Track robot lens dirt detection method and system based on time sequence learning

Abstract

The invention discloses a track robot lens smudginess detection method and system based on time sequence learning, and belongs to the technical field of breeding. A track robot lens smudginess detection system based on time sequence learning is characterized by comprising a data acquisition module arranged at a pig raising area; the advancing track is arranged in a pig raising area, and the robot can move along the advancing track and is used for installing a data acquisition module. According to the track robot lens smudginess detection method and system based on time sequence learning, through mutual cooperation of the time sequence feature preprocessing module, the time sequence feature extraction module and other components, images at the shot position of the lens can be arranged according to time, the images are analyzed for a long time period, and the detection accuracy is improved. According to the method, the smudginess area on the lens is divided into the temporary smudginess image and the persistent smudginess image, and after the persistent smudginess is judged, the staff is notified to intervene, so that the false alarm rate can be reduced, the intervention demand of the staff is reduced, and the burden of the staff is reduced.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of breeding, and particularly relates to a track robot lens dirt detection method and system based on time sequence learning. BACKGROUND

[0002] In a modernized breeding farm, as an important intelligent device, a track robot needs to be equipped with an observation camera to monitor the appearance, behavior, body shape, back fat and pig distribution of pigs. However, the observation camera is easily polluted by dirt in the pig raising area during use, which causes the images captured by the observation camera to be unclear.

[0003] The commonly used lens cleaning detection method in the industry is as follows: A detection method based on single-time image quality evaluation, which determines whether the lens is dirty by checking the dirt in the photos by intercepting multiple individual segments in the image and taking multiple individual photos. This method determines whether the lens is dirty by using a single-time image. However, the following problems exist in the use of this method: It cannot distinguish between temporary dirt such as water droplet, water stain, fog pollution and temporary shielding pollution on the lens and persistent dirt, which leads to a deviation between the determination result and the actual situation. In actual use, the false positive rate of the above method is high, which requires frequent dispatch of workers and causes great work pressure on the workers. Therefore, the application proposes a track robot lens dirt detection method and system based on time sequence learning. SUMMARY

[0004] The application aims to provide a track robot lens dirt detection method and system based on time sequence learning to solve the problem that the existing detection method based on single-time image quality evaluation cannot distinguish between dirt on the lens and cannot clearly know whether the dirt on the lens is temporary or persistent.

[0005] To achieve the above-mentioned purpose, the application provides the following technical solution, a track robot lens dirt detection system based on time sequence learning, characterized by comprising: A data acquisition module arranged at the pig raising area, wherein the data acquisition module comprises an observation camera; A running track arranged at the pig raising area and a robot capable of moving along the running track and used for installing the data acquisition module; A time sequence feature preprocessing module installed on the robot, comprising an image time calibration module, an image recognition module, an image interception module and a data acquisition processing module; Further comprising a time sequence feature extraction module; the time sequence feature extraction module comprises an image extraction module, a data extraction module, and an image analysis module; The time sequence feature extraction module is used for comparing and analyzing the unclear images identified and intercepted by the time sequence feature preprocessing module, judging the causes of the unclear images, and classifying the unclear images into temporary dirty images and persistent dirty images according to the causes of the unclear images. A control module; A power module; the power module is used for providing power for the data acquisition module, the time sequence feature preprocessing module, the time sequence feature extraction module, and the control module.

[0006] Preferably, the data acquisition module comprises an observation camera, a lens light intensity sensor used for recording the light intensity at the lens of the observation camera, a temperature sensor used for measuring the temperature around the observation camera, a humidity sensor used for measuring the humidity around the observation camera, an ambient light intensity sensor used for measuring the light intensity in the pig raising area, and a timer used for recording time; the lens light intensity sensor, the temperature sensor, the humidity sensor, and the timer are all installed on the robot, and the ambient light intensity sensor is arranged in the pig raising area.

[0007] Preferably, the image time calibration module is used for calibrating the time length of the image captured by the observation camera with the time recorded by the timer, and replacing the time axis of the image captured by the observation camera with the time recorded by the timer.

