Bird recognition method and apparatus based on bird posture and image quality

Abstract

Embodiments of the present disclosure disclose a bird identification method and device based on bird posture and image quality. A specific embodiment of the method comprises: in response to detecting the presence of bird activity in the feeding area, collecting real-time monitoring video corresponding to the feeding area; video sampling is performed on the real-time monitoring video; for the to-be-detected image, the following processing steps are performed: bird posture identification is performed on the to-be-detected image; determine the image quality result corresponding to the to-be-detected image; perform bird identification on the to-be-detected image; according to the bird posture identification result and the image quality result, the bird identification result is optimized; according to the obtained optimized bird identification result sequence, generate a global bird identification result for the real-time monitoring video. This embodiment greatly improves the bird identification accuracy.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the field of computer technology, and more specifically to a method and apparatus for bird recognition based on bird posture and image quality. Background Technology

[0002] With the rapid development of artificial intelligence (AI) technology, its application in fields such as biometrics and species conservation is of great significance, especially for animal behavior analysis and protection. Currently, particularly in bird identification, the accuracy of bird identification is hampered by issues such as bird posture and image quality degradation.

[0003] The information disclosed in this background section is only intended to enhance the understanding of the background of the inventive concept, and therefore may contain information that does not form prior art known to those skilled in the art. Summary of the Invention

[0004] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0005] Some embodiments of this disclosure propose bird recognition methods and apparatus based on bird posture and image quality to solve one or more of the technical problems mentioned in the background section above.

[0006] In a first aspect, some embodiments of this disclosure provide a bird recognition method based on bird posture and image quality. The method includes: in response to detecting bird activity within a feeding area, acquiring real-time monitoring video corresponding to the feeding area; performing video sampling on the real-time monitoring video to obtain a sequence of images to be detected; for each image in the sequence of images to be detected, performing the following processing steps: performing bird posture recognition on the image to be detected to generate a bird posture recognition result; determining the image quality result corresponding to the image to be detected; performing bird recognition on the image to be detected to determine a bird recognition result; optimizing the bird recognition result based on the bird posture recognition result and the image quality result to obtain an optimized bird recognition result; and generating a global bird recognition result for the real-time monitoring video based on the obtained optimized bird recognition result sequence.

[0007] Optionally, before acquiring real-time monitoring video of the feeding area in response to detecting bird activity in the feeding area, the method further includes: detecting heat source changes and / or pressure changes in the feeding area using an infrared sensor and / or a pressure sensor to determine whether bird activity exists in the feeding area.

[0008] Optionally, the above-mentioned video sampling of the real-time monitoring video to obtain the image sequence to be detected includes: downsampling the real-time monitoring video at preset time intervals to obtain the image sequence to be detected.

[0009] Optionally, the above-mentioned video sampling of the real-time monitoring video to obtain the image sequence to be detected includes: downsampling the real-time monitoring video at preset time intervals to obtain a candidate image sequence; for each candidate image in the candidate image sequence, performing the following filtering steps: performing image binarization processing on the candidate image to obtain a binarized image; performing dilation processing on the binarized image to obtain a dilated image; determining the image difference between the dilated image and the background image to obtain at least one difference region, wherein the background image is an image when there are no obstructions in the feeding area; in response to the existence of a difference region with a corresponding area larger than a preset area in the at least one difference region, determining the candidate image as the image to be detected.

[0010] Optionally, the above-mentioned bird pose recognition of the image to be detected to generate a bird pose recognition result includes: determining the probability distribution of bird poses corresponding to the image to be detected by using a pre-trained bird pose recognition model, wherein the probability distribution of bird poses represents the confidence level of the image to be detected containing birds corresponding to different bird poses; and determining the bird pose recognition result based on the bird pose distribution.

[0011] Optionally, determining the image quality result corresponding to the image to be detected includes: scoring the image quality of the image to be detected using a pre-trained image quality scoring model to obtain the image quality result.

[0012] Optionally, the above-mentioned bird identification of the image to be detected to determine the bird identification result includes: performing bird identification on the image to be detected using a pre-trained fine-grained bird identification model to determine the bird identification result.

[0013] Optionally, the above-mentioned optimization of the bird recognition results based on the bird posture recognition results and the image quality results to obtain optimized bird recognition results includes: using the bird posture recognition results and the image quality results as optimization weights, performing confidence-weighted fusion on the bird recognition results to obtain the optimized bird recognition results.

[0014] Optionally, the above-mentioned confidence-weighted fusion of the bird recognition results using the bird posture recognition results and the image quality results as optimization weights to obtain the optimized bird recognition results includes: using the bird posture recognition results and the image quality results as optimization weights to perform confidence-weighted fusion of the bird recognition results to obtain the weighted fused bird recognition results; and using a balance factor to balance the distribution of the bird recognition results and the weighted fused bird recognition results to obtain the optimized bird recognition results.

[0015] Optionally, the above-mentioned confidence-weighted fusion of the bird recognition results using the bird pose recognition results and the image quality results as optimization weights to obtain the optimized bird recognition results includes: performing confidence-weighted fusion of the bird recognition results using the bird pose recognition results and the image quality results as optimization weights to obtain the weighted fused bird recognition results; and normalizing the weighted fused bird recognition results through an activation layer to obtain the optimized bird recognition results.

