second preset animal are labeled as negative samples. The activity features of each sample are used as the input feature vector of the model. Through supervised learning, the weight parameters of the logistic regression model are trained, enabling the model to maximize the differentiation between positive and negative samples based on the input activity features. In the actual collected data, negative samples may exist where only mice are present in the environment, yet the image identifies the mice; conversely, there may be cases where both mice and snakes are present in the environment, but the image only identifies the mice. To improve the accuracy of the negative sample data source, this embodiment also employs manual verification to confirm the thermal imaging images. To improve efficiency, manual verification can be performed only on thermal imaging images of certain error-prone terrain features, such as piles of rocks and bushes. In other embodiments, during the model training data acquisition phase, at least two thermal imaging observation tube cameras can be deployed to collect thermal imaging images from multiple angles for error-prone terrain features, allowing for double or multiple verifications.

[0058] The analysis module is also used to extract the activity features of the current second preset animal when only the second preset animal is identified, input the activity features of the current second preset animal into the first preset animal identifiability scoring model, and obtain the output of the probability of the existence of the first preset animal.

[0059] Specifically, the activity characteristics of the current second preset animal are input into the identifiability scoring model of the first preset animal:

[0060]

[0061] in, These are the weights obtained during training. It is an eigenvalue. It is a bias term.

[0062] The probability value P of the existence of the first pre-defined animal is calculated using the sigmoid function:

[0063]

[0064] The probability threshold can be set according to the actual application. In this embodiment, a snake is considered to be possible when the probability value is greater than 0.7.

[0065] The information notification module generates animal warning messages when the number of mice exceeds a threshold, a snake is detected, or the probability of a snake's presence exceeds a threshold. The animal warning message includes the warning type, time, specific details, and an image screenshot. For example: Snake warning, time: October 15, 2024, 02:30, snake activity was found on the east side of the farm; another example: Mouse warning, time: October 15, 2024, 11:23, the number of mice exceeded the limit, with 5 mice detected.

[0066] like Figure 2 As shown, based on the above system, this embodiment also provides an environmental sensing method for a farm, including the following steps:

[0067] S1. Acquire thermal imaging images of the area outside the farm within a first preset time period;

[0068] S2. Acquire visible light images outside the farm within the second preset time period;

[0069] S3. Analyze the thermal imaging image to identify and mark the first and second preset animals; and analyze the visible light image to identify and mark the first and second preset animals; and also identify the species of the first preset animal; wherein the first preset animal is a snake and the second preset animal is a mouse.

[0070] S4. Identify the terrain features within the image range, including mud, piles of rocks, bushes, waterlogged areas, crop fields, sandy soil, garbage dumps, brick and stone roads, cement ground, etc.

[0071] S5. Extract the activity features of the second preset animal when both the first and second preset animals are identified simultaneously, and also extract the activity features when the second preset animal appears alone, and construct an activity feature training dataset; train the first preset animal identifiability scoring model based on the activity feature training dataset of the second preset animal; the activity features include time, ambient temperature data, number of mice, average thermal signal intensity of mice, average thermal signal intensity of background, duration of mouse activity, activity coverage of mice, and terrain features corresponding to the activity coverage of mice.

[0072] S6. When only the second preset animal is identified, extract the activity features of the current second preset animal, input the activity features of the current second preset animal into the first preset animal identifiability scoring model, and obtain the output of the probability of the existence of the first preset animal.

[0073] Specifically, the activity characteristics of the current second preset animal are input into the identifiability scoring model of the first preset animal:

[0074]

[0075] in, These are the weights obtained during training. It is an eigenvalue. It is a bias term.

[0076] The probability value P of the existence of the first pre-defined animal is calculated using the sigmoid function:

[0077]

[0078] The probability threshold can be set according to the actual application. In this embodiment, a snake is considered to be possible when the probability value is greater than 0.7.

[0079] S7. Generate animal warning information when the number of mice exceeds the number threshold, a snake is detected, or the probability of a snake's presence is greater than the probability threshold.