[0008] Preferably, the data processing module arranges the data in sequence according to the time recorded by the timer, to form a lens light intensity value curve, a temperature value curve, a humidity value curve, and an ambient light intensity value curve.

[0009] Preferably, the image recognition module is used for recognizing the images captured by the observation camera when the robot moves along the track, and classifying the images into clear images and unclear images according to the image clarity; the image interception module is used for intercepting the unclear images identified by the image recognition module, to obtain the unclear images; and the data processing module is used for processing the data collected by the data acquisition module.

[0010] A lens dirty detection method for a track robot based on time sequence learning, which adopts a lens dirty detection system for a track robot based on time sequence learning, and comprises the following steps when in use: S1, data acquisition, collecting data through a data acquisition module; S2, time sequence feature preprocessing; S201, image time calibration, arranging the images captured by an observation camera in sequence according to the time recorded by a timer; S202, unclear image screening, through the image recognition module, the image captured by the observation camera is identified according to the clarity of the image, and the image is divided into unclear image, clear image; S203, data processing; through the acquisition data processing module, the data collected by the data acquisition module is received and processed; S204, data extraction, through the acquisition data processing module, the data collected by the data acquisition module is collected according to the time period in which the unclear image is intercepted by the image interception module; S3, time sequence feature extraction, according to the reason of unclear image formation, the unclear image is divided into temporary dirty image and persistent dirty image; S301, analyze the reason of unclear image generation, and classify the unclear image caused by water droplet water stain pollution, fog pollution and temporary shielding pollution as temporary dirty image; S302, classify the reason of unclear image caused by non-temporary dirty as persistent dirty.

[0011] Preferably, the method of collecting data by the data acquisition module in S1 is that the observation camera moves along the track with the robot, and the image of the pig in the pig raising area is shot; Lens light sensor is used to collect the light intensity data around the lens of observation camera; Temperature sensor is used to collect temperature data around the lens of observation camera; Humidity sensor is used to collect humidity data around the lens of observation camera; Ambient light intensity sensor is used to measure the light intensity data in the pig raising area.

[0012] Preferably, the method of receiving and processing the data collected by the data acquisition module through the acquisition data processing module in S203 is that the acquisition data processing module arranges the data collected by the lens light sensor, temperature sensor, humidity sensor and lens light intensity sensor in sequence according to the time recorded by the timer, and obtains the lens light intensity value curve, temperature value curve, humidity value curve and ambient light intensity value curve.

[0013] Preferably, the method of determining the unclear image caused by water droplet water stain pollution, fog pollution and temporary shielding pollution in S301 is: Time sequence feature extraction module, scanning image, can obtain the value of image clarity and process the value, in the collected image, the value of image clarity gradually increases, and the lens light intensity value curve in the same time period of the collected image is in high value or gradually increases, and the image is classified as water droplet water stain pollution; The temporal feature extraction module compares and analyzes the captured unclear image with the data collected by the data acquisition module for the corresponding time period of the captured unclear image. In the captured unclear image, the image clarity value gradually increases, the humidity value curve shows that the humidity value is in a high state, and the value in the lens light intensity value curve is gradually increasing or in a high state. The image segment is classified as fog pollution. The temporal feature extraction module compares and analyzes the captured unclear images with the data collected by the data acquisition module for the corresponding time period of the captured unclear images. Among the captured unclear images, those whose duration of unclear images is within the set time length are classified as unclear images caused by temporary occlusion pollution.

[0014] Preferably, unclear images caused by unremovable obstructions are classified as meaningless images: two sensor switches are set in front of and behind the obstruction. When the track robot moves to the sensor switches, the control module records the activation time of the two sensor switches. The control module marks the image during the time period from the activation of the first sensor switch to the activation of the second sensor switch as meaningless image.