[0016] Secondly, some embodiments of this disclosure provide a bird recognition device based on bird posture and image quality. The device includes: an acquisition unit configured to acquire real-time monitoring video corresponding to a feeding area in response to detecting bird activity within that area; a video sampling unit configured to perform video sampling on the real-time monitoring video to obtain a sequence of images to be detected; an execution unit configured to perform the following processing steps on each image to be detected in the sequence of images to be detected: performing bird posture recognition on the image to be detected to generate a bird posture recognition result; determining an image quality result corresponding to the image to be detected; performing bird recognition on the image to be detected to determine a bird recognition result; optimizing the bird recognition result based on the bird posture recognition result and the image quality result to obtain an optimized bird recognition result; and a generation unit configured to generate a global bird recognition result for the real-time monitoring video based on the obtained optimized bird recognition result sequence.

[0017] Optionally, the bird recognition device based on bird posture and image quality further includes: a detection unit configured to, before acquiring real-time monitoring video corresponding to the feeding area in response to detecting bird activity in the feeding area, the method further includes: detecting heat source changes and / or pressure changes in the feeding area using an infrared sensor and / or a pressure sensor to determine whether bird activity exists in the feeding area.

[0018] Optionally, the video sampling unit is further configured to perform video downsampling on the real-time monitoring video at preset time intervals to obtain a sequence of images to be detected.

[0019] Optionally, the video sampling unit is further configured to: perform video downsampling on the real-time monitoring video at preset time intervals to obtain a candidate image sequence; for each candidate image in the candidate image sequence, perform the following filtering steps: perform image binarization processing on the candidate image to obtain a binarized image; perform dilation processing on the binarized image to obtain a dilated image; determine the image difference between the dilated image and the background image to obtain at least one difference region, wherein the background image is the image when there are no obstructions in the feeding area; in response to the existence of a difference region with a corresponding area larger than a preset area in the at least one difference region, determine the candidate image as the image to be detected.

[0020] Optionally, the execution unit is further configured to: determine the probability distribution of bird poses corresponding to the image to be detected by using a pre-trained bird pose recognition model, wherein the probability distribution of bird poses represents the confidence level of containing birds corresponding to different bird poses in the image to be detected; and determine the bird pose recognition result based on the bird pose distribution.

[0021] Optionally, the execution unit is further configured to: score the image quality of the image to be detected using a pre-trained image quality scoring model to obtain the image quality result.

[0022] Optionally, the execution unit is further configured to: perform bird recognition on the image to be detected using a pre-trained fine-grained bird recognition model, so as to determine the bird recognition result.

[0023] Optionally, the execution unit is further configured to: use the bird pose recognition result and the image quality result as optimization weights to perform confidence-weighted fusion on the bird recognition result to obtain the optimized bird recognition result.

[0024] Optionally, the execution unit is further configured to: use the bird pose recognition result and the image quality result as optimization weights to perform confidence-weighted fusion on the bird recognition result to obtain a weighted fusion bird recognition result; and use a balance factor to balance the distribution of the bird recognition result and the weighted fusion bird recognition result to obtain the optimized bird recognition result.

[0025] Optionally, the execution unit is further configured to: perform confidence-weighted fusion of the bird recognition results using the bird pose recognition results and the image quality results as optimization weights to obtain a weighted fusion bird recognition result; and perform normalization processing on the weighted fusion bird recognition result through an activation layer to obtain the optimized bird recognition result.

[0026] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation of the first aspect above.

[0027] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in any of the implementations of the first aspect above.

[0028] The various embodiments of this disclosure have the following beneficial effects: the bird recognition method based on bird posture and image quality of some embodiments of this disclosure improves the accuracy of bird recognition. Specifically, the reason for the low accuracy of bird recognition is that it is difficult to accurately identify birds due to the interference of bird posture and image quality degradation. In practice, birds have diverse posture types in the natural environment, and different posture types will cause significant fluctuations in recognition accuracy. In addition, the influence of weather (e.g., rain, snow, haze) and / or obstructions (e.g., dirt) will lead to image quality degradation. Using low image quality images for bird recognition will greatly affect the accuracy and reliability of recognition. Based on this, the bird recognition method based on bird posture and image quality of some embodiments of this disclosure firstly, in response to the detection of bird activity in the feeding area, acquires real-time monitoring video corresponding to the feeding area. By detecting bird activity in the feeding area before starting the acquisition of real-time monitoring video, unnecessary power consumption caused by long-term video acquisition can be effectively avoided, and video storage space can be saved. Secondly, the real-time monitoring video is sampled to obtain the image sequence to be detected. In practice, bird activity is continuous, reflected in real-time monitoring videos as similar bird behaviors appearing in consecutive video frames. Video downsampling can effectively reduce the amount of data processing. Next, for each image in the above-mentioned image sequence to be detected, the following processing steps are performed: First, bird pose recognition is performed on the image to generate a bird pose recognition result, thereby determining the bird pose within the image. Second, the image quality result corresponding to the image to be detected is determined, thereby quantifying the image quality. Third, bird recognition is performed on the image to be detected to determine the bird recognition result. Fourth, based on the bird pose recognition result and the image quality result, the bird recognition result is optimized to obtain an optimized bird recognition result. This optimizes the bird recognition result by weighting and fusing it from the perspectives of bird pose and image quality. Finally, based on the obtained optimized bird recognition result sequence, a global bird recognition result is generated for the above-mentioned real-time monitoring video. Finally, based on the optimized bird recognition result corresponding to each image to be detected, the bird recognition result corresponding to the real-time monitoring video is determined. This method greatly improves the accuracy of bird recognition. Attached Figure Description

