A non-contact detection device and detection method for the fullness of a bovine rumen

Abstract

The present invention discloses a non-contact detection device and method for the fullness degree of a bovine rumen. The device includes a one-way channel, a first light source module, a second light source module, an image acquisition module, a synchronous trigger unit, and an analysis and calculation module. The one-way channel controls the forward movement of the cow. The analysis and calculation module identifies the picture of the bovine rumen area, determines whether the bovine rumen area is located at the horizontal center position of the imaging picture. If it is at the center position, it sends a control signal to the synchronous trigger unit, receives the first picture, the second picture, and the natural light picture projected onto the bovine rumen area respectively, and calculates the fullness index based on the first picture, the second picture, and the natural light picture. The first light source module and the second light source module project onto the bovine rumen area respectively, and the image acquisition module acquires the picture of the bovine rumen area and the exposed picture. The synchronous trigger unit executes the projection and acquisition instructions. The present invention can achieve rapid, automatic, and accurate detection of the fullness index.

Description

Technical Field

[0001] The invention relates to the field of bovine rumen fullness detection, and in particular to a non-contact bovine rumen fullness detection device and a detection method. Background Art

[0002] The fullness of the cow's rumen can reflect the cow's rumination and digestion status in real time, help accurately control the feeding amount and nutritional ratio, prevent food accumulation or malnutrition, ensure digestive health, improve milk production efficiency and reduce feed waste. It is one of the key indicators for evaluating cattle feeding and digestive health.

[0003] At present, the detection of cattle rumen fullness mainly relies on manual visual inspection, lacks a quantitative measurement mechanism, and the detection results are greatly affected by the operator's experience and have subjective errors. The existing patent "A cattle rumen fullness measurement device and cattle rumen fullness measurement system" (patent number: CN 112545492 B) changes the size of the light channel by physically pressing to drive the compression part to deform, and measures the cattle rumen fullness by detecting the light intensity. This method requires contact with individual cattle and external force intervention may change the actual filling state of the rumen, resulting in inaccurate measurement. The existing patent "A cattle rumen fullness detection device and detection method" (application number 202411627435.9) uses ultrasonic ranging technology, but still needs to contact the cattle's body surface through a fixed device. In actual application, the operation is difficult and the efficiency is low. Summary of the invention

[0004] The purpose of the present invention is to provide a non-contact bovine rumen fullness detection device and detection method, aiming to solve the problem of rapid detection and accuracy of bovine rumen fullness.

[0005] The present invention provides a non-contact bovine rumen filling degree detection device, comprising:

[0006] One-way channel, light source module, image acquisition module, synchronous trigger unit and analysis and calculation module;

[0007] The light source module includes a first light source and a second light source, the image acquisition module includes a first image acquisition unit and a second image acquisition unit, and the analysis and calculation module includes a recognition unit and a calculation unit;

[0008] The one-way channel is used to guide the cattle to move forward in a preset direction;

[0009] The first light source and the second light source are installed above and below the one-way channel, respectively, and are used to project the cow rumen area at different angles;

[0010] The first image acquisition unit is used to acquire a picture of the cow's abdomen after the cow enters the one-way channel under natural light, and send it to the recognition unit;

[0011] The recognition unit is used to recognize the rumen area of the cow in the cow abdominal picture. When it determines that the rumen area of the cow is located at the horizontal center position in the cow abdominal picture, it sends a control signal to the synchronous trigger unit;

[0012] The synchronous trigger unit is used to receive the control signal sent by the recognition unit, control the first light source to turn on, then control the second image acquisition unit to acquire the first picture of the first light source projecting on the rumen area of the cow, then control the first light source to turn off, the second light source to turn on, control the second image acquisition unit to acquire the second picture of the second light source projecting on the rumen area of the cow, and finally control the second light source to turn off, and control the second image acquisition unit to acquire the natural light picture of the rumen area of the cow only under natural light;

[0013] The second image acquisition unit sends the first picture, the second picture and the natural light picture of the rumen area of the cow to the calculation unit;

[0014] The calculation unit is used to receive the first picture, the second picture and the natural light picture of the rumen area of the cow, and calculate the rumen fullness index.