[0080] This embodiment overcomes the technical bottleneck of traditional monitoring methods, which struggle to directly detect cold-blooded dangerous animals like snakes in complex farm environments. By innovatively establishing a correlation model between snake and mouse activity characteristics, it achieves a shift from direct detection to indirect prediction. Even under unfavorable conditions such as nighttime, obstruction by grass, or snake body temperature similar to the ambient temperature, the system can accurately assess the probability of snake presence by analyzing specific mouse activity patterns and leveraging the obvious and easily detectable infrared signatures of mice. This enables effective perception of hidden threats, significantly improving the accuracy and reliability of early warning systems.

[0081] This solution can not only issue an alarm when the number of snakes or rats exceeds the limit, but also provide probabilistic risk warnings based on models. This allows farm managers to be aware of potential danger areas in advance and avoid accidental contact with snakes during inspections or operations, thereby effectively protecting the lives of personnel and farmed animals. It solves the problem that the existing monitoring system has limited functions and cannot perform intelligent risk assessment.

[0082] This solution can also make efficient use of the existing monitoring equipment in the farm, and upgrade and transform the monitoring system at a low cost, realizing all-weather automated monitoring and early warning, and reducing the pressure and risk of manual patrols at night.

[0083] Example 2

[0084] The difference between this embodiment and Embodiment 1 is that the thermal imaging acquisition module in this embodiment uses a thermal imager with a pan-tilt unit to scan a preset area outside the farm at a preset rate within a first preset time and acquire thermal images. Scanning by rotating the pan-tilt unit allows for coverage of a wider area.

[0085] It also includes a sound acquisition module and a sound recognition module.

[0086] The sound acquisition module uses a microphone to collect sound data within the farm.

[0087] The sound recognition module is used to analyze whether the sound data is the sound of pigs. In this embodiment, when analyzing the sound data, the collected sound data is first preprocessed such as denoising, and sound features are extracted. The extracted sound features are then input into a pre-trained voiceprint recognition model to obtain the output result of whether the sound category is the sound of pigs.

[0088] If it is the sound of pigs squealing, it is also used to determine whether the decibel level exceeds the decibel threshold. If it does, the scanning rate of the thermal imaging acquisition module is increased.

[0089] In this embodiment, during the normal activity period of the pigs within a first preset time period, such as a resting period without feeding or significant external interference, sound data is collected for several consecutive days, for example, 7 days. The average decibel value is calculated as a baseline decibel value, and 10 dB is added to this as a decibel threshold. For example, if the measured baseline decibel value is 65 dB, the decibel threshold is set to 65 + 10 = 75 dB. The decibel threshold is also recalculated and calibrated periodically to adapt to changes in the pig herd size or seasonal environmental noise variations. In other embodiments, a manually set fixed value can also be used.

[0090] In this embodiment, the gimbal rotation strategy is as follows: First, it rotates continuously at an initial rate to cover a set rotation angle (the set rotation angle is determined based on the device's installation location and the area to be monitored; for example, if the device is against a farm fence, the rotation angle is set between 30° and 175°). Then, it returns to the initial position and performs interval rotation. That is, it first takes an image, then rotates at a set interval (e.g., 5 seconds) before taking a second image at a fixed angle, repeating this process until the set rotation angle is completely covered. Then, it rotates back to the initial position from the current position, repeating this process. Adjusting the scanning rate, i.e., adjusting the set time, shortens the set time (e.g., 1 second) to increase the scanning rate. Using interval rotation, the device remains stationary during image capture, avoiding motion blur in the image frames. After only mice are detected, the center of the image is determined based on the position of each mouse, and the rotation angle is readjusted accordingly. After a snake is detected, the rotation angle is readjusted to center the snake in the image.