[0015] Beneficial effects: I. The present invention provides a method and system for detecting dirt in the lens of a track robot based on temporal learning. The temporal feature extraction module scans unclear images screened by temporal feature preprocessing, determines the cause of the unclear images, classifies unclear images caused by water droplet contamination, fog contamination, or temporary occlusion contamination as temporary dirt images, and classifies images not caused by temporary dirt as permanent dirt images. After determining that an image is a permanent dirt image, the system notifies the staff to intervene. This can reduce the false alarm rate, reduce the amount of staff intervention required, and reduce the burden on the staff.

[0016] II. The present invention provides a method and system for detecting dirt in the lens of an orbital robot based on temporal learning. The temporal feature extraction module classifies images where the image clarity value gradually increases and the lens illumination intensity value curve is at a high or gradually increasing state within the same time period as the image acquisition as water droplet / water stain pollution. The temporal feature extraction module also classifies images where the image clarity value gradually increases, the humidity value curve is at a high level, and the lens illumination intensity value curve is gradually increasing or at a high value as fog pollution. Furthermore, the temporal feature extraction module compares and analyzes the captured unclear images with the data collected by the data acquisition module during the corresponding time period. If the captured unclear images contain rapidly appearing and rapidly disappearing obstructions, the images are classified as temporary obstruction pollution. This paper categorizes temporary contamination into methods for identifying water droplet and water stain contamination, fog contamination, and temporary obstruction contamination. By identifying and judging temporary dirt, it is easier to record and analyze the quantity and frequency of temporary dirt, and assist staff in understanding the condition of lens dirt and judging the type of dirt.

[0017] Third, the present invention provides a method and system for detecting dirt on the lens of an orbital robot based on time-series learning. If it is determined that there is temporary obstruction and contamination on the lens of the observation camera, it is necessary to remove the obstruction at the time period corresponding to the trajectory of the orbital robot. If it cannot be removed, the image of the time period in which the orbital robot travels near the temporary obstruction can be marked as meaningless image by the time-series feature extraction module, so as to avoid performing useless identification and differentiation work and reduce the amount of abnormal error reports. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the lens contamination system for an orbital robot based on time-series learning in this invention; Figure 2 This is a flowchart illustrating the steps of the time-series learning-based lens contamination detection method for orbital robots in this invention. Figure 3 This is a schematic diagram of step S2 in the time-series learning-based lens dirt detection method for orbital robots in this invention; Figure 4 This is a schematic diagram of step S3 in the time-series learning-based lens dirt detection method for orbital robots in this invention; Figure 5 This is a flowchart of the image sieving process acquired by the observation camera in this invention; Figure 6 This is a flowchart of the process for screening meaningless images in the images acquired by the observation camera in this invention. Detailed Implementation

[0019] The specific embodiments of the present invention will be described in detail below, but it should be understood that the scope of protection of the present invention is not limited to the specific embodiments.

[0020] like Figure 1 As shown in the figure, the present invention provides a time-series learning-based lens contamination detection system for orbital robots, comprising: The data acquisition module is set up in the pig farming area. The data acquisition module includes an observation camera, a lens light intensity sensor for recording the light intensity at the lens of the observation camera, a temperature sensor for measuring the temperature around the observation camera, a humidity sensor for measuring the humidity around the observation camera, an ambient light intensity sensor for measuring the light intensity in the pig farming area, and a timer for recording time (an observation camera with its own time recording function can also be used). The pig farming area is equipped with a travel track and robots that can move along the track. Observation cameras are mounted on the robots and move with them. During the movement, the observation cameras can record videos of the pigs on the travel path.

[0021] Currently, there are several ways to use observation cameras to observe the number of pigs, one of which is: Target detection: The observation camera can be the TS-CM01 pig and cattle face recognition camera. The TS-CM01 pig and cattle face recognition camera identifies and locates pigs through a pig face recognition device. It adopts an integrated design, and the supporting equipment can identify, locate, and number pigs, and enclose them with bounding boxes, and calculate the total number of pigs.

[0022] In a specific embodiment, the observation camera calculates the total number of pigs by superimposing the observation camera and all numbered pigs along the entire route the robot travels along the track.