[0029] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0030] Figure 1This is a flowchart of some embodiments of the bird recognition method based on bird posture and image quality according to the present disclosure;

[0031] Figure 2 This is a flowchart of some other embodiments of the bird recognition method based on bird posture and image quality according to the present disclosure;

[0032] Figure 3 This is a diagram showing the spatial relationship between the infrared sensor, the feeding area, and the birds;

[0033] Figure 4 This is a schematic diagram showing the positional relationship between the pressure sensor, the feeding area, and the birds.

[0034] Figure 5 This is a diagram showing the positional relationship between the infrared sensor, pressure sensor, feeding area, and birds.

[0035] Figure 6 This is a schematic diagram of the structure of some embodiments of a bird recognition device based on bird posture and image quality according to the present disclosure;

[0036] Figure 7 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0037] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0038] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0039] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0040] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0041] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0042] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0043] refer to Figure 1 The diagram illustrates a flow 100 of some embodiments of a bird recognition method based on bird pose and image quality according to the present disclosure. This bird recognition method based on bird pose and image quality includes the following steps:

[0044] Step 101: In response to the detection of bird activity in the feeding area, real-time monitoring video of the feeding area is collected.

[0045] In some embodiments, the execution entity (e.g., a computing device) of the bird recognition method based on bird posture and image quality can acquire real-time monitoring video corresponding to the feeding area in response to the detection of bird activity within the feeding area. The feeding area can be an area specifically for feeding birds. In practice, setting up a feeding area can increase the probability of attracting birds and thus improve the probability of birds being included in the acquired real-time monitoring video. Furthermore, wild animals often avoid humans; therefore, feeding areas are often set up in areas with low human density for acquiring real-time monitoring video. In this case, the camera used to capture the real-time monitoring video can be powered by a lithium battery or a solar cell. To ensure power efficiency, the detection of bird activity within the feeding area is used as a condition to trigger the acquisition of real-time monitoring video.

[0046] As an example, bird sensors can be placed toward the feeding area. Specifically, bird sensors can be implemented using millimeter-wave radar. By combining the echoes of radar waves emitted by the millimeter-wave radar, the presence of birds in the feeding area can be determined.

[0047] It should be noted that the aforementioned computing devices can be either hardware or software. When the computing device is hardware, it can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or as a single server or a single terminal device. When the computing device is software, it can be installed on the hardware devices listed above. It can be implemented as, for example, multiple software programs or software modules used to provide distributed services, or as a single software program or software module. No specific limitations are made here.

[0048] Step 102: Perform video sampling on the real-time monitoring video to obtain the image sequence to be detected.

[0049] In some embodiments, the aforementioned executing entity can perform video sampling on the real-time monitoring video to obtain a sequence of images to be detected. This sequence of images to be detected is an ordered sequence of images obtained by downsampling the real-time monitoring video. In practice, frame rate downsampling can be used to perform video sampling on the real-time monitoring video to obtain the sequence of images to be detected.

[0050] As an example, the video frame rate of real-time monitoring video can be 30 FPS (Frames Per Second). Therefore, the video frame rate of the monitoring video can be reduced, for example, to 15 FPS, to obtain the image sequence to be detected.

[0051] Step 103: For each image in the image sequence to be detected, perform the following processing steps:

[0052] Step 1031: Perform bird pose recognition on the image to be detected to generate bird pose recognition results.

[0053] In some embodiments, the aforementioned execution entity can perform bird pose recognition on the image to be detected to generate bird pose recognition results. The bird pose recognition results represent the confidence that the image to be detected contains a bird in different poses. In practice, these include frontal pose, lateral pose, and back pose. A frontal pose indicates that the bird is facing the camera directly, ensuring that the acquired image to be detected contains the bird's front view. A lateral pose indicates that the bird is facing the camera sideways, ensuring that the acquired image to be detected contains the bird's side view. A back pose indicates that the bird is facing away from the camera, ensuring that the acquired image to be detected contains the bird's back view. Specifically, considering the limited computing power at the edge (camera), lightweight recognition models, such as MobileNet or ResNet models, can be used to perform bird pose recognition on the image to be detected to generate bird pose recognition results. For the recognition model used for bird pose recognition, an image dataset labeled with bird poses can be collected and divided into a training sample set and a validation sample set. Furthermore, to enrich the number of samples, data augmentation methods such as flipping and cropping can be used. This improves the robustness of the subsequently trained recognition model for different poses. During model training, the cross-entropy loss function is used to optimize the model. Stochastic gradient descent or the Adam optimizer is employed to control weight updates during iteration. Furthermore, during model training or testing, the accuracy for different poses is statistically analyzed to map corresponding weights. For example, an 80% accuracy rate in identifying a bird's forward pose corresponds to a weight of 0.8. Similarly, a 95% accuracy rate in identifying a bird's sideways pose corresponds to a weight of 0.95, and a 60% accuracy rate in identifying a bird's backward pose corresponds to a weight of 0.6. In practice, the recognition model can iterate continuously based on an ever-expanding pool of training samples; therefore, the (confidence) weights corresponding to different bird poses are continuously optimized.