[0091] This embodiment introduces a sound acquisition and recognition module to construct a three-dimensional early warning system that links the internal and external systems, achieving an upgrade from passive scanning to proactive response. The system no longer relies solely on external visual information but uses stress responses of pigs within the farm (such as high-decibel squeals) as trigger signals for high-risk events. Once an anomaly is detected in the pig herd, the system can instantly increase the scanning rate of the external thermal imaging module, accelerating the search for and location of potential threats and significantly improving the real-time performance of dangerous animal early warnings.

[0092] Example 3

[0093] The difference between this embodiment and embodiment two is that, in this embodiment, the information notification module is also used to record the time when the user views the animal warning information.

[0094] The analysis module is also used to record the location of the mouse when a snake appears, and when the probability of the snake's presence is greater than a probability threshold, and mark these locations as dangerous locations.

[0095] The image recognition module is also used to analyze thermal imaging images to identify whether a person is present. If a person is present, it determines the distance between the person and the dangerous location, as well as the speed at which the person approaches the dangerous location.

[0096] The information notification module is also used to play an audio alert when the current time and the viewing time are less than 1 hour apart, and when the distance between a person and the dangerous location is less than a distance threshold.

[0097] It is also used to continuously play an audio alert when the current time and the viewing time are greater than or equal to 1 hour, and the distance between the person and the dangerous location is less than a distance threshold. The closer the person is to the dangerous location, the louder the sound. It is also used to stop playing the audio alert when the person stops moving. After the person resumes movement, if the resumed movement speed is significantly lower than the movement speed before stopping, the audio alert remains stopped. If the resumed movement speed is similar to the movement speed before stopping, the audio alert continues to play. In this embodiment, "significantly lower than" means less than 10% of the average speed before stopping; "similar" means the resumed movement speed is within 10% of the average movement speed before stopping.

[0098] In other embodiments, the on / off state of the sound alert can also be controlled by recognizing whether the person is looking at the camera after they stop moving. For example, if the person is looking at the camera, the sound continues to play; if they are not looking at the camera, the sound stops playing.

[0099] This embodiment further realizes the intelligent linkage between early warning information and personnel status. By recording the time when users view the early warning information and combining it with real-time recognition of personnel location, movement speed and direction, it effectively distinguishes between two behavioral states: active inspection and unintentional approach, thereby dynamically adjusting the reminder strategy.

[0100] When it is determined that the personnel may be knowingly and intentionally approaching a dangerous location to handle the situation, the warnings should be reduced or kept silent to avoid alerting the suspect or interfering with the work of the staff. When it is determined that the personnel may be unintentionally approaching the danger under unknown circumstances, the warnings should be proactively issued through gradually increasing and continuous auditory warnings to effectively prevent accidents from occurring and ensure personnel safety.

[0101] The above are merely embodiments of the present invention. The invention is not limited to the fields covered by these embodiments. Commonly known structures and characteristics in the solutions are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are able to access all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. An environmental sensing system for a livestock farm, characterized in that, include: The thermal imaging acquisition module is used to acquire thermal images of the area outside the farm within a first preset time. The image recognition module is used to analyze thermal imaging images, identify and mark a first preset animal and a second preset animal; The analysis module is used to extract the activity characteristics of the second preset animal when the first preset animal and the second preset animal are identified simultaneously, and to construct an identification scoring model for the first preset animal based on the activity characteristics of the second preset animal. It is also used to extract the activity features of the current second preset animal when only the second preset animal is identified, input the activity features of the current second preset animal into the first preset animal identifiability scoring model, and obtain the output of the probability of the existence of the first preset animal; The first preset animal is a snake, and the second preset animal is a mouse.

2. The environmental sensing system for a farm according to claim 1, characterized in that: It also includes a visible light acquisition module for acquiring visible light images outside the farm within a second preset time period; and an image recognition module for analyzing the visible light images, identifying and marking the first and second preset animals. It is also used to identify the species of a first-preset animal; It is also used to identify terrain features within an image range.

