Livestock and poultry body condition detection method, device, equipment and system

Abstract

This application discloses a method, apparatus, equipment, and system for detecting the body condition of livestock and poultry, applied in the field of animal husbandry technology. To address the problems of low efficiency, low accuracy, and high cost in livestock and poultry body condition detection, the method involves acquiring images of at least one target part of the livestock or poultry to be detected; identifying the part type corresponding to each of the at least one target part images; and inputting the at least one target part image into a pre-established body condition detection model based on the part type to obtain the body condition category of the livestock or poultry to be detected. The body condition detection model is trained based on sample images of different target parts. This application achieves non-contact detection of livestock and poultry by combining images of at least one target part of the livestock or poultry to be detected with a body condition detection model, avoiding stress reactions in the livestock or poultry during the detection process. No human intervention is required during the detection process, improving detection efficiency and accuracy. Furthermore, the detection equipment is simple and low-cost.

Description

Technical Field

[0001] This application relates to the field of animal husbandry technology, and in particular to a method, device, equipment and system for detecting the physical condition of livestock and poultry. Background Technology

[0002] Body condition of livestock and poultry is a core indicator for measuring their reproductive performance, health level, and economic benefits of breeding, directly determining the production efficiency and breeding quality of large-scale farms. For example, the ideal body condition of breeding pigs needs to be maintained within a reasonable range. Excessive body condition can easily lead to dystocia, while excessive thinness can cause delayed estrus and a decrease in piglet survival rate. Both of these will shorten the lifespan of breeding pigs and increase breeding costs. Precise and efficient body condition monitoring is also a key support for the large-scale breeding and modern breeding of various livestock and poultry such as cattle and sheep.

[0003] Currently, livestock and poultry body condition assessment mainly relies on two traditional methods: manual assessment and ultrasonic backfat meter testing. Taking breeding pig body condition testing as an example, manual assessment depends on the farmer's visual observation and limb palpation, which is highly subjective, lacks accuracy, and is inefficient, leading to unscientific adjustments to feeding programs and making it difficult to adapt to the needs of large-scale monitoring. Although ultrasonic backfat meter testing can achieve quantitative assessment, it is prone to causing stress reactions in breeding pigs during the testing process, resulting in high stress risk. It also requires professional training for testing personnel, has a high operating threshold, and has high equipment purchase and maintenance costs, making it difficult to popularize in small and medium-sized pig farms.

[0004] Therefore, how to improve the efficiency and accuracy of livestock and poultry condition testing, avoid stress reactions in livestock and poultry, and reduce overall costs have become problems that need to be solved by those skilled in the art. Summary of the Invention

[0005] The purpose of this application is to provide a method, apparatus, equipment and system for detecting the physical condition of livestock and poultry, so as to avoid stress reactions in livestock and poultry during the detection process, improve detection efficiency and accuracy, and the detection equipment is simple and low in cost.

[0006] To address the aforementioned technical problems, the embodiments of this application provide the following technical solutions: This application provides a method for detecting the body condition of livestock and poultry, including: Acquire images of at least one target part of the livestock or poultry to be detected; Identify the part type corresponding to at least one of the target part images; Based on the body part type, at least one image of the target body part is input into a pre-established body condition detection model to obtain the body condition category of the livestock to be detected. The body condition detection model is trained based on sample images of different target areas.

[0007] In one implementation, the body condition detection model is trained based on sample images of different target body parts, including: Identify the different body parts corresponding to various livestock and poultry; Acquire sample images of different target parts of multiple livestock and poultry; Based on the actual body condition of the livestock and poultry, the body condition categories of the sample images of different target parts are labeled; The pre-defined classification model is trained using labeled sample images to obtain a body condition detection model.

[0008] In one implementation, training a preset classification model using labeled sample images to obtain a body condition detection model includes: The labeled sample images are preprocessed to obtain preprocessed sample images. The preprocessed sample images are input into each input terminal of the preset classification model in batches. The preset classification model performs feature extraction, feature fusion and classification analysis on each batch of input sample images to obtain a trained body condition detection model. One input terminal corresponds to one part type. For each batch of input sample images, the data of one input terminal is randomly set to zero.

[0009] In one implementation, the step of inputting at least one image of the target body part into a pre-established body condition detection model based on the body part type to obtain the body condition category of the livestock or poultry to be detected includes: Image preprocessing is performed on at least one of the target regions to obtain preprocessed target region images; Based on the part type corresponding to at least one of the target part images, the preprocessed target part images are respectively input to the input terminal of the corresponding part type in the pre-established body condition detection model; The body condition detection model extracts, fuses, and classifies features from the input target area image to obtain the body condition category of the livestock to be detected.

[0010] In one embodiment, before labeling the body condition categories of the sample images of different target parts based on the actual body condition of the livestock and poultry, the method further includes: Distortion correction is performed on the target area sample image to obtain the corrected target area sample image; Perform field detection on the corrected target area sample image to determine the field boundaries; Based on the field boundaries, the corrected target area sample image is cropped to obtain the target area sample image after removing the background. Then, based on the actual body condition of the livestock and poultry, the body condition categories of the sample images of different target parts are labeled, including: Based on the actual physical condition of the livestock and poultry, the physical condition categories of sample images of different target parts after removing the background are labeled.

[0011] In one embodiment, the at least one target region image includes a first region image and / or a second region image.

[0012] In one embodiment, the first part image is a buttock image, and the second part image is a back image.

[0013] In one embodiment, the image of the at least one target part is acquired by at least one image acquisition device mounted on a track machine; the track machine travels along tracks laid out in each livestock pen of the livestock farm.

[0014] In one implementation, it further includes: The body condition grade of the livestock and poultry to be tested is determined based on the body condition category; The feeding method of the feeder is adjusted based on the body condition level. The physical condition categories include valid physical condition types and invalid physical condition categories, and the valid physical condition types correspond one-to-one with the physical condition levels.