[0023] In this way, the goal of counting pigs was achieved by observing the images recorded by the camera.

[0024] In a specific embodiment, the lens illumination intensity sensor is mounted on the robot, and the lens illumination intensity sensor needs to be aligned with the lens of the observation camera to measure the lens illumination intensity value at the observation camera location.

[0025] In a specific embodiment, the temperature sensor is installed on the robot to measure the temperature near the camera lens. Since the temperature around the camera generally does not change significantly, the temperature sensor does not need to be too close to the camera. It is only necessary to avoid placing the temperature sensor near the heat dissipation vents of electronic components such as the camera, so as not to affect the temperature sensor readings due to the heat emitted from the heat dissipation vents of the electronic components, so that the temperature sensor can measure the accurate temperature in the pig farming area.

[0026] In a specific embodiment, a humidity sensor is mounted on the robot to measure the humidity level near the lens of the observation camera. The humidity sensor can be located below the observation camera in a position that does not affect the light from the lens.

[0027] In a specific embodiment, an ambient light intensity sensor is installed in the pig farming area to measure the ambient light intensity value within the pig farming area. Multiple ambient light intensity sensors are required, and these sensors are installed at various locations within the pig farming area. The entire pig farming area can be divided into three zones: A, B, and C. An ambient light intensity sensor is installed in each zone, and the corresponding ambient light intensity sensor in each zone is used to measure the ambient light intensity value within that zone.

[0028] In a specific embodiment, a timer is mounted on the robot to record time.

[0029] The temporal feature preprocessing module includes an image time calibration module, an image recognition module, an image cropping module, and an acquisition data processing module; the temporal feature preprocessing module is installed on the robot.

[0030] The image time calibration module is used to calibrate the total time of images captured by the observation camera with the time recorded by the timer. It replaces the timeline of the images captured by the observation camera with the timeline of the timer, arranging the images according to the time recorded by the timer. All data acquired by the data acquisition module is used with the time recorded by the timer as the horizontal axis. After extracting unclear images, other data within the corresponding time period can be extracted using the corresponding time period information.

[0031] The image recognition module is used to identify images captured by the observation camera as the robot moves along the travel track, and classifies them into clear images and unclear images according to their clarity.

[0032] In a specific embodiment, the method for classifying images into clear and unclear images based on their clarity is as follows: Quantitatively evaluate the image clarity indicators (such as edge intensity, contrast, and frequency domain energy). Specifically, numerical analysis is performed on the edge intensity, contrast, and frequency domain energy of the image to obtain the numerical change curves of the edge intensity, contrast, and frequency domain energy of the image. The change values ​​within a predetermined time period (generally set to 6-15 seconds) of the curve are compared with the judgment threshold. Images in which any one of the edge intensity, contrast, or frequency domain energy is lower than the corresponding judgment threshold are judged as unclear images.

[0033] In a more specific embodiment, the image sharpness is determined by contrast, the image is converted into a frame sequence, and the image is converted into a grayscale image (which reduces computational load). Then, the following steps are performed: Feature extraction: Calculate global / local brightness, contrast, grayscale distribution, etc. for each frame.

[0034] Change detection: Detecting brightness abrupt changes using thresholding or statistical models (such as Gaussian mixture models).

[0035] Spatial localization: Image segmentation techniques (such as GrabCut and U-Net) can be used to find the darkened areas.

[0036] Post-processing: Filter out transient changes (the change value of data for a predetermined time period is lower than the judgment threshold), and judge images whose contrast data change value exceeds the judgment threshold as unclear images.

[0037] The image cropping module is used to crop out the parts of the image that the image recognition module identifies as unclear, thus obtaining the unclear image.

[0038] The data acquisition and processing module is used to process the data acquired by the data acquisition module. It arranges the data sequentially according to the time recorded by the timer to form the lens light intensity curve, temperature curve, humidity curve, and ambient light intensity curve.