3. The environmental sensing system for a farm according to claim 2, characterized in that: The activity characteristics include time, ambient temperature data, number of mice, average thermal signal intensity of mice, average thermal signal intensity of background, duration of mouse activity, coverage area of ​​mouse activity, and terrain features corresponding to the mouse activity area.

4. The environmental sensing system for a farm according to claim 3, characterized in that: The analysis module is also used to construct an activity feature training dataset based on activity features; and to train a first preset animal identifiability scoring model based on the second preset animal activity feature training dataset.

5. The environmental sensing system for a farm according to claim 4, characterized in that: The first preset animal identifiability scoring model adopts a logistic regression model. During construction, the activity feature training dataset is concentrated. The activity feature samples that simultaneously identify the first preset animal and the second preset animal are marked as positive samples, and the samples that only identify the second preset animal are marked as negative samples. The activity features of each sample are used as the input feature vector of the model. The weight parameters of the logistic regression model are trained through supervised learning.

6. The environmental sensing system for a farm according to claim 5, characterized in that: It also includes an information notification module, which generates animal warning information when the number of mice exceeds a threshold, a snake is detected, or the probability of a snake being present is greater than a probability threshold.

7. The environmental sensing system for a farm according to claim 6, characterized in that: The thermal imaging acquisition module is used to scan a preset area outside the farm at a preset rate within a first preset time and acquire thermal imaging images. It also includes a sound acquisition module and a sound recognition module. The sound acquisition module is used to collect sound data within the farm. The sound recognition module is used to analyze whether the sound data is the sound of pigs. If so, it determines whether the decibel level exceeds the decibel threshold. If it does, it increases the scanning rate of the thermal imaging acquisition module.

8. The environmental sensing system for a farm according to claim 7, characterized in that: The information notification module is also used to record the time when a user views animal warning information; The analysis module is also used to record the location of the mouse when a snake appears, and when the probability of the snake's presence is greater than a probability threshold, and mark these locations as dangerous locations. The image recognition module is also used to analyze thermal imaging images to identify whether a person is present. If a person is present, it determines the distance between the person and the dangerous location, as well as the speed at which the person approaches the dangerous location. The information notification module is also used to play an audio alert when the current time and the viewing time are less than 1 hour apart, and when the distance between a person and the dangerous location is less than a distance threshold. It is also used to continuously play sound alerts when the current time and the viewing time are greater than or equal to 1 hour, and the distance between a person and the dangerous location is less than a distance threshold. The closer the person is to the dangerous location, the louder the sound.

9. An environmental sensing method for a livestock farm, using the system described in any one of claims 2-8, characterized in that, Includes the following steps: S1. Acquire thermal imaging images of the area outside the farm within a first preset time period; S2. Acquire visible light images outside the farm within the second preset time period; S3. Analyze the thermal imaging image to identify and mark the first and second preset animals; and analyze the visible light image to identify and mark the first and second preset animals. S4. Identify the terrain features within the image area; S5. Extract the activity features of the second preset animal when both the first and second preset animals are identified simultaneously, and also extract the activity features when the second preset animal appears alone, and construct an activity feature training dataset; train the first preset animal identifiability scoring model based on the activity feature training dataset of the second preset animal. S6. When only the second preset animal is identified, extract the activity features of the current second preset animal, input the activity features of the current second preset animal into the first preset animal identifiability scoring model, and obtain the output of the probability of the existence of the first preset animal. S7. Generate animal warning information when the number of mice exceeds the number threshold, a snake is detected, or the probability of a snake's presence is greater than the probability threshold.

10. The environmental sensing method for a livestock farm according to claim 9, characterized in that: The first preset animal is a snake, and the second preset animal is a mouse; Landform features include mud, piles of rocks, shrubs, waterlogged areas, crop fields, sandy soil, garbage dumps, brick and stone roads, and cement ground; Activity characteristics include time, ambient temperature data, number of mice, average thermal signal intensity of mice, average thermal signal intensity of background, duration of mouse activity, coverage area of ​​mouse activity, and geomorphological features corresponding to the coverage area of ​​mouse activity.

Images