[0015] The present invention also provides a non-contact method for detecting the rumen fullness of cows, guiding the cows to move forward in a preset direction through a one-way channel;

[0016] The first image acquisition unit acquires the cow abdominal picture after the cow enters the one-way channel under natural light, sends the cow abdominal picture to the recognition unit, recognizes the rumen area of the cow in the cow abdominal picture, and when it determines that the rumen area of the cow is located at the horizontal center position in the cow abdominal picture, it sends a control signal to the synchronous trigger unit;

[0017] The synchronous trigger unit controls the first light source to turn on, then controls the second image acquisition unit to acquire the first picture of the first light source projecting on the rumen area of the cow, then controls the first light source to turn off, the second light source to turn on, controls the second image acquisition unit to acquire the second picture of the second light source projecting on the rumen area of the cow, and finally controls the second light source to turn off, and controls the second image acquisition unit to acquire the natural light picture of the rumen area of the cow only under natural light;

[0018] Send the first picture, the second picture and the natural light picture of the rumen area of the cow to the calculation unit, and calculate the rumen fullness index.

[0019] By adopting the embodiment of the present invention, it is possible to completely avoid physical contact, eliminate the stress response of animals, and there is no need for manual intervention from cow guiding and positioning to result output, which can greatly improve the detection efficiency.

[0020] The above description is only an overview of the technical solution of the present invention. In order to understand the technical means of the present invention more clearly, it is implemented in accordance with the content of the specification. And in order to make the above and other purposes, features and advantages of the present invention more obvious and understandable, the following specifically describes the embodiments of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0022] Figure 1 is a schematic diagram of a non-contact bovine rumen fullness detection device according to an embodiment of the present invention;

[0023] Figure 2 is a schematic circuit structure diagram of a non-contact bovine rumen fullness detection device according to an embodiment of the present invention.

[0024] DESCRIPTION OF THE REFERENCE NUMERALS:

[0025] 1: Device main body; 2: One-way channel; 3: First light source; 4: Second light source; 5: Image acquisition module; 6: Synchronous trigger unit; 7: Analysis and calculation module. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] The following will clearly and completely describe the technical solutions of the present invention in conjunction with the embodiments. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0027] Device Embodiment

[0028] According to an embodiment of the present invention, a non-contact bovine rumen fullness detection device is provided. Figure 1 is a schematic diagram of a non-contact bovine rumen fullness detection device according to an embodiment of the present invention, as Figure 1 shown, specifically including:

[0029] A one-way channel 2, a light source module, an image acquisition module 5, a synchronous trigger unit 6, and an analysis and calculation module 7 provided on the device main body 1. The light source module includes a first light source 3 and a second light source 4. The image acquisition module 5 includes an image acquisition first unit and an image acquisition second unit. The analysis and calculation module 7 includes an identification unit and a calculation unit;

[0030] The one-way channel 2 is used to guide the cattle to move forward in a preset direction;

[0031] The first light source 3 and the second light source 4 are respectively installed above and below the one-way channel 2, and are respectively used to project the rumen area of the cattle at different angles;

[0032] The first image acquisition unit is used to acquire the abdominal pictures of the cattle after they enter the one-way channel 2 under natural light and send them to the recognition unit;

[0033] The recognition unit is used to recognize the rumen area in the abdominal pictures of the cattle. When it is judged that the rumen area is located at the horizontal center position in the abdominal pictures of the cattle, it sends a control signal to the synchronous trigger unit;

[0034] The synchronous trigger unit 6 is used to receive the control signal sent by the recognition unit, control the first light source 3 to turn on, then control the second image acquisition unit to acquire the first picture of the first light source 3 projecting the rumen area of the cattle, then control the first light source 3 to turn off, turn on the second light source 4, control the second image acquisition unit to acquire the second picture of the second light source 4 projecting the rumen area of the cattle, and finally control the second light source 4 to turn off, and control the second image acquisition unit to acquire the natural light picture of the rumen area of the cattle only under natural light;

[0035] The second image acquisition unit sends the first picture, the second picture and the natural light picture of the rumen area of the cattle to the calculation unit;

[0036] The calculation unit is used to receive the first picture, the second picture and the natural light picture of the rumen area of the cattle and calculate the rumen fullness index.