[0039] The temporal feature extraction module includes an image extraction module, a data extraction module, and an image analysis module; It should be added that the temporal feature extraction module is used to compare and analyze the unclear images identified and extracted by the temporal feature preprocessing module, determine the cause of the unclear images, and classify the unclear images according to the cause of their generation: Temporary dirty images; persistent dirty images; Temporary soiled areas can be further subdivided into: water droplet and water stain pollution, fog pollution, and temporary shielding pollution.

[0040] The temporal feature extraction module is installed on the robot; In a specific embodiment, the image extraction module is used to extract unclear images, and the data extraction module is used to extract data collected by the data acquisition module for the corresponding time period of the unclear images; The image analysis module is used to compare the unclear images extracted by the image extraction module and the unclear images extracted by the data extraction module within the corresponding time period with the data collected by the data acquisition module to determine the type of pollution.

[0041] The power module provides power to the data acquisition module, the time series feature preprocessing module, the time series feature extraction module, and the control module. The control module is used to control the opening and closing of the data acquisition module, the time series feature preprocessing module, and the time series feature extraction module; like Figures 2-6 As shown, a method for detecting dirt in the lens of an orbital robot based on time-series learning includes the following steps: S1. Data Acquisition: Data is acquired through the data acquisition module. The observation camera moves along the travel track with the robot and can capture images of pigs in the pig farming area. The lens illumination sensor is used to collect and observe the light intensity data around the camera lens; Temperature sensors are used to collect temperature data around the camera lens. The humidity sensor is used to collect humidity data around the camera lens. The ambient light intensity sensors are divided into multiple units, which are distributed in areas A, B, and C of the pig farming area to collect the ambient light intensity values ​​of area A, area B, and area C.

[0042] S2. Temporal feature preprocessing; S201, Image Time Calibration: Arrange the images captured by the observation camera in order according to the time recorded by the timer; S202. Unclear Image Screening: The image recognition module identifies the images captured by the observation camera based on their clarity, classifying them into unclear and clear images. The image cropping module then extracts the unclear images. During cropping, images from a certain period before and after the unclear image are extracted. The cropped images are arranged in order according to the time recorded by the timer. It should be noted that when cropping unclear images, the duration of the unclear images must exceed the set time, which is 1-3 minutes.

[0043] S203, Data Processing; The data acquisition and processing module receives and processes the data acquired by the data acquisition module. In a specific embodiment, the data acquisition and processing module arranges the data acquired by the lens light sensor, temperature sensor, humidity sensor, and lens light intensity sensor in sequence according to the time recorded by the timer, and obtains the lens light intensity curve, temperature curve, humidity curve, and ambient light intensity curve.

[0044] In a specific embodiment, the lens illumination intensity sensor is used to collect data on the illumination intensity at the lens and transmit the data to the data acquisition and processing module in the form of an electrical signal. The data acquisition and processing module receives and processes the value of the light intensity at the lens of the observation camera collected by the lens illumination intensity sensor and the time recorded by the timer, forming a lens illumination intensity value curve with time on the horizontal axis and the lens illumination intensity value on the vertical axis.

[0045] In a specific embodiment, the temperature sensor is used to collect temperature data and transmit the data to the data acquisition and processing module in the form of an electrical signal. The data acquisition and processing module receives and processes the temperature data collected by the temperature sensor and the time recorded by the timer to form a temperature value curve with time on the horizontal axis and temperature value on the vertical axis. In a specific embodiment, the humidity sensor is used to collect humidity data and transmit the data to the data acquisition and processing module in the form of an electrical signal. The data acquisition and processing module receives and processes the humidity data collected by the humidity sensor and the time recorded by the timer to form a humidity value curve with time on the horizontal axis and humidity value on the vertical axis.