[0015] In one implementation, adjusting the feeding of the feeder based on the body condition level includes: Based on the various body condition levels obtained within a preset time period, the current body condition level of the livestock and poultry to be tested is determined; The current material feed rate is determined based on the current body condition level and the pre-established mapping relationship between body condition level and material feed rate. The feeding behavior of the feeder is adjusted according to the current feeding amount.

[0016] In one implementation, determining the current body condition level of the livestock to be tested based on various body condition levels obtained within a preset time period includes: From the various physical condition levels obtained within the preset time period, determine the target physical condition level with the largest number of physical condition levels. The target body condition level is used as the current body condition level of the livestock and poultry to be tested.

[0017] Another aspect of this application provides a livestock and poultry body condition detection device, comprising: The acquisition module is used to acquire images of at least one target part of the livestock or poultry to be detected; The identification module is used to identify the part type corresponding to at least one of the target part images respectively; The detection module is used to input at least one image of the target body part into a pre-established body condition detection model based on the body part type to obtain the body condition category of the livestock to be detected; wherein the body condition detection model is trained based on sample images of different target body parts.

[0018] Another aspect of this application provides an electronic device, including: Memory, used to store computer programs; A processor is used to execute the computer program to implement the steps of the livestock and poultry condition detection method as described above.

[0019] Another aspect of this application provides a livestock and poultry condition detection system, including: a track laid along each livestock and poultry pen in a livestock and poultry farm, a track machine traveling along the track, an image acquisition device mounted on the track machine, and electronic equipment as described above. The track machine is used to move to a designated location of the livestock to be detected based on a movement command, and to acquire images of at least one target part of the livestock to be detected through the image acquisition device and send them to the electronic device.

[0020] As can be seen from the above technical solutions, the embodiments of this application have the following advantages: This application provides a method for detecting the body condition of livestock and poultry, including: acquiring images of at least one target part of the livestock and poultry to be detected; identifying the part type corresponding to each of the at least one target part images; and inputting the at least one target part image into a pre-established body condition detection model based on the part type to obtain the body condition category of the livestock and poultry to be detected; wherein the body condition detection model is trained based on sample images of different target parts.

[0021] Therefore, this application demonstrates that a body condition detection model can be pre-established based on sample images of one or more different target parts of livestock and poultry. When detecting the body condition of livestock and poultry, images of at least one target part of the livestock and poultry can be acquired, and the part type of each target part image can be identified. These target part images are then input into the body condition detection model, which analyzes each target part image according to its corresponding part type to determine the body condition category of the livestock and poultry. This application achieves non-contact detection of livestock and poultry by combining images of at least one target part of the livestock and poultry with a body condition detection model, avoiding stress reactions during detection. The detection process requires no human intervention, improving efficiency and accuracy. Furthermore, the detection equipment is simple and low-cost.

[0022] In addition, this application also provides corresponding livestock and poultry body condition detection devices, electronic devices, and livestock and poultry body condition detection systems for livestock and poultry body condition detection methods, further making the methods more practical. The devices, electronic devices, and livestock and poultry body condition detection systems have corresponding advantages. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the prior art and embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 A flowchart illustrating a method for detecting the body condition of livestock and poultry provided in an embodiment of this application; Figure 2 This is a schematic diagram of an image preprocessing process provided in an embodiment of this application; Figure 3 This is a schematic diagram of the training process of a body condition detection model provided in an embodiment of this application; Figure 4 This application provides a schematic diagram of a livestock and poultry body condition detection process. Figure 5 A schematic diagram of a track machine provided in an embodiment of this application; Figure 6 This is a schematic diagram of the overall architecture of a livestock and poultry body condition detection system provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of a livestock and poultry body condition detection device provided in an embodiment of this application; Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0025] This application provides a method, apparatus, equipment, and system for detecting the physical condition of livestock and poultry, which avoids stress reactions in livestock and poultry during the detection process, improves detection efficiency and accuracy, and the detection equipment is simple and low in cost.

[0026] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0027] It is important to note that livestock and poultry body condition is a core indicator for measuring their reproductive performance, health level, and economic benefits, directly determining the production efficiency and breeding quality of large-scale farms. For example, the ideal body condition for breeding pigs needs to be maintained within a reasonable range. Excessive fatness can lead to dystocia, while insufficient fatness can result in delayed estrus and decreased piglet survival rates, both of which shorten the lifespan of breeding pigs and significantly increase breeding costs. Therefore, monitoring the body condition of breeding livestock and poultry is necessary. Currently, livestock and poultry body condition testing mainly relies on two traditional methods: manual assessment and ultrasonic backfat testing. However, both methods have certain limitations and cannot meet the development needs of precision farming and intelligent breeding.

[0028] Taking the body condition assessment of breeding pigs as an example, the manual assessment method relies on the visual observation and limb palpation of the farmers, combined with a 1-5 point scale to judge the body condition of the breeding pigs. Its core drawback is that it is highly subjective and lacks precision. It requires a high degree of reliance on the farming experience of the assessors. Different people have different understandings and grasps of the body condition scoring standards, which can easily lead to scoring deviations and result in a lack of scientific basis for adjusting feeding programs. In addition, manual assessment is inefficient and difficult to adapt to the large-scale monitoring needs of large-scale pig farms.

[0029] Ultrasonic backfat measuring instruments measure the thickness of fat in specific areas of breeding pigs using the principle of ultrasonic reflection, thus achieving a quantitative assessment of body condition. However, its technical limitations are also prominent. On the one hand, it carries a high risk of stress, as the breeding pigs need to be forcibly restrained during the testing process, which can easily trigger stress reactions such as struggling and panic. On the other hand, it has a high operational threshold and high cost, requiring professional training for testing personnel and regular calibration and maintenance of the equipment, resulting in high purchase and maintenance costs, making it difficult to popularize in small and medium-sized pig farms.