[0037] The specific design of the device is as follows:

[0038] The equipment main body 1 adopts a high-strength aluminum alloy frame, the surface is sprayed with an anti-corrosion coating, the protection level reaches IP68, it can withstand the humidity and ammonia corrosion in the pasture environment, and an integrated height adjustment mechanism is integrated. The height adjustment mechanism realizes vertical movement through the guide rail, and is friction-fixed to the equipment main body 1 with the one-way channel 2 through the fixing knob. The adjustment range is 1.2 - 1.8 m, which adapts to the rumen positions of cattle with different body types.

[0039] The one-way channel 2 is installed in the production area and serves as the overall support of the equipment. The two sides of the channel are vertically fixed on the ground, with a width of 700 mm. Adjustable limit baffles (adjustment range ±50 mm) are provided on both sides. The inner side of the baffles is covered with a flexible rubber layer (thickness 10 mm). The channel floor is embedded with anti-slip striped steel plates (strip depth 2 mm), and the inclination angle is 2° to guide the cattle to move forward naturally.

[0040] The first light source 3 and the second light source 4 are used to project onto the bovine rumen area respectively; the first light source 3 and the second light source 4 adopt an 850nm near-infrared light source array and pulse modulation technology to ensure short-time high-brightness projection. The projection range covers the left flank area of the rumen.

[0041] The first light source 3 is installed above the left rear of the one-way channel 2, 1.5m above the ground, with a projection inclination angle of 30°. It uses a 16×16 array of 850nm near-infrared LEDs, with a pulse modulation peak power of 200mW, a duty cycle of 1:10, and the spot diameter coverage range of 800 - 1000mm.

[0042] The second light source 4 is installed below the left rear of the one-way channel 2, 0.8m above the ground, with a projection inclination angle of 45°. The second light source 4 and the first light source 3 are in an asymmetric geometric distribution to enhance the projection contour difference and compensate for the projection deformation error caused by body position offset. It also uses a near-infrared light source of the same wavelength of 850nm, with a gradient design of light intensity distribution (central light intensity > edge light intensity by 20%).

[0043] The image acquisition module 5 is fixed on the device main body 1 in a way perpendicular to the one-way channel 2. The vertical distance is close to the height of the bovine rumen and is adjustable. It is installed in the center of the right side wall of the one-way channel 2 (facing the rumen projection surface), and includes a high-frame-rate global shutter camera and a supporting narrow-band filter. Specifically, it is a global shutter CMOS camera with a frame rate ≥120fps, equipped with an 850nm narrow-band filter (bandwidth ±5nm, transmittance > 90%) installed in the center of the right side wall of the channel, 1.2m away from the central axis of the channel, and the optical axis coincides with the normal line of the rumen projection surface.

[0044] The synchronization trigger unit 6 uses an FPGA hardware trigger circuit to control the light source pulse and the camera exposure timing. The delay time from the light source turning on to exposure < 100μs, and the time-sharing trigger accuracy is ±10μs to ensure no light pollution within the image acquisition interval of the two light sources. The synchronization trigger unit 6 is used to control the time-sharing trigger timing of the first light source 3 and the second light source 4 and the image acquisition second unit to ensure the synchronization of light source irradiation and image acquisition while ensuring no crosstalk in the images of the two light sources; the exposure time of the image acquisition second unit is strictly synchronized with the light source pulse (exposure time 1ms) to suppress ambient light noise.

[0045] The analysis and calculation module 7 is built-in with an embedded GPU (computing power ≥4TFLOPS), and combines a bovine rumen area detection model based on deep vision to automatically identify the bovine rumen area and judge the position of the bovine rumen area. The deep vision is implemented using YOACT and Cascade Mask R-CNN algorithms.