[0046] In a specific embodiment, the ambient light intensity sensor is used to collect the ambient light intensity value of the pig farming area and transmit the data to the data acquisition and processing module in the form of an electrical signal. The data acquisition and processing module receives and processes the light intensity value in the pig farming area collected by the ambient light intensity sensor and the time recorded by the timer to form an ambient light intensity value curve with time on the horizontal axis and light intensity value on the vertical axis. S204. Data extraction: The data acquisition and processing module extracts data from the lens light intensity curve, temperature curve, humidity curve, and ambient light intensity curve of the unclear image within the time period captured by the image cropping module. S3. Temporal feature extraction: Based on the cause of unclear image formation, unclear images are divided into temporary dirty images and persistent dirty images. S301. Analyze the causes of unclear image generation and classify unclear images caused by water droplet contamination, fog contamination, and temporary obstruction contamination as temporary dirt; unclear images caused by light pollution can also be classified as temporary dirt.

[0047] In a specific embodiment, the identification of water droplet and water stain contamination is as follows: the image of temporary dirt gradually becomes clear (the image before the unclear image has a situation where it quickly changes from clear to dirty, and the unclear image has a clear gradual change process from dirty to clear), and during the process of the unclear image becoming clear, the value in the lens illumination intensity value curve is in an increasing or high value state.

[0048] The temporal feature extraction module has the functions of image clarity identification and data processing. By scanning the image, it can obtain the image clarity value and process the value. If the image clarity value gradually increases in the acquired image, and the lens illumination intensity value curve is in a high value or gradually increasing state during the same time period as the acquired image, the image is classified as water droplet or water stain pollution; the lens illumination intensity value is above 60Lx, which is the high value state mentioned above.

[0049] In a specific embodiment, fog pollution is identified by examining the humidity curve of an image with temporary dirt over the same time period. The humidity curve shows a high humidity value, and as the unclear image becomes clearer, the lens illumination intensity curve shows an increasing or high value. The unclear image exhibits a process of transitioning from clear to dirty, and also a process of transitioning from dirty to clear.

[0050] The temporal feature extraction module compares and analyzes the captured unclear image with the data collected by the data acquisition module for the corresponding time period. In the captured unclear image, the image clarity value gradually increases, the humidity value curve shows a high humidity value, and the value in the lens illumination intensity value curve is gradually increasing or at a high value. This segment of the image is classified as fog pollution. A humidity value above 60% is considered a high value, and a lens illumination intensity value above 60 Lx is considered a high value as mentioned above.

[0051] In a specific embodiment, temporary occlusion contamination is identified as follows: Images of temporary dirt appear and disappear rapidly, which is abnormal. Simultaneously, the light intensity recorded by the lens illumination curve remains stable without significant fluctuations during the same time period. In this case, the obstructions corresponding to the time period along the robot's trajectory need to be removed. If removal is not possible, the abnormal situation for that time period is marked, and images of the area reached are labeled as meaningless.

[0052] The temporal feature extraction module compares and analyzes the captured unclear images with the data collected by the data acquisition module for the corresponding time period. In the captured unclear images, the duration of the unclear images is less than the set time, which can be 1 minute. The overall duration of the captured images is 1 minute, indicating that the obstructions in the unclear images appear and disappear quickly. This is caused by the obstructions in the pig farming area blocking the observation camera. The above images are classified as temporary obstruction pollution. The overall duration of temporary obstruction pollution is not long and does not require special cleaning.

[0053] To handle temporary obstruction contamination, in a specific embodiment, two proximity switches (which can be used) can be placed before and after the obstruction. When the tracked robot moves to the proximity switches, the switches record the corresponding signals. The control module and timer record the time period between the two proximity switches, and the images during this time period are labeled as meaningless. This removes images where lens contamination cannot be determined, avoiding unnecessary identification and differentiation work and reducing the number of error reports. The temporal feature extraction module labels the images during the time period between the proximity switches as meaningless, and no further identification is performed.

[0054] In a specific embodiment, light pollution is identified as follows: The intensity of light varies at different locations and times in the pig farming area. To differentiate the light pollution affecting the lens image in specific time periods and areas, unclear images classified as temporary occlusion pollution are further screened. Data collected by the lens illumination sensor during the same time period as the unclear images identified as temporary occlusion pollution are extracted. If the values ​​of the data collected by the lens illumination sensor during the same time period as the unclear images identified as temporary occlusion pollution fluctuate significantly (i.e., the difference between the maximum and minimum values ​​of the collected data is large), and the fluctuation exceeds 80 Lx, it indicates that the temporary occlusion pollution is light pollution.