[0054] Step 1032: Determine the image quality result corresponding to the image to be detected.

[0055] In some embodiments, the aforementioned execution entity can determine the image quality result corresponding to the image to be detected. The image quality result characterizes whether the image to be detected is clear or whether it is occluded. In practice, the image quality result can be represented by a score value between 0 and 1; that is, the lower the score, the less clear or occluded the image to be detected. The higher the score, the clearer or unoccluded the image to be detected. Specifically, a lightweight convolutional neural network, such as the MobileNet or ResNet model, can be used to determine the image quality result corresponding to the image to be detected. Images labeled with corresponding image quality results can be collected as training and validation samples. The loss function can be the mean squared error loss function. During the training phase, an optimizer such as Adam can be used to control the weight updates of the model during the iterative process.

[0056] Step 1033: Perform bird identification on the image to be detected to determine the bird identification result.

[0057] In some embodiments, the aforementioned execution entity can perform bird identification on the image to be detected to determine the bird identification result. The bird classification result represents the classification result and corresponding confidence score for different bird types. In practice, lightweight models such as ResNet, EfficientNet, or ViT (Vision Transformer) can be used to perform bird identification on the image to be detected to determine the bird identification result. A multi-class cross-entropy loss function can be used to ensure the trained model's ability to distinguish fine-grained categories. The output format is a 1×N vector, representing the recognition confidence score for N bird categories.

[0058] Step 1034: Based on the bird posture recognition results and image quality results, optimize the bird recognition results to obtain optimized bird recognition results.

[0059] In some embodiments, the aforementioned execution entity can optimize the bird recognition results based on the bird pose recognition results and image quality results to obtain optimized bird recognition results. In practice, the aforementioned execution entity can use the maximum confidence score contained in the bird recognition results and the image quality results as weights to update the confidence scores for N bird categories contained in the bird recognition results, thereby obtaining bird recognition results containing the updated confidence scores corresponding to the bird categories. Specifically, the update method is simplified to the following formula:

[0060] P adjusted (c)=w pose ×w quality ×P model (c)

[0061] Among them, P adjusted(c) Represents the updated confidence level for bird category c. pose The maximum confidence level represented by the bird pose recognition results. quality Characterizing image quality results. P model (c) Characterize the confidence level of bird posture recognition information for bird category C.

[0062] Step 104: Based on the obtained optimized bird recognition result sequence, generate global bird recognition results for real-time monitoring video.

[0063] In some embodiments, the aforementioned executing entity can generate a global bird identification result for real-time monitoring video based on the obtained optimized bird identification result sequence. In practice, since the optimized bird identification result includes updated confidence scores for N bird categories, a statistical approach can be used to determine the bird category with the highest frequency and corresponding highest confidence score in the bird identification result sequence as the global bird identification result.

[0064] As an example, bird categories can include: c1, c2, and c3. The bird identification result sequence can include [0.1, 0.3, 0.7], [0.1, 0.2, 0.9], and [0.1, 0.2, 0.9]. Here, the confidence score of the first dimension of the bird identification result corresponds to c1, the confidence score of the second dimension corresponds to c2, and the confidence score of the third dimension corresponds to c3. Therefore, c3 is the global bird identification result.

[0065] The various embodiments of this disclosure have the following beneficial effects: the bird recognition method based on bird posture and image quality of some embodiments of this disclosure improves the accuracy of bird recognition. Specifically, the reason for the low accuracy of bird recognition is that it is difficult to accurately identify birds due to the interference of bird posture and image quality degradation. In practice, birds have diverse posture types in the natural environment, and different posture types will cause significant fluctuations in recognition accuracy. In addition, the influence of weather (e.g., rain, snow, haze) and / or obstructions (e.g., dirt) will lead to image quality degradation. Using low image quality images for bird recognition will greatly affect the accuracy and reliability of recognition. Based on this, the bird recognition method based on bird posture and image quality of some embodiments of this disclosure firstly, in response to the detection of bird activity in the feeding area, acquires real-time monitoring video corresponding to the feeding area. By detecting bird activity in the feeding area before starting the acquisition of real-time monitoring video, unnecessary power consumption caused by long-term video acquisition can be effectively avoided, and video storage space can be saved. Secondly, the real-time monitoring video is sampled to obtain the image sequence to be detected. In practice, bird activity is continuous, reflected in real-time monitoring videos as similar bird behaviors appearing in consecutive video frames. Video downsampling can effectively reduce the amount of data processing. Next, for each image in the above-mentioned image sequence to be detected, the following processing steps are performed: First, bird pose recognition is performed on the image to generate a bird pose recognition result, thereby determining the bird pose within the image. Second, the image quality result corresponding to the image to be detected is determined, thereby quantifying the image quality. Third, bird recognition is performed on the image to be detected to determine the bird recognition result. Fourth, based on the bird pose recognition result and the image quality result, the bird recognition result is optimized to obtain an optimized bird recognition result. This optimizes the bird recognition result by weighting and fusing it from the perspectives of bird pose and image quality. Finally, based on the obtained optimized bird recognition result sequence, a global bird recognition result is generated for the above-mentioned real-time monitoring video. Finally, based on the optimized bird recognition result corresponding to each image to be detected, the bird recognition result corresponding to the real-time monitoring video is determined. This method greatly improves the accuracy of bird recognition.