[0030] Therefore, this application provides a method for detecting the body condition of livestock and poultry, which enables non-contact detection, avoids stress reactions in livestock and poultry during the detection process, requires no human intervention, improves detection efficiency and accuracy, and uses simple and low-cost equipment. Please refer to... Figure 1 , Figure 1 This is a flowchart illustrating a method for detecting the body condition of livestock and poultry, provided in an embodiment of this application. The method includes: S110: Obtain an image of at least one target part of the livestock or poultry to be detected.

[0031] It should be noted that, in this embodiment of the application, when conducting body condition testing on the livestock and poultry to be tested, at least one target part image of the livestock and poultry to be tested can be acquired. This target part image can be acquired non-contactly by an image acquisition device installed on the track-mounted machine. The target part can be the waist, buttocks, back, etc., and the specific part can be predetermined according to requirements. In this embodiment of the application, at least one target part image of the livestock and poultry to be tested can be acquired, meaning multiple target parts can be predetermined. During the actual testing process, at least one corresponding target part image can be acquired for at least one target part, and at least one target part image can be acquired for subsequent body condition testing. Alternatively, target part images corresponding to two or more target parts can be acquired.

[0032] S120: Identify the part type corresponding to at least one target part image.

[0033] It is understandable that after acquiring at least one image of a target region, each image of the target region can be identified to determine the type of region corresponding to that image. Different images of different target regions correspond to different types of regions.

[0034] S130: Based on the body part type, input at least one target body part image into a pre-established body condition detection model to obtain the body condition category of the livestock to be detected; wherein, the body condition detection model is trained based on sample images of different target body parts.

[0035] In practical applications, target parts corresponding to each body part type can be predetermined in advance. Multiple livestock and poultry can be selected as sample objects. For each livestock or poultry, multiple sample images corresponding to each target part are acquired, resulting in a large number of sample images containing target parts corresponding to each body part type. Based on these sample images, a body condition detection model is trained to obtain the model. During the body condition detection process for the livestock or poultry to be detected, after determining the body part type corresponding to each target part image, the body condition detection model can be further used to analyze and identify each target part image in conjunction with the identified body part types to obtain the body condition category of the livestock or poultry to be detected.

[0036] In one implementation, the body condition detection model in S130 above is trained based on sample images of different target body parts, and may include: Identify the different body parts corresponding to various livestock and poultry; Acquire sample images of different target parts of multiple livestock and poultry; Based on the actual body condition of livestock and poultry, body condition categories are labeled for sample images of different target parts; The pre-defined classification model is trained using labeled sample images to obtain a body condition detection model.

[0037] It should be noted that at least one body part type can be predetermined, such as the waist, rump, and back. For multiple livestock and poultry, a track-mounted camera can capture multiple images of each animal and poultry, obtaining one or more target body part sample images corresponding to each body part type. This yields the actual body condition of the animal and poultry, thus determining its body condition category. The body condition category can then be labeled for each target body part sample image. In practical applications, a backfat meter can be used to measure the backfat value of livestock and poultry to determine their actual body condition and category. For example, by selecting target body part sample images from a preset number of days before (e.g., 3 days) and a preset number of days after (e.g., 3 days) of backfat meter measurement, and labeling these target body part sample images (e.g., using labelImg), the body condition category of the target body part image can be labeled, as well as information such as the animal's pen position, thus obtaining labeled sample images corresponding to each animal and poultry.

[0038] To improve the quality of sample data, abnormal sample data can be removed during the annotation process. For example, for a particular livestock or poultry, the body condition category determined by the breeder can be obtained, and combined with the backfat value measured by the backfat meter to determine the body condition category, the body condition deviation of the livestock or poultry can be identified. Based on this body condition deviation, abnormal sample data can be identified. For example, the body condition category can include body condition levels. Sample images of target parts with body condition deviations exceeding two levels are identified as abnormal sample data and removed, thereby obtaining sample images of each normal target part. These normal target part sample images are then labeled to obtain labeled sample images, and a training set (e.g., 50,000 sets) is determined based on these labeled sample images. Finally, a pre-set classification model is trained using all the labeled sample images to obtain a trained body condition detection model.

[0039] In this embodiment, the body condition category may include various valid body condition categories and invalid body condition categories. Each valid body condition category corresponds to a specific body condition level; that is, there is a one-to-one correspondence between valid body condition categories and body condition levels. Wireless body condition categories may include unqualified posture or no pig. Specifically, each body condition level may include levels 1 to 5. For example, a backfat value <10mm is level 1, a backfat value of 10-14mm is level 2, a backfat value of 15-18mm is level 3, a backfat value of 19-22mm is level 4, and a backfat value >22mm is level 5. Unqualified posture includes images of livestock lying down, curled up, or otherwise unable to be assessed for body condition.

[0040] In one embodiment, before labeling the body condition categories of sample images of different target parts based on the actual body condition of livestock and poultry, the method may further include: Distortion correction is performed on the target area sample image to obtain the corrected target area sample image; Perform field detection on the corrected target area sample image to determine the field boundaries; Based on the field boundaries, the target area sample image after correction is cropped to obtain the target area sample image after removing the background.

[0041] It should be noted that breeding pig pens are typically 3. A body measuring 0.7 meters is difficult to capture entirely with a regular camera. Therefore, in practical applications, a 180-degree fisheye camera can be used for image acquisition. However, fisheye cameras introduce significant distortion, causing image deformation of sample parts and affecting body condition assessment. Therefore, before labeling different target body part sample images based on the actual body condition of livestock and poultry, as described above, ... Figure 2 As shown, distortion correction, field detection, and field cropping can be performed on sample images of different target parts first. Specifically, latitude and longitude distortion correction can be used to correct the distortion of sample images of each target part. Then, field detection is performed on the corrected target part sample images to obtain the field boundaries in the corrected target part sample images. Then, the field sample images are cropped according to the determined field boundaries to remove the background, thus obtaining target part sample images containing only the corresponding parts. Specifically, when performing field detection, a large number (e.g., 5000) of field images can be acquired in advance using an image acquisition device, and the field images can be labeled using labelImg. Then, YOLOv11 is used to train a model on the labeled field images to obtain a field detection model, and this field detection model is used to perform field detection on the corrected target part sample images.