[0055] S302. The cause of unclear images due to non-temporary dirt is classified as persistent dirt.

[0056] In summary, this invention provides a method and system for detecting dirt on the lens of an orbital robot based on temporal learning. Through the cooperation of components such as the temporal feature preprocessing module and the temporal feature extraction module, the system can analyze the causes of image blurring by the observation camera, classifying the causes of image blurring into temporary and persistent contamination. After determining persistent contamination, the system notifies staff to intervene, which can reduce the false alarm rate, reduce the need for staff intervention, and reduce the burden on staff.

[0057] By identifying water droplet and water stain pollution, fog pollution, and temporary obstruction pollution, it is easy to identify unclear images caused by water droplet and water stain pollution, fog pollution, and temporary obstruction pollution. Pollution caused by temporary dirt does not require manual cleaning, reducing the frequency of staff attendance for maintenance of observation camera lenses.

[0058] If it is determined that the lens of the observation camera is temporarily obstructed or contaminated, and the obstruction cannot be removed, the images during the time period when the orbital robot travels to the temporary obstruction can be marked as meaningless images to avoid useless identification and differentiation work and reduce the number of abnormal errors.

[0059] The above-disclosed embodiments are merely a few specific examples of the present invention. However, the embodiments of the present invention are not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A dirt detection system for a track robot lens based on time-series learning, characterized in that, include: A data acquisition module is installed in the pig farming area to collect images, temperature values, humidity values, and lens light intensity values ​​within the pig farming area; The temporal feature preprocessing module receives and processes images, temperature values, humidity values, and light intensity values ​​collected by the data acquisition module, and extracts unclear images and the temperature, humidity, and light intensity values ​​within the time period of the unclear images. The temporal feature extraction module includes an image extraction module, a data extraction module, and an image analysis module; The temporal feature extraction module is used to receive and process the unclear images captured by the temporal feature preprocessing module, as well as the temperature, humidity, and light intensity values ​​within the time period of the unclear images. It analyzes the causes of unclear image generation, classifies unclear images caused by water droplet contamination, fog contamination, and temporary occlusion contamination as temporary dirty images, and classifies unclear images other than temporary dirty images as persistent dirty images.

2. The time-series learning-based lens contamination detection system for orbital robots as described in claim 1, characterized in that, The data acquisition module includes an observation camera, a lens light intensity sensor for recording the light intensity at the camera lens, a temperature sensor for measuring the temperature around the camera, a humidity sensor for measuring the humidity around the camera, an ambient light intensity sensor for measuring the light intensity in the pig farming area, and a timer for recording time. It also includes a travel track set in the pig farming area and a robot that can move along the travel track and is used to install the data acquisition module. The lens light intensity sensor, temperature sensor, humidity sensor, and timer are all mounted on the robot, and the ambient light intensity sensor is located in the pig farming area.

3. The time-series learning-based lens contamination detection system for an orbital robot as described in claim 2, characterized in that, The time-series feature preprocessing module includes an image time calibration module, an image recognition module, an image cropping module, and an acquisition data processing module. The image time calibration module is used to calibrate the duration of the images captured by the observation camera with the time recorded by the timer, and to replace the time axis of the images captured by the observation camera with the time recorded by the timer.

4. The dirt detection system for a track robot lens based on time-series learning as described in claim 3, characterized in that, The data acquisition and processing module arranges the data sequentially based on the time recorded by the timer, forming a lens illumination intensity curve, a temperature curve, a humidity curve, and an ambient light intensity curve.