[0066] Further reference Figure 2 This illustrates a flowchart 200 of another embodiment of a bird recognition method based on bird posture and image quality. The flowchart 200 of this bird recognition method based on bird posture and image quality includes the following steps:

[0067] Step 201: Detect changes in heat sources and / or pressure within the feeding area using infrared sensors and / or pressure sensors to determine whether bird activity exists within the feeding area.

[0068] In some embodiments, the implementer of the bird identification method based on bird posture and image quality (e.g., a computing device) can detect changes in heat sources and / or pressure within the feeding area using infrared sensors and / or pressure sensors to determine whether bird activity exists within the feeding area.

[0069] As an example, see Figure 2 The diagram shows the positional relationship between the infrared sensor, the feeding area, and the birds. The infrared sensor 2 can be directed towards the feeding area 1 to detect changes in heat sources within the feeding area 1. When a bird 3 is present in the feeding area 1, the infrared sensor 1 will detect the presence of a heat source, thus indicating bird activity within the feeding area 1.

[0070] As yet another example, see Figure 4 The diagram shows the positional relationship between the pressure sensor, the feeding area, and the birds. A pressure sensor 4 is installed below the feeding area 1. When there are birds 3 in the feeding area 1, the pressure sensor 4 can detect the pressure change in the feeding area 1, which can indicate the presence of bird activity in the feeding area 1.

[0071] As another example, see Figure 5 The diagram shows the positional relationship between the infrared sensor, pressure sensor, feeding area, and birds. Infrared sensor 2 is positioned towards feeding area 1 to detect changes in heat sources within feeding area 1. Pressure sensor 4 is positioned below feeding area 1 to detect changes in pressure within feeding area 1. When birds 3 are present in feeding area 1, infrared sensor 2 will detect changes in heat sources, and pressure sensor 4 will detect changes in pressure, thus indicating the presence of bird activity in feeding area 1.

[0072] Step 202: In response to the detection of bird activity in the feeding area, real-time monitoring video of the feeding area is collected.

[0073] In some embodiments, the specific implementation of step 202 and its resulting technical effects can be found in [reference needed]. Figure 1 Step 102 in the corresponding embodiment will not be repeated here.

[0074] Step 203: Downsample the real-time monitoring video at preset time intervals to obtain the image sequence to be detected.

[0075] In some embodiments, the aforementioned execution entity can downsample the real-time monitoring video at preset time intervals to obtain a sequence of images to be detected. In practice, the preset time interval is shorter than the acquisition duration of the real-time monitoring video. For example, the acquisition duration of the real-time monitoring video can be 120 seconds. The preset time interval can be 5 seconds, that is, frames are sampled from the real-time monitoring video every 5 seconds to obtain the sequence of images to be detected.

[0076] In some optional implementations of certain embodiments, the execution entity performs video sampling on the real-time monitoring video to obtain a sequence of images to be detected, including:

[0077] The first step is to downsample the real-time monitoring video at preset time intervals to obtain a candidate image sequence.

[0078] In practice, due to the randomness of bird movement, i.e., when video is downsampled at fixed time intervals, the candidate images in the resulting candidate image sequence may not contain birds.

[0079] The second step involves performing the following filtering steps for each candidate image in the above candidate image sequence:

[0080] The first sub-step involves performing image binarization on the candidate images to obtain the binarized images.

[0081] In practice, binarization can reduce the amount of subsequent data processing and alleviate the data processing pressure on the edge (camera).

[0082] The second sub-step involves dilating the binarized image to obtain the dilated image.

[0083] In practice, the Canndy operator can be used to dilate the binarized image to obtain the dilated image. Specifically, during the binarization process, a threshold needs to be set for image binarization. If the threshold is not selected properly, the binarization may not be accurate enough, especially for small areas with indistinct boundaries. After binarization, the area boundaries may shrink. Dilation can restore the area boundaries to some extent.

[0084] The third sub-step involves determining the image difference between the dilated image and the background image to obtain at least one difference set region.

[0085] The background image is an image taken when there are no obstructions within the feeding area. Specifically, the background image can be a binarized image of the feeding area when there are no obstructions. At least one difference region can represent at least one obstruction.