[0046] The device also includes a network module. The network module uses an RFID antenna to identify the individual identity of the cow and upload the fullness index to the cloud for business docking with the application system.

[0047] Method Embodiment

[0048] According to an embodiment of the present invention, a non-contact method for detecting the fullness of a bovine rumen is provided, and it is implemented based on a non-contact device for detecting the fullness of a bovine rumen disclosed in the device embodiment. Specifically, it includes:

[0049] 1. Guide the cow to move forward in a preset direction through the one-way channel 2. The first image acquisition unit acquires a picture of the cow's abdomen after the cow enters the one-way channel under natural light, and sends the picture of the cow's abdomen to the recognition unit. When it is determined that the rumen area of the cow is located at the horizontal center position of the picture of the cow's abdomen, a control signal is sent to the synchronous trigger unit 6;

[0050] For rumen area recognition and positioning, instance segmentation models such as YOACT and Cascade Mask R-CNN are used for real-time recognition, and high-accuracy recognition is achieved through means such as confidence adjustment and area secondary verification.

[0051] When the centroid of the recognized result area is at the horizontal center position of the image, high-speed time-sharing projection imaging is automatically triggered. When the rumen area of the cow is at the horizontal center position of the imaging image, the synchronous trigger unit 6 controls the time-sharing trigger timing of the first light source 3 and the second light source 4 and the image acquisition module. The synchronous trigger unit 6 executes the following steps according to the following timing:

[0052] The synchronous trigger unit 6 triggers the first light source 3 to turn on, the light source duration is 10 ms, and the exposure time of the image acquisition module is strictly synchronized with the light source pulse (exposure time 1 ms) to suppress ambient light noise. The second image acquisition unit is synchronously triggered to capture an image ; Turn off the first light source 3: Wait for 2 ms to eliminate afterglow; The synchronous trigger unit 6 switches to trigger the second light source 4 to turn on, and the second image acquisition unit is synchronously triggered to capture an image ; Turn off the second light source 4.

[0053] Acquire a natural light picture of the rumen area of the cow when there is no first light source 3 and second light source 4, only natural light, and send it to the calculation unit;

[0054] 2. The calculation unit obtains the first picture, the second picture and the natural light picture of the rumen area of the cow, and calculates the effective projection areas of the first picture and the second picture based on the first picture, the second picture and the natural light picture of the rumen area of the cow;

[0055] Since the same-band light source is adopted, the projection signals of the two light sources need to be separated by high-speed time-sharing triggering. Within a very short time interval, the change of ambient light can be ignored, and differential processing can effectively extract the effective projection area. Through differential processing, the ambient light and sensor noise are eliminated, and the rumen shadow areas projected independently by the two light sources are accurately extracted, and finally the effective projection area is calculated. 、 The experiment shows that the algorithm can still maintain a signal-to-noise ratio ≥ 28 dB under 2000 lux ambient light.

[0056] Calculating the effective projection areas of the first picture and the second picture based on the first picture, the second picture and the natural light picture of the bovine rumen area specifically includes: calculating the effective projection areas A1 and A2 using Formula 1, and Formula 1 is as follows:

[0057] ,

[0058] Formula 1;

[0059] where is the pixel value of the natural light picture of the bovine rumen area at the coordinate , is the pixel value of the first picture at the coordinate , is the pixel value of the second picture at the coordinate ; , is the standard deviation of the natural light picture of the bovine rumen area, and the threshold is set to 3 times the standard deviation of the natural light picture of the bovine rumen area, covering 99.7% of the noise fluctuations (based on the 3σ principle of normal distribution); is the step function, which is used for binary segmentation of the projection area. When , otherwise it is 0, and v is the function parameter.

[0060] 3. Calculate the deformation difference index of the first picture and the second picture;

[0061] The asymmetric layout of the dual light sources results in differences in the projection contours. The deformation caused by the body position offset is quantified through the gradient difference, solving the problem that the single light source is insensitive to the contour deformation. Calculate the deformation difference index . When the cow leans sideways, the projection area of the first light source 3 (rear upper) shrinks, and the projection area of the second light source 4 (front lower) expands, and the gradient difference increases significantly. The experimental data shows that when the body position offset is 5°, the value increases by 3.2 times, and the sensitivity is 4 times higher than that of the single light source area method.