5. The dirt detection system for a track robot lens based on time-series learning as described in claim 4, characterized in that, The image recognition module is used to identify images captured by the camera as the robot moves along the travel track, and classifies the images into clear images and unclear images according to their clarity. The image cropping module is used to crop out the parts of the image that the image recognition module identifies as unclear, thus obtaining the unclear image. The data acquisition and processing module is used to process the data acquired by the data acquisition module. It also includes a control module and a power supply module. The power supply module is used to provide power to the data acquisition module, the time series feature preprocessing module, the time series feature extraction module, and the control module.

6. A method for detecting dirt in the lens of an orbital robot based on time-series learning, characterized in that, The orbital robot lens dirt detection system based on time-series learning as described in any one of claims 1-5 is used, and the following steps are included in its use: S1. Data acquisition: Data is acquired through the data acquisition module. S2. Temporal feature preprocessing; S201, Image Time Calibration: Arrange the images captured by the observation camera in order according to the time recorded by the timer; S202. Unclear Image Screening: The image recognition module identifies the images captured by the observation camera according to their clarity and classifies them into unclear images and clear images. S203, Data Processing; The data acquisition module receives and processes the data acquired by the data acquisition module. S204. Data extraction: The data is collected by the data acquisition module according to the time period of the unclear image captured by the image cropping module. S3. Temporal feature extraction: Based on the cause of unclear image formation, unclear images are divided into temporary dirty images and persistent dirty images. S301. Analyze the causes of unclear image generation and classify unclear images caused by water droplet contamination, fog contamination, or temporary obstruction contamination as temporary dirty images. S302. Classify unclear images caused by non-temporary dirt as persistent dirt images.

7. The method for detecting dirt in the lens of an orbital robot based on time-series learning as described in claim 6, characterized in that, The method for collecting data through the data acquisition module mentioned in S1 is as follows: the observation camera moves along the travel track with the robot to take pictures of the pigs in the pig farming area. The lens illumination sensor is used to collect and observe the light intensity data around the camera lens; Temperature sensors are used to collect temperature data around the camera lens. The humidity sensor is used to collect humidity data around the camera lens. An ambient light intensity sensor is used to measure light intensity data in pig farming areas.

8. The method for detecting dirt in the lens of an orbital robot based on time-series learning as described in claim 6, characterized in that, The method mentioned in S203 for receiving and processing data collected by the data acquisition module is as follows: the data acquisition module arranges the data collected by the lens light sensor, temperature sensor, humidity sensor and lens light intensity sensor in sequence according to the time recorded by the timer, and obtains the lens light intensity curve, temperature curve, humidity curve and ambient light intensity curve.

9. The method for detecting dirt in the lens of an orbital robot based on time-series learning as described in claim 6, characterized in that, The method mentioned in S301 for determining unclear images caused by water droplet contamination, fog contamination, or temporary obstruction contamination is as follows: The temporal feature extraction module scans the image, obtains the numerical value of the image clarity, and processes the value. If the image clarity value gradually increases in the acquired image, and the lens illumination intensity value curve is in a high value or gradually increasing state during the same time period as the acquired image, the image is classified as water droplet and water stain pollution. The temporal feature extraction module compares and analyzes the captured unclear images with the data collected by the data acquisition module during the corresponding time period of the captured unclear images. In the captured unclear images, the image clarity value gradually increases, the humidity value curve shows that the humidity value is in a high state, and the value in the lens light intensity value curve is gradually increasing or in a high state. The above images are classified as fog pollution. The temporal feature extraction module compares and analyzes the captured unclear images with the data collected by the data acquisition module for the corresponding time period of the captured unclear images. Among the captured unclear images, those whose duration of unclear images is within the set time length are classified as unclear images caused by temporary occlusion pollution.

10. The method for detecting dirt in the lens of an orbital robot based on time-series learning as described in claim 6, characterized in that, Unclear images caused by unremovable obstructions are classified as meaningless images: Two sensor switches are set in front of and behind the obstruction. When the track robot moves to the sensor switch, the control module records the activation time of the two sensor switches. The control module marks the image during the time period from the activation of the first sensor switch to the activation of the second sensor switch as meaningless image.

Images