[0086] The fourth sub-step involves determining the candidate image as the image to be detected in response to the existence of a difference region with a corresponding area larger than a preset area in at least one difference region.

[0087] In practice, the difference region can represent bird occlusion or dirt occlusion. However, birds often have a certain size, so the corresponding difference region also has a certain area. Therefore, by setting a preset region area as a threshold, candidate images containing only small difference regions or not containing any difference regions are filtered out. That is, only candidate images that may contain birds are retained as the images to be detected. This method achieves further image filtering based on image sampling at fixed time intervals.

[0088] Step 204: For each image in the image sequence to be detected, perform the following processing steps:

[0089] Step 2041: Perform bird pose recognition on the image to be detected to generate bird pose recognition results.

[0090] In some embodiments, the execution entity performs bird pose recognition on the image to be detected to generate a bird pose recognition result, which may include the following steps:

[0091] The first step is to determine the probability distribution of bird poses corresponding to the above-mentioned images to be detected by using a pre-trained bird pose recognition model.

[0092] The aforementioned bird pose probability distribution represents the confidence level of containing different bird poses within the image to be detected. In practice, considering the limited computing power at the edge (camera) and the fact that birds occupy a certain portion of the frame when they appear in the image to be detected, the bird pose recognition model can be constructed using five serially connected convolutional layers. Specifically, the bird pose recognition model can include: convolutional layer A, convolutional layer B, convolutional layer C, convolutional layer D, convolutional layer E, and fully connected layer F. Convolutional layers A, B, and C each contain three parallel-connected convolutional layers and one fusion layer. Taking convolutional layer A as an example, convolutional layer A includes: convolutional layer A1, convolutional layer A2, convolutional layer A3, and fusion layer A4. The kernel of convolutional layer A1 is 1×1. The kernel of convolutional layer A2 is 3×3. The kernel of convolutional layer A3 is 5×5. The fusion layer A4 is used to superimpose the three features output by convolutional layers A1, A2, and A3. Feature extraction under different receptive fields is achieved by setting convolutional layers A1, A2, and A3. The feature vector output by the fully connected layer has a dimension of 1×3, which includes the confidence scores for forward, lateral, and backward poses.

[0093] The second step is to determine the bird posture recognition results based on the above bird posture distribution.

[0094] In practice, the aforementioned implementing entities can use the bird posture corresponding to the highest confidence level in the bird posture distribution as the bird posture recognition result.

[0095] Step 2042: Use a pre-trained image quality scoring model to score the image quality of the image to be detected and obtain the image quality result.

[0096] In some embodiments, the aforementioned execution entity can use a pre-trained image quality scoring model to score the image to be detected, thereby obtaining an image quality result. In practice, considering that both bird pose recognition and image quality scoring are based on the image to be detected, meaning that image features are reused in both processes, to further alleviate the data processing pressure at the edge (camera), the image quality scoring model may include: convolutional layer A, convolutional layer B, convolutional layer C, convolutional layer G, convolutional layer H, convolutional layer I, fully connected layer J, and fully connected layer H. The image quality scoring model shares convolutional layers A, B, and C with the bird pose recognition model. Convolutional layers D and E are deep feature extractions of the output features of convolutional layer C, with parameters primarily targeting bird pose recognition; therefore, convolutional layers G, H, and I are also selected in the image quality scoring model. Furthermore, considering that the image quality result is a 1×1 vector, corresponding to values ​​between 0 and 1. Therefore, fully connected layers J and H are set to transform the feature dimension of the output of convolutional layer I to 1×1.

[0097] Step 2043: Using a pre-trained fine-grained bird recognition model, perform bird recognition on the image to be detected to determine the bird recognition result.

[0098] In some embodiments, the aforementioned execution entity can use a pre-trained fine-grained bird recognition model to perform bird recognition on the image to be detected, thereby determining the bird recognition result. In practice, further considering the computational pressure at the edge (camera), a lightweight fine-grained bird recognition model and a non-lightweight fine-grained bird recognition model can be pre-set. The execution entity can adaptively select either a lightweight or non-lightweight fine-grained bird recognition model to perform bird recognition on the image to be detected, depending on the current complexity, to determine the bird recognition result. Specifically, the lightweight fine-grained bird recognition model can use the EfficientNet-B1 model as the backbone network. The non-lightweight fine-grained bird recognition model can use the EfficientNet-B7 model as the backbone network. To further improve feature processing efficiency and reduce computational pressure at the edge, the input to the fine-grained bird recognition model is the feature map after feature processing through convolutional layers A, B, and C. This approach can significantly improve the efficiency of feature utilization and processing. Furthermore, during the training phase, since the bird pose recognition model, image quality scoring model, and fine-grained bird recognition model share common model components (convolutional layer A, convolutional layer B, and convolutional layer C), in order to improve model training efficiency, the bird pose recognition model, image quality scoring model, and fine-grained bird recognition model are trained in parallel. When updating convolutional layers A, B, and C, a voting method is used to update the weights.

[0099] Step 2044: Using the bird pose recognition result and image quality result as optimization weights, perform confidence-weighted fusion on the bird recognition result to obtain the optimized bird recognition result.