[0062] Calculating the deformation difference index of the first picture and the second picture specifically includes: calculating the deformation difference index D based on Formula 2, and Formula 2 is as follows:

[0063] Formula 2;

[0064] Wherein, is the gradient operator of the valid pixel point i with coordinates in the first picture, and is used to calculate the edge intensity of the pixel points in the first picture; is the gradient operator of the valid pixel point i with coordinates in the second picture, and is used to calculate the edge intensity of the pixel points in the second picture; is the number of valid pixel points in the overlapping area of the double light sources, The larger the value, the more significant the difference in the projection contours of the double light sources, and the more severe the body position shift or local rumen deformation is reflected. is the number of valid pixel points in the overlapping area of the double light sources, The larger the value, the more significant the difference in the projection contours of the double light sources, and the more severe the body position shift or local rumen deformation is reflected.

[0065] 4. Calculate the overlapping area of the first picture and the second picture, and calculate the dynamic weight;

[0066] First, calculating the overlapping area of the first picture and the second picture specifically includes: calculating the overlapping area of the first picture and the second picture based on Formula 3, and Formula 3 is as follows:

[0067] Formula 3;

[0068] is the threshold for determining the overlapping area.

[0069] Second, calculating the dynamic weight specifically includes calculating the dynamic weight based on Formula 4, and Formula 4 is as follows:

[0070] When the body position is stable, that is , , the area mean value is preferably adopted;

[0071] When the body position shifts, that is , , the weight of the enhanced deformation difference is increased.

[0072] The overlapping area of the double light source projection reflects the reliable measurement area when the body position is stable. The higher its proportion, the higher the measurement confidence. Introducing the dynamic weight adaptively balances the contributions of the area and the deformation index.

[0073] 5. By fusing the four core variables of the effective projection area of the double light sources ( , ), the overlapping area ( ), the deformation difference index ( ), and the dynamic weight ( ), non-contact accurate detection of rumen fullness is realized. Among them, the area mean value Reflect the overall filling state of the rumen, proportion of the overlapping area Characterize the body position stability, deformation difference index Quantify the dynamic offset of the projection contour, dynamic weight Adaptively adjust the contribution weights of the area and deformation features according to the proportion of the overlapping area. Calculate the filling degree index based on the above variables .

[0074] Calculate the filling degree index using Formula 5 based on the effective projection area, deformation difference index, overlapping area, and dynamic weight. Formula 5 is as follows:

[0075] ;

[0076] Wherein, is the average value of the projection area in the standard filling state, is the maximum allowable deformation difference; is the dynamic weight, with a value range of 0 or 1; is the ratio of the average value of the double-light-source projection area to the standard area; is the normalized value of the deformation difference, used to compensate for the area distortion caused by body position offset, The lower it is, the stronger the deformation compensation, suitable for the activity interference scenario.

[0077] 6. Identify the individual identity of the cow through the network module and upload the filling degree index to the cloud.

[0078] The present invention scores the filling degree index, specifically including;

[0079] Based on clinical palpation verification and pasture management requirements, establish a non-linear piecewise scoring system, with K1, K2, K3, and K4 as the scoring thresholds: when the quantization value is greater than K1, the score is 1 point; when the quantization value is greater than K2 and less than or equal to K1, the score is 2 points; when the quantization value is greater than K3 and less than or equal to K2, the score is 3 points; when the quantization value is greater than K4 and less than or equal to K3, the score is 4 points; when the quantization value is less than or equal to K4, the score is 5 points.

[0080] A non-contact cow rumen filling degree detection device and detection method of the present invention can quickly and accurately calculate the filling degree of the rumen part. Compared with the prior art, the present invention has the following advantages.