[0042] Accordingly, the process of labeling sample images of different target parts with body condition categories based on the actual body condition of livestock and poultry can include: Based on the actual physical condition of livestock and poultry, the physical condition categories of sample images of different target parts after background removal are labeled.

[0043] In other words, after obtaining the sample images of different target parts after removing the background, the above annotation method can be used to annotate the sample images of different target parts after removing the background according to the actual condition of the livestock and poultry.

[0044] In one implementation, the process of training a preset classification model using labeled sample images to obtain a body condition detection model may include: (1) Perform image preprocessing on the labeled sample images to obtain the preprocessed sample images.

[0045] It should be noted that, in this embodiment of the application, after obtaining the labeled sample images, each labeled sample image can be further preprocessed to obtain preprocessed sample images. Specifically, the image preprocessing in this embodiment includes scaling each labeled sample image proportionally to its width and height, padding with a (0,0,0) black border, and adjusting the image to 384 pixels. A resolution of 384 is used to obtain the preprocessed sample images.

[0046] (2) Input the preprocessed sample images into each input terminal of the preset classification model in batches. The preset classification model performs feature extraction, feature fusion and classification analysis on each batch of input sample images to obtain the trained body condition detection model.

[0047] One input terminal corresponds to one part type. For each batch of input sample images, the data of one input terminal is randomly set to zero.

[0048] Specifically, the preset classification model in this embodiment has multiple input terminals, each corresponding to a body part type. During the training of the preset classification model using pre-processed sample images, these pre-processed sample images can be input into each input terminal of the preset classification model in batches. That is, for each sample image in the current batch, the sample image can be input into one input terminal of the preset classification model corresponding to the body part type of the target body part according to the target body part in these sample images, and the input data of one input terminal can be randomly set to zero. After the preset classification model obtains each sample image in the current batch through each input terminal, it performs feature extraction, feature fusion and classification analysis on these sample images to obtain a trained body condition detection model.

[0049] In one implementation, the process of inputting preprocessed sample images in batches to the input terminals of a preset classification model, and performing feature extraction, feature fusion, and classification analysis on each batch of input sample images through the preset classification model to obtain a trained body condition detection model, may include: The preprocessed sample images are divided into multiple sample image groups; among them, the sample images of each part in each sample image group are of different parts, collected at the same time, and all correspond to the same livestock. During the current training round, the preprocessed sample images of each part in the current sample image group are input into the input terminal of the preset classification model corresponding to the corresponding part type, and the input data of one of the input channels is randomly set to zero; the preset classification model includes multiple input terminals, sub-networks connected to each input terminal, connection function layers connected to each sub-network, and residual networks connected to the connection function layers. By extracting features from sample images of corresponding body parts through each sub-network, feature information corresponding to each sample image is obtained. By fusing the various feature information through a connection function layer, a fused feature is obtained; A residual network is used to classify and analyze the fused features to obtain the body condition classification prediction results; If the training in the current round does not meet the training termination condition, the parameters of each sub-network, connection function layer and residual network are updated based on the body condition classification prediction results. The next set of sample images is then used as the current set of sample images. The process of inputting the preprocessed sample images of each part in the current set of sample images into the input terminal of the preset classification model corresponding to the corresponding part type is returned to be executed in the current round of training. This process continues until the training in the current round meets the training termination condition, and the trained body condition detection model is obtained.

[0050] In this embodiment, the labeled sample images can be divided into multiple sample image groups according to the livestock and poultry and the collection time. The sample images in each sample image group have different body parts, the same collection time, and all correspond to the same livestock or poultry. For example, if the body part category includes the buttocks and back, then each sample image group includes buttocks sample images and back sample images of the same livestock or poultry collected at the same time. These sample image groups are used to train a preset classification model.

[0051] Specifically, such as Figure 3 As shown, the preset classification model in this embodiment includes n input terminals, sub-networks connected to each input terminal, connection function layers connected to the n sub-networks, and residual networks connected to the connection function layers. Each input terminal corresponds to a body part type, and n is greater than or equal to 2. That is, for the current training round, each pre-processed sample image in the current sample image group can be input to the input terminal in the preset classification model corresponding to the body part type of the sample image. For example, input terminal 1 corresponds to body part type 1, input terminal 2 corresponds to body part type 2, ..., input terminal n corresponds to body part type n. Therefore, sample image 1 corresponding to body part type 1 can be input to input terminal 1, sample image 2 corresponding to body part type 2 can be input to input terminal 2, ..., and sample image n corresponding to body part type n can be input to input terminal n. Then, feature information 1 is extracted from sample image 1 through sub-network 1, feature information 2 is extracted from sample image 2 through sub-network 2, ..., and feature information 3 is extracted from sample image 3 through sub-network 3. The feature information from feature information 1 to feature information n is then fused through a connection function layer to obtain fused features. The fused features are then classified and analyzed using a residual network to obtain the body condition classification prediction results for this round.

[0052] After obtaining the body condition classification prediction results, a loss function can be determined based on these results. It can then be checked whether the loss value meets the preset loss requirement or whether the current training round has reached the preset number of rounds. If the loss value meets the preset loss requirement or the current training round has reached the preset number of rounds, the training round meets the training termination condition. The final network parameters for each sub-network, connection function layer, and residual network are then determined, resulting in a well-trained body condition detection model. If the loss value does not meet the preset loss requirement or the current training round has not reached the preset number of rounds, the training round does not meet the training termination condition. In this case, the parameters of each sub-network, connection function layer, and residual network can be updated based on the body condition classification prediction results, and the next training round can be performed until the training terminates, resulting in a well-trained body condition detection model.