[0100] In some embodiments, the aforementioned executing entity can use the bird pose recognition result and image quality result as optimization weights to perform confidence-weighted fusion on the bird recognition result to obtain the optimized bird recognition result. Specifically, refer to the formula corresponding to the update method shown in step 1034. However, considering that the maximum confidence value range of the bird pose recognition result is 0-1, and the range of the image quality result is 0-1, using the maximum confidence value of the bird pose recognition result and the image quality result as weights may result in a decrease in the updated confidence value of the optimized bird recognition result by an order of magnitude, exhibiting extremely low confidence. Therefore, for the formula corresponding to the update method in step 1034, the image quality result can be increased by ten times. This can offset the problem of the order of magnitude reduction to some extent.

[0101] In some optional implementations of certain embodiments, the execution entity performs confidence-weighted fusion of the bird recognition results using the bird pose recognition results and the image quality results as optimization weights to obtain the optimized bird recognition results, including:

[0102] The first step is to use the bird pose recognition results and the image quality results as optimization weights to perform confidence-weighted fusion on the bird recognition results, and obtain the weighted fusion bird recognition results.

[0103] The second step is to balance the distribution of the bird identification results and the bird identification results after weight fusion by using a balance factor, so as to obtain the optimized bird identification results.

[0104] In practice, the aforementioned implementing entity can use the maximum confidence score, image quality result, and balance factor contained in the bird identification result as weights to update the confidence scores for N bird categories in the bird identification result, thus obtaining a bird identification result containing the updated confidence scores corresponding to the bird categories. See the following formula for details:

[0105] P adjusted (c)=(1-α)×(w pose ×w quality ×P model (c))+α×P model (c)

[0106] Among them, P adjusted (c) Represents the updated confidence level for bird category c. pose The maximum confidence level represented by the bird pose recognition results. quality Characterizing image quality results. P model (c) Characterizes the confidence level of bird posture recognition information for bird category C. represents the balance factor.

[0107] In some optional implementations of certain embodiments, the execution entity performs confidence-weighted fusion of the bird recognition results using the bird pose recognition results and the image quality results as optimization weights to obtain the optimized bird recognition results, including:

[0108] The first step is to use the bird pose recognition results and the image quality results as optimization weights to perform confidence-weighted fusion on the bird recognition results, and obtain the weighted fusion bird recognition results.

[0109] In practice, the formula corresponding to the update method shown in step 1034 can be used for weighted fusion, or the formula corresponding to the update method in step 2044 can be used for weighted fusion.

[0110] The second step is to normalize the weighted fusion bird recognition results through the activation layer to obtain the optimized bird recognition results.

[0111] In practice, the bird recognition results after weighted fusion are normalized using the Softmax function of the activation layer to obtain the optimized bird recognition results. This involves remapping the confidence scores to a range of 0-1 with a reduced confidence level.

[0112] Step 205: Based on the obtained optimized bird recognition result sequence, generate global bird recognition results for the real-time monitoring video.

[0113] In some embodiments, the specific implementation of step 205 and its resulting technical effects can be found in [reference needed]. Figure 1 Step 105 in the corresponding embodiment will not be repeated here.

[0114] from Figure 2 It can be seen that, with Figure 1 Compared to the descriptions of some corresponding embodiments, this disclosure further optimizes the model structures of the bird pose recognition model, image quality scoring model, and fine-grained bird recognition model while minimizing the computational burden on the edge (camera). By sharing the model structure, redundant feature processing is avoided, alleviating the computational burden on the edge. Furthermore, by setting a balancing factor or activation layer, the problem of a decrease in confidence level after weight optimization is overcome. This further ensures the accuracy of bird recognition.

[0115] Further reference Figure 6 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of a bird recognition device based on bird posture and image quality. These device embodiments are similar to... Figure 1 Corresponding to the method embodiments shown, this bird recognition device based on bird posture and image quality can be specifically applied to various electronic devices.

[0116] like Figure 6As shown, a bird recognition device 600 based on bird posture and image quality in some embodiments includes: an acquisition unit 601, a video sampling unit 602, an execution unit 603, and a generation unit 604. The acquisition unit 601 is configured to acquire real-time monitoring video corresponding to the feeding area in response to detecting bird activity within the feeding area; the video sampling unit 602 is configured to perform video sampling on the real-time monitoring video to obtain a sequence of images to be detected; the execution unit 603 is configured to perform the following processing steps for each image to be detected in the sequence of images to be detected: performing bird posture recognition on the image to be detected to generate a bird posture recognition result; determining the image quality result corresponding to the image to be detected; performing bird recognition on the image to be detected to determine a bird recognition result; optimizing the bird recognition result based on the bird posture recognition result and the image quality result to obtain an optimized bird recognition result; and the generation unit 604 is configured to generate a global bird recognition result for the real-time monitoring video based on the obtained optimized bird recognition result sequence.

[0117] In some optional implementations of some embodiments, the bird recognition device based on bird posture and image quality further includes: a detection unit (not shown in the figure), configured to, before acquiring real-time monitoring video corresponding to the feeding area in response to detecting bird activity in the feeding area, the method further includes: detecting heat source changes and / or pressure changes in the feeding area using an infrared sensor and / or a pressure sensor to determine whether bird activity exists in the feeding area.