[0081] Non-contact and stress-free: Completely avoid physical contact, eliminate the animal stress response, and meet the requirements of animal welfare;

[0082] Strong anti-interference ability: The non-visible light band time-division differential imaging technology effectively reduces the environmental light interference;

[0083] High-precision analysis: By fusing multi-dimensions of double projection area, overlapping area and morphological similarity, the accuracy of the single light source solution is effectively improved;

[0084] Fully automated operation: No manual intervention is required from cattle guiding and positioning to result output, which can greatly improve the detection efficiency;

[0085] Low cost and easy to deploy: No 3D reconstruction hardware is required, only 2D image processing algorithms are needed, and the equipment cost is effectively reduced;

[0086] Evaluate the filling degree to facilitate quantification of good and bad.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the technical solutions of the present invention deviate from the scope of the present solution.

Claims

1. A non-contact bovine rumen fullness detection device, characterized in that: It includes a one-way channel, a light source module, an image acquisition module, a synchronous trigger unit and an analysis and calculation module; The light source module includes a first light source and a second light source, the image acquisition module includes a first image acquisition unit and a second image acquisition unit, and the analysis and calculation module includes a recognition unit and a calculation unit; The one-way channel is used to guide the cattle to move forward in a preset direction; The first light source and the second light source are installed above and below the one-way channel, respectively, and are used to project the rumen area of ​​the cow at different angles; The first image acquisition unit is used to acquire a picture of the cow's abdomen after the cow enters the one-way channel under natural light, and send it to the recognition unit; The identification unit is used to identify the cow rumen area in the cow abdomen picture, and when it is determined that the cow rumen area is located in the horizontal center of the cow abdomen picture, send a control signal to the synchronization trigger unit; The synchronous trigger unit is used to receive the control signal sent by the identification unit, control the first light source to turn on, and then control the second image acquisition unit to acquire a first picture of the cattle rumen area projected by the first light source, and then control the first light source to turn off, the second light source to turn on, and control the second image acquisition unit to acquire a second picture of the cattle rumen area projected by the second light source, and finally control the second light source to turn off, and control the second image acquisition unit to acquire a natural light picture of the cattle rumen area under natural light only; The second image acquisition unit sends the first picture, the second picture and the natural light picture of the cow rumen area to the computing unit; The calculation unit is used to receive the first picture, the second picture and the natural light picture of the cattle rumen area, and calculate the cattle rumen filling index; The calculation unit is specifically used to calculate the effective projection area of ​​the first image and the second image based on the first image and the second image and the natural light image of the bovine rumen area, calculate the deformation difference index of the first image and the second image, calculate the overlapping area of ​​the first image and the second image, calculate the dynamic weight, and calculate the fullness index based on the effective projection area, the deformation difference index, the overlapping area and the dynamic weight; The method of calculating the effective projection areas of the first image and the second image based on the first image and the second image and the natural light image of the cow rumen region specifically includes: using formula 1 to calculate the effective projection areas A1 and A2, and formula 1 is as follows: ， Formula 1; in Natural light picture of the rumen area of ​​cattle at coordinates The pixel value at For the first image at coordinates The pixel value at For the second image at coordinates The pixel value at ; , is the standard deviation of the natural light image of the cow rumen area, the threshold It is set to 3 times the standard deviation of the natural light image of the cow rumen area; is a step function used to binarize the projection area. , otherwise it is 0, v is the function parameter; The calculating of the deformation difference index between the first image and the second image specifically includes: calculating the deformation difference index D based on Formula 2, where Formula 2 is as follows: Formula 2: in, The coordinates in the first picture are The gradient operator of the effective pixel point i is used to calculate the edge strength of the pixel point in the first picture; The coordinates in the second picture are The gradient operator of the effective pixel point i is used to calculate the edge strength of the pixel point of the second picture; is the number of effective pixels in the overlapping area of ​​light source No. 1 and light source No. 2; Calculating the overlapping area of ​​the first image and the second image specifically includes: calculating the overlapping area of ​​the first image and the second image based on Formula 3, where Formula 3 is as follows: Formula 3: To determine the threshold of the overlapping area; Calculating the dynamic weight specifically includes calculating the dynamic weight based on Formula 4, which is as follows: When the body position is stable, , Take 1 When the body position shifts, , Take 0, formula 4; The calculation of the fullness index based on the effective projection area, the deformation difference index, the overlapping area and the dynamic weight specifically includes: calculating the fullness index based on formula 5, which is as follows: , Formula 5; in, is the mean projection area under standard filling conditions, is the maximum allowable deformation difference; is a dynamic weight, with a value range of 0 or 1; It is the ratio of the mean projection area of ​​dual light sources to the standard area; It is the normalized value of deformation difference, which is used to compensate for the area distortion caused by body position deviation.