[0053] In practical applications, at least one target part image in the embodiments of this application includes a first part image and / or a second part image; that is, the part type can include a first part type and / or a second part type. For example, each group of sample images can include a first part sample image and a second part sample image. The preset classification model can adopt a dual-input classification model, which includes two input terminals and two sub-networks. Furthermore, since the rump and back of livestock and poultry can well reflect their body condition, in the embodiments of this application, the first part type can be the rump, and the second part type can be the back. Therefore, the first part sample image can be an image of the rump of livestock and poultry, and the second part sample image can be an image of the back of livestock and poultry. Figure 4 As shown, preprocessed images of livestock hindquarters and backs can be input to the corresponding body part types. The corresponding feature information is extracted through the respective sub-networks, and the feature information is fused using a concat connection function layer. The fused features are then sent to a ResNet residual network for classification analysis to obtain the body condition classification prediction result. In this embodiment, body part images of different body parts are jointly trained to obtain a body condition detection model, which effectively avoids the problem of insufficient features from a single type of image and helps improve the model's detection accuracy.

[0054] It should also be noted that, in order to enable the trained body condition detection model to recognize the body condition of any type of body part image, in this embodiment of the application, when training the body condition detection model, for each batch of sample images input into the preset classification model, the image pixel values ​​of different body part types of sample images can be set to 0 with a preset probability (e.g., 10% probability) before being input into the model for model training.

[0055] In one embodiment, the process in S130 above, which involves inputting an image of at least one target body part into a pre-established body condition detection model based on body part type to obtain the body condition category of the livestock or poultry to be detected, may include: At least one target region image is preprocessed to obtain a preprocessed target region image; Based on the part type corresponding to at least one target part image, the preprocessed target part images are input to the input terminal of the corresponding part type in the pre-established body condition detection model; The body condition detection model extracts, fuses, and classifies features from the input target area image to obtain the body condition category of the livestock and poultry to be detected.

[0056] It is understood that, in the process of detecting the body condition of livestock and poultry to be tested in this embodiment of the application, after determining the part type of each target part image, each target part image can be preprocessed, that is, distortion correction, pen detection, and pen cropping are performed. Then, the cropped target part images are preprocessed to obtain preprocessed target part images. Then, the input terminal corresponding to each target part image is determined according to the part type of each target part image, and each target part image is input to the body condition detection model and the input terminal corresponding to the part type. Through the sub-network of each target part image input in the body condition detection model, feature extraction is performed on the corresponding target part images to obtain target feature information corresponding to each target part image. Then, feature fusion is performed on each target feature information through the connection function layer to obtain overall feature information. The overall feature information is classified and analyzed through the residual network to obtain the body condition category of the livestock and poultry to be tested.

[0057] In one implementation, based on the body part type corresponding to at least one target body part image, the preprocessed target body part images are respectively input to one input terminal of the corresponding body part type in a pre-established body condition detection model; the process of obtaining the body condition category of the livestock or poultry to be detected by performing feature extraction, feature fusion, and classification analysis on the input target body part images through the body condition detection model may include: Based on the body part type of each target body part image, at least one preprocessed target body part image is input into the target input terminal corresponding to the body condition detection model. By using the target sub-network connected to each target input in the body condition detection model, feature extraction is performed on the corresponding target part images to obtain target feature information corresponding to each target part image; The overall feature information is obtained by fusing the feature information of each target through the connection function layer in the body condition detection model. By classifying and analyzing the overall feature information using residual networks, the body condition categories of the livestock and poultry to be tested can be obtained.

[0058] It should be noted that in practical applications, the number of target body part images acquired can be less than the number of inputs to the body condition detection model. That is, when establishing the body condition detection model, sample images corresponding to n different body part types can be used for model training. In subsequent use of the body condition detection model, the number of target body part types collected for the livestock to be detected can be less than or equal to n, to better meet the actual needs of livestock farms and improve flexibility. When the number of target body part types collected for the livestock to be detected is less than n, there may be situations where the corresponding type of target body part image is not input to the body condition detection model's input. In this case, the input information for that input can be set to empty or zero.

[0059] In practical applications, to improve the accuracy of body condition detection, the body part types in this embodiment may include the buttocks and back. Therefore, the at least one target body part image collected for the livestock to be detected in this application includes a buttocks image and / or a back image, and the trained body condition detection model is also trained based on the buttocks sample image and the back sample image. Thus, in this embodiment, the buttocks image and / or back image of the livestock to be detected can be input to the input terminal corresponding to the body condition detection model to obtain the body condition category of the livestock to be detected. Whether to collect a buttocks image, a back image, or both, can be determined according to the actual situation, and this embodiment does not impose any special limitations here.

[0060] In one embodiment, an image of at least one target location is acquired by at least one image acquisition device mounted on a track machine; the track machine travels along tracks laid out in each livestock pen of the livestock farm.

[0061] It is understood that in this embodiment of the application, tracks for the movement of the track-mounted machine can be pre-laid along each livestock pen in the livestock farm. Image acquisition devices can be installed on the track-mounted machine. In practical applications, to simultaneously acquire images of two or more body parts, at least one image acquisition device can be installed. The position and angle of the image acquisition device can be determined according to the body parts to be photographed. For example, if images of the buttocks and back need to be captured, two image acquisition devices (e.g., cameras) can be installed. One image acquisition device can be installed directly below the track-mounted machine, specifically a fisheye camera with a resolution of 4 megapixels and a field of view of 180 degrees. A robotic arm can be installed on the track-mounted machine, and another image acquisition device can be installed at the front end of the robotic arm. This image acquisition device can specifically be an industrial camera with a resolution of 4 megapixels and a field of view of 90 degrees. Of course, to improve the image acquisition quality, a supplementary lighting module can also be installed. The supplementary lighting module can use infrared supplementary lighting to avoid strong light affecting the livestock.

[0062] It should also be noted that a track-mounted machine controller can be set up. This controller can use an API interface to move the track-mounted machine to a designated location, acquire location information, and control the extension or retraction of the robotic arm. When the robotic arm is extended, its onboard camera can capture images of the livestock's rump. A schematic diagram of the track-mounted machine is shown below. Figure 5 As shown. In addition, API interfaces are a type of software interface. This solution can also use other types of software interfaces, and can also achieve control directly through hardware signals, such as IO (Input / Output) high and low level signals, pulse control signals, etc., without the need for a software interface.