[0118] In some optional implementations of some embodiments, the video sampling unit 602 is further configured to: perform video downsampling on the real-time monitoring video at preset time intervals to obtain a sequence of images to be detected.

[0119] In some optional implementations of certain embodiments, the video sampling unit 602 is further configured to: perform video downsampling on the real-time monitoring video at preset time intervals to obtain a candidate image sequence; for each candidate image in the candidate image sequence, perform the following filtering steps: perform image binarization processing on the candidate image to obtain a binarized image; perform dilation processing on the binarized image to obtain a dilated image; determine the image difference between the dilated image and the background image to obtain at least one difference region, wherein the background image is an image in the feeding area when there are no obstructions; in response to the existence of a difference region with a corresponding area larger than a preset area in the at least one difference region, determine the candidate image as an image to be detected.

[0120] In some optional implementations of some embodiments, the execution unit 603 is further configured to: determine the probability distribution of bird postures corresponding to the image to be detected by using a pre-trained bird posture recognition model, wherein the probability distribution of bird postures represents the confidence that the image to be detected contains birds corresponding to different bird postures; and determine the bird posture recognition result based on the bird posture distribution.

[0121] In some optional implementations of some embodiments, the execution unit 603 is further configured to: perform image quality scoring on the image to be detected using a pre-trained image quality scoring model to obtain the image quality result.

[0122] In some alternative implementations of some embodiments, the execution unit 603 is further configured to: perform bird recognition on the image to be detected using a pre-trained fine-grained bird recognition model to determine the bird recognition result.

[0123] In some optional implementations of some embodiments, the execution unit 603 is further configured to: perform confidence-weighted fusion of the bird recognition results with the bird posture recognition results and the image quality results as optimization weights to obtain the optimized bird recognition results.

[0124] In some optional implementations of some embodiments, the execution unit 603 is further configured to: perform confidence-weighted fusion of the bird recognition results with the bird pose recognition results and the image quality results as optimization weights to obtain a weighted fused bird recognition result; and perform distribution balancing of the bird recognition results and the weighted fused bird recognition results through a balancing factor to obtain the optimized bird recognition result.

[0125] In some optional implementations of some embodiments, the execution unit 603 is further configured to: perform confidence-weighted fusion on the bird recognition results using the bird pose recognition results and the image quality results as optimization weights to obtain a weighted fused bird recognition result; and perform normalization processing on the weighted fused bird recognition result through an activation layer to obtain the optimized bird recognition result.

[0126] It is understandable that the units described in the bird recognition device 600 based on bird posture and image quality are similar to those in the reference device. Figure 1 The steps in the described method correspond accordingly. Therefore, the operations, features, and beneficial effects described above for the method are also applicable to the bird recognition device 600 based on bird posture and image quality, and the units contained therein, and will not be repeated here.

[0127] The following is for reference. Figure 7It shows a schematic diagram of the structure of an electronic device (e.g., a computing device) 700 suitable for implementing some embodiments of the present disclosure. Figure 7 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0128] like Figure 7 As shown, the electronic device 700 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory 702 or a program loaded from a storage device 708 into a random access memory 703. The random access memory 703 also stores various programs and data required for the operation of the electronic device 700. The processing unit 701, the read-only memory 702, and the random access memory 703 are interconnected via a bus 704. An input / output interface 705 is also connected to the bus 704.

[0129] Typically, the following devices can be connected to the input / output interface 705: input devices 706 including, for example, a touchscreen, touchpad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; output devices 707 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 708 including, for example, magnetic tape, hard disk, etc.; and communication devices 709. Communication device 709 allows electronic device 700 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 7 An electronic device 700 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 7 Each box shown can represent a device or multiple devices as needed.

[0130] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 709, or installed from a storage device 708, or installed from a read-only memory 702. When the computer program is executed by the processing device 701, it performs the functions defined in the methods of some embodiments of this disclosure.

[0131] It should be noted that, in some embodiments of this disclosure, the computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0132] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and end-to-end networks (e.g., ad hoc end-to-end networks), as well as any currently known or future-developed networks.

[0133] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to: in response to detecting bird activity within a feeding area, acquire real-time monitoring video corresponding to the feeding area; perform video sampling on the real-time monitoring video to obtain a sequence of images to be detected; for each image to be detected in the sequence of images to be detected, perform the following processing steps: perform bird posture recognition on the image to be detected to generate a bird posture recognition result; determine the image quality result corresponding to the image to be detected; perform bird recognition on the image to be detected to determine a bird recognition result; optimize the bird recognition result based on the bird posture recognition result and the image quality result to obtain an optimized bird recognition result; and generate a global bird recognition result for the aforementioned real-time monitoring video based on the obtained optimized bird recognition result sequence.

[0134] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0135] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0136] The units described in some embodiments of this disclosure can be implemented in software or hardware. The described units can also be housed in a processor; for example, a processor may be described as including an acquisition unit, a video sampling unit, an execution unit, and a generation unit. The names of these units do not necessarily limit