2. A non-contact method for detecting bovine rumen fullness, based on the device of claim 1, characterized in that: Guide cattle in a preset direction through one-way channels; The first image acquisition unit acquires a picture of the cow's abdomen after the cow enters the one-way channel under natural light, sends the picture of the cow's abdomen to the recognition unit, recognizes the cow's rumen area in the cow's abdomen picture, and sends a control signal to the synchronization trigger unit when it is determined that the cow's rumen area is located at the horizontal center of the cow's abdomen picture; The synchronous trigger unit controls the first light source to turn on, and then controls the second image acquisition unit to acquire the first image of the cattle rumen area projected by the first light source, and then controls the first light source to turn off, and the second light source to turn on, and controls the second image acquisition unit to acquire the second image of the cattle rumen area projected by the second light source, and finally controls the second light source to turn off, and controls the second image acquisition unit to acquire the natural light image of the cattle rumen area under natural light only; Sending the first picture, the second picture and the natural light picture of the cattle rumen area to a calculation unit to calculate the cattle rumen filling index; The calculation of the cattle rumen filling index specifically includes: Based on the first image and the second image and the natural light image of the cow rumen area, the effective projection area of ​​the first image and the second image is calculated, the deformation difference index of the first image and the second image is calculated, the overlapping area of ​​the first image and the second image is calculated, the dynamic weight is calculated, and the fullness index is calculated based on the effective projection area, the deformation difference index, the overlapping area and the dynamic weight; The method of calculating the effective projection areas of the first image and the second image based on the first image and the second image and the natural light image of the cow rumen region specifically includes: using formula 1 to calculate the effective projection areas A1 and A2, and formula 1 is as follows: ， Formula 1; in Natural light picture of the rumen area of ​​cattle at coordinates The pixel value at For the first image at coordinates The pixel value at For the second image at coordinates The pixel value at ; , is the standard deviation of the natural light image of the cow rumen area, the threshold It is set to 3 times the standard deviation of the natural light image of the cow rumen area; is a step function used to binarize the projection area. , otherwise it is 0, v is the function parameter; The calculating of the deformation difference index between the first image and the second image specifically includes: calculating the deformation difference index D based on Formula 2, where Formula 2 is as follows: Formula 2: in, The coordinates in the first picture are The gradient operator of the effective pixel point i is used to calculate the edge strength of the pixel point in the first picture; The coordinates in the second picture are The gradient operator of the effective pixel point i is used to calculate the edge strength of the pixel point of the second picture; is the number of effective pixels in the overlapping area of ​​light source No. 1 and light source No. 2; Calculating the overlapping area of ​​the first image and the second image specifically includes: calculating the overlapping area of ​​the first image and the second image based on Formula 3, where Formula 3 is as follows: Formula 3: To determine the threshold of the overlapping area; Calculating the dynamic weight specifically includes calculating the dynamic weight based on Formula 4, which is as follows: When the body position is stable, , Take 1; When the body position shifts, , Take 0, formula 4; The calculation of the fullness index based on the effective projection area, the deformation difference index, the overlapping area and the dynamic weight specifically includes: calculating the fullness index based on formula 5, which is as follows: Formula 5: in, is the mean projection area under standard filling conditions, is the maximum allowable deformation difference; is a dynamic weight, with a value range of 0 or 1; It is the ratio of the mean projection area of ​​dual light sources to the standard area; It is the normalized value of deformation difference, which is used to compensate for the area distortion caused by body position deviation.

3. The method according to claim 2, characterized in that The method also includes scoring a filling index.

Images