[0063] In one embodiment, the livestock and poultry body condition detection method may further include: The body condition grade of the livestock and poultry to be tested is determined based on the body condition category; Adjust the feeding of the feeder based on the body condition level; Among them, the physical condition categories include valid physical condition types and invalid physical condition categories, with a one-to-one correspondence between valid physical condition types and physical condition levels.

[0064] It is understood that, in the embodiments of this application, after obtaining the body condition category of the livestock to be tested, the body condition level of the livestock to be tested can be determined according to the body condition category, and the feeding situation of the feeder corresponding to the livestock to be tested can be adjusted according to the body condition level, so as to achieve precise feeding of the livestock to be tested.

[0065] Specifically, as mentioned above, the body condition category can include various valid and invalid body condition categories. Each valid body condition category corresponds to a specific body condition level, while invalid body condition categories can include unacceptable posture or no pigs. Specifically, each body condition level can include levels 1 to 5. For example, backfat value <10mm is level 1, backfat value 10-14mm is level 2, backfat value 15-18mm is level 3, backfat value 19-22mm is level 4, and backfat value >22mm is level 5. Unacceptable posture includes images of livestock lying down or curled up, making body condition assessment impossible.

[0066] In one embodiment, the process of adjusting the feeder's feeding based on body condition level may include: Based on the various body condition levels obtained within a preset time period, the current body condition level of the livestock and poultry to be tested is determined. The current material feed rate is determined based on the current material condition level and the pre-established mapping relationship between the material condition level and the material feed rate. Adjust the feeder's feeding behavior according to the current feed volume.

[0067] It should be noted that in this embodiment, a mapping relationship between body condition level and feed amount can be established in advance. After determining the body condition level of the livestock to be tested based on the body condition category of the livestock to be tested, the current body condition level of the livestock to be tested can be determined based on the body condition level obtained from each test within the most recent preset time period (e.g., 10 days). Then, based on the current body condition level and the pre-established mapping relationship between body condition level and feed amount, the current feed amount can be determined, and the feed amount of the feeder can be adjusted according to the current feed amount.

[0068] In one implementation, the process of determining the current body condition level of the livestock or poultry to be tested based on the various body condition levels obtained within a preset time period may include: From the various physical condition levels obtained within a preset time period, determine the target physical condition level with the most physical condition levels. The target body condition level is used as the current body condition level of the livestock and poultry to be tested.

[0069] Specifically, after determining the various body condition levels obtained within the preset time period, the frequency of each body condition level can be further determined, and the body condition level with the most occurrences can be taken as the target body condition level. This target body condition level can then be used as the current body condition level of the livestock to be tested.

[0070] For example, each pen can be inspected multiple times a day (e.g., twice) to obtain two body condition categories. Over 10 days, this would be 20 inspections, resulting in 20 body condition categories. Within each body condition category, 7 are identified as level 3, 3 as level 2, 2 as level 4, and 8 as having unacceptable postures. The level 3 category, which has the most occurrences, is then output, indicating that the current body condition type of the livestock or poultry is level 3.

[0071] It should also be noted that, please refer to Figure 6 The diagram shows the overall architecture of the livestock and poultry body condition detection system. In practical applications, the livestock and poultry body condition detection method in this embodiment can be achieved through methods such as... Figure 6 The AI ​​(Artificial Intelligence) box shown demonstrates how adjusting the feeder's dispensing based on body condition levels can be achieved through... Figure 6The feeder linkage module is implemented in this embodiment. The track machine control module can control the track machine to move to the location of the livestock to be detected, acquire images of the target area through image acquisition equipment, and input these images into the AI ​​box. The AI ​​box can call the body condition detection model to analyze and detect the target area image, obtain the body condition category of the livestock to be detected, and send this category to the feeder linkage module to adjust the feeder. In this embodiment, the AI ​​box is an edge computing box with NPU (Neural Processing Unit) computing power, which can be installed on the track machine to provide real-time computing power support for the body condition detection model's recognition and processing.

[0072] Therefore, this application demonstrates that a body condition detection model can be pre-established based on sample images of one or more different target parts of livestock and poultry. When detecting the body condition of livestock and poultry, images of at least one target part of the livestock and poultry can be acquired, and the part type of each target part image can be identified. These target part images are then input into the body condition detection model, which analyzes each target part image according to its corresponding part type to determine the body condition category of the livestock and poultry. This application achieves non-contact detection of livestock and poultry by combining images of at least one target part of the livestock and poultry with a body condition detection model, avoiding stress reactions during detection. The detection process requires no human intervention, improving efficiency and accuracy. Furthermore, the detection equipment is simple and low-cost.

[0073] This application also provides a corresponding apparatus for livestock and poultry body condition detection methods, further enhancing the practicality of the methods. The apparatus can be described from both functional module and hardware perspectives. The livestock and poultry body condition detection apparatus provided in this application is described below. This apparatus is used to implement the livestock and poultry body condition detection method provided in this application. In this embodiment, the livestock and poultry body condition detection apparatus may include or be divided into one or more program modules. These one or more program modules are stored in a storage medium and executed by one or more processors to complete the livestock and poultry body condition detection method disclosed in the above embodiments. The program module referred to in this application is a series of computer program instruction segments capable of performing specific functions, which is more suitable than the program itself for describing the execution process of the livestock and poultry body condition detection apparatus in the storage medium. The following description will specifically introduce the functions of each program module in this embodiment. The livestock and poultry body condition detection apparatus described below can be referred to in correspondence with the livestock and poultry body condition detection method described above.

[0074] From the perspective of functional modules, see Figure 7 , Figure 7 This application provides a structural diagram of a livestock and poultry body condition detection device, which may include: The acquisition module 11 is used to acquire an image of at least one target part of the livestock or poultry to be detected; The recognition module 12 is used to identify the part type corresponding to at least one target part image; The detection module 13 is used to input at least one target part image into a pre-established body condition detection model based on the part type to obtain the body condition category of the livestock to be detected; wherein, the body condition detection model is trained by the training module based on sample images of different target parts.

[0075] In one implementation, the training module includes: The first determining unit is used to determine the different body types corresponding to multiple livestock and poultry; The acquisition unit is used to acquire sample images of different target parts of multiple livestock and poultry. The annotation unit is used to annotate the body condition category of sample images of different target parts based on the actual body condition of livestock and poultry. The training unit is used to train a pre-defined classification model using labeled sample images to obtain a body condition detection model.

[0076] In one implementation, the training unit includes: The first processing subunit is used to perform image preprocessing on the labeled sample images to obtain preprocessed sample images. The second processing subunit is used to input the preprocessed sample images into each input terminal of the preset classification model in batches. The preset classification model performs feature extraction, feature fusion and classification analysis on each batch of input sample images to obtain a trained body condition detection model. One input terminal corresponds to one part type. For each batch of input sample images, the data of one input terminal is randomly set to zero.

[0077] In one embodiment, the detection module 13 includes: The first processing unit is used to perform image preprocessing on at least one target region image to obtain a preprocessed target region image. The input unit is used to input the preprocessed target part images to the corresponding part type input terminals of the pre-established body condition detection model based on the part type corresponding to at least one target part image. The analysis unit is used to extract features, fuse features, and classify the input target area image through the body condition detection model to obtain the body condition category of the livestock and poultry to be detected.

[0078] In one embodiment, the device further includes: The correction unit is used to correct the distortion of the target area sample image to obtain the corrected target area sample image. The detection unit is used to perform field detection on the corrected target area sample image and determine the field boundaries. The cropping unit is used to crop the target area sample image based on the field boundaries to obtain the target area sample image after removing the background. Therefore, the annotation unit is specifically used for: Based on the actual physical condition of livestock and poultry, the physical condition categories of sample images of different target parts after background removal are labeled.

[0079] In one embodiment, at least one target region image includes a first region image and / or a second region image.

[0080] In one embodiment, the first part image is a buttocks image, and the second part image is a back image.

[0081] In one embodiment, an image of at least one target location is acquired by at least one image acquisition device mounted on a track machine; the track machine travels along tracks laid out in each livestock pen of the livestock farm.

[0082] In one embodiment, the device further includes: The determination module is used to determine the body condition level of the livestock and poultry to be tested based on the body condition category; An adjustment module is used to adjust the feeding behavior of the feeder based on the body condition level; Among them, the physical condition categories include valid physical condition types and invalid physical condition categories, with a one-to-one correspondence between valid physical condition types and physical condition levels.

[0083] In one embodiment, the adjustment module includes: The second determining unit is used to determine the current body condition level of the livestock and poultry to be tested based on the various body condition levels obtained within a preset time period. The third determining unit is used to determine the current material feed rate based on the current body condition level and the pre-established mapping relationship between body condition level and material feed rate. The adjustment unit is used to adjust the feeding status of the feeder according to the current feeding amount.

[0084] In one embodiment, the second determining unit includes: The first determining subunit is used to determine the target physical condition level with the most physical condition levels from the various physical condition levels obtained within a preset time period. The second determining subunit is used to use the target body condition level as the current body condition level of the livestock and poultry to be tested.

[0085] It should be noted that the livestock and poultry condition detection device provided in this application embodiment has the same beneficial effects as the livestock and poultry condition detection method provided in the above embodiments, and for a detailed description of the livestock and poultry condition detection method involved in this application embodiment, please refer to the above embodiments, which will not be repeated here.

[0086] The livestock and poultry condition detection device mentioned above is described from the perspective of functional modules. Furthermore, this application also provides an electronic device, which is described from the perspective of hardware. Figure 8 A structural diagram of an electronic device provided in an embodiment of this application, such as... Figure 8 As shown, the electronic device includes: a memory 20 for storing computer programs; The processor 21 is used to execute computer programs to implement the steps of the livestock and poultry condition detection method as described in the above embodiments.

[0087] The electronic devices provided in this embodiment may include, but are not limited to, smartphones, tablets, laptops, or desktop computers.

[0088] The processor 21 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 21 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 21 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 21 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor 21 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0089] The memory 20 may include one or more computer-readable storage media, which may be non-transitory. The memory 20 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the memory 20 may be an internal storage unit of an electronic device, such as a server hard drive. In other embodiments, the memory 20 may be an external storage device of an electronic device, such as a plug-in hard drive on a server, a smart media card (SMC), a secure digital card (SD), a flash card, etc. Furthermore, the memory 20 may include both internal and external storage units of the electronic device. The memory 20 can be used not only to store application software and various types of data installed on the electronic device, such as code in the process of executing the livestock and poultry condition detection method, but also to temporarily store data that has been output or will be output. In this embodiment, the memory 20 is used to store at least the following computer program 201, which, after being loaded and executed by the processor 21, can implement the relevant steps of the livestock and poultry condition detection method disclosed in any of the foregoing embodiments. In addition, the resources stored in the memory 20 may also include an operating system 202 and data 203, and the storage method may be temporary storage or permanent storage. The operating system 202 may include Windows, Unix, Linux, etc. The data 203 may include, but is not limited to, data corresponding to livestock and poultry condition detection results.

[0090] In some embodiments, the electronic device may further include a display screen 22, an input / output interface 23, a communication interface 24, a power supply 25, and a communication bus 26. The display screen 22 and input / output interface 23, such as a keyboard, are user interfaces; optional user interfaces may also include standard wired interfaces, wireless interfaces, etc. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. The display may also be appropriately referred to as a screen or display unit, used to display information processed in the electronic device and to display a visual user interface. The communication interface 24 may optionally include a wired interface and / or a wireless interface, such as a Wi-Fi interface, a Bluetooth interface, etc., typically used to establish communication connections between the electronic device and other electronic devices. The communication bus 26 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 8 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0091] Those skilled in the art will understand that Figure 8 The structures shown do not constitute a limitation on electronic devices and may include more or fewer components than those shown.

[0092] Another aspect of this application provides a livestock and poultry condition detection system, including: a track laid along each livestock and poultry pen, a track machine traveling along the track, an image acquisition device installed on the track machine, and electronic equipment as described above. The track machine is used to move to a designated location of the livestock to be inspected based on movement commands, and to acquire images of at least one target part of the livestock to be inspected through an image acquisition device and send them to an electronic device.

[0093] In addition, this application embodiment also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described livestock and poultry condition detection method.

[0094] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0095] It should also be noted that, in this specification, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0096] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting the body condition of livestock and poultry, characterized in that, include: Acquire images of at least one target part of the livestock or poultry to be detected; Identify the part type corresponding to at least one of the target part images; Based on the body part type, at least one image of the target body part is input into a pre-established body condition detection model to obtain the body condition category of the livestock to be detected. The body condition detection model is trained based on sample images of different target areas.

2. The method for detecting the body condition of livestock and poultry according to claim 1, characterized in that, The body condition detection model is trained based on sample images of different target body parts, including: Identify the different body parts corresponding to various livestock and poultry; Acquire sample images of different target parts of multiple livestock and poultry; Based on the actual body condition of the livestock and poultry, the body condition categories of the sample images of different target parts are labeled; The pre-defined classification model is trained using labeled sample images to obtain a body condition detection model.

3. The method for detecting the body condition of livestock and poultry according to claim 2, characterized in that, The step of training a pre-defined classification model using labeled sample images to obtain a body condition detection model includes: The labeled sample images are preprocessed to obtain the preprocessed sample images. The preprocessed sample images are input into each input terminal of the preset classification model in batches. The preset classification model performs feature extraction, feature fusion and classification analysis on each batch of input sample images to obtain a trained body condition detection model. One input terminal corresponds to one part type. For each batch of input sample images, the data of one input terminal is randomly set to zero.

4. The method for detecting the body condition of livestock and poultry according to claim 3, characterized in that, Based on the body part type, at least one image of the target body part is input into a pre-established body condition detection model to obtain the body condition category of the livestock or poultry to be detected, including: Image preprocessing is performed on at least one of the target regions to obtain preprocessed target region images; Based on the part type corresponding to at least one of the target part images, the preprocessed target part images are respectively input to the input terminal of the corresponding part type in the pre-established body condition detection model; The body condition detection model extracts, fuses, and classifies features from the input target area image to obtain the body condition category of the livestock to be detected.

5. The method for detecting the body condition of livestock and poultry according to claim 2, characterized in that, Before labeling the body condition categories of the sample images of different target parts based on the actual body condition of the livestock and poultry, the method further includes: Distortion correction is performed on the target area sample image to obtain the corrected target area sample image; Perform field detection on the corrected target area sample image to determine the field boundaries; Based on the field boundaries, the corrected target area sample image is cropped to obtain the target area sample image after removing the background. Then, based on the actual body condition of the livestock and poultry, the body condition categories of the sample images of different target parts are labeled, including: Based on the actual physical condition of the livestock and poultry, the physical condition categories of sample images of different target parts after removing the background are labeled.

6. The method for detecting the body condition of livestock and poultry according to claim 1, characterized in that, The at least one target region image includes a first region image and / or a second region image.

7. The method for detecting the body condition of livestock and poultry according to claim 6, characterized in that, The first image is a buttocks image, and the second image is a back image.

8. The method for detecting the body condition of livestock and poultry according to claim 1, characterized in that, The image of at least one target part is acquired by at least one image acquisition device set on the track machine; the track machine travels along the tracks set up in each livestock pen of the livestock farm.

9. The method for detecting the body condition of livestock and poultry according to any one of claims 1 to 8, characterized in that, Also includes: The body condition grade of the livestock and poultry to be tested is determined based on the body condition category; The feeding method of the feeder is adjusted based on the body condition level. The physical condition categories include valid physical condition types and invalid physical condition categories, and the valid physical condition types correspond one-to-one with the physical condition levels.

10. The method for detecting the body condition of livestock and poultry according to claim 9, characterized in that, Adjusting the feeder's feeding based on the body condition level includes: Based on the various body condition levels obtained within a preset time period, the current body condition level of the livestock and poultry to be tested is determined; The current material feed rate is determined based on the current body condition level and the pre-established mapping relationship between body condition level and material feed rate. The feeding behavior of the feeder is adjusted according to the current feeding amount.

11. The method for detecting the body condition of livestock and poultry according to claim 10, characterized in that, The process of determining the current body condition level of the livestock to be tested based on the various body condition levels obtained within a preset time period includes: From the various physical condition levels obtained within the preset time period, determine the target physical condition level with the largest number of physical condition levels. The target body condition level is used as the current body condition level of the livestock and poultry to be tested.

12. A livestock and poultry body condition detection device, characterized in that, include: The acquisition module is used to acquire images of at least one target part of the livestock or poultry to be detected; The identification module is used to identify the part type corresponding to at least one of the target part images respectively; The detection module is used to input at least one image of the target body part into a pre-established body condition detection model based on the body part type to obtain the body condition category of the livestock to be detected; wherein the body condition detection model is trained based on sample images of different target body parts.

13. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the livestock and poultry condition detection method as described in any one of claims 1 to 11.

14. A livestock and poultry body condition detection system, characterized in that, include: The track laid along each livestock pen in the livestock farm, the track machine traveling along the track, the image acquisition device installed on the track machine, and the electronic device as described in claim 13. The track machine is used to move to a designated location of the livestock to be detected based on a movement command, and to acquire images of at least one target part of the livestock to be detected through the image acquisition device and send them to the electronic device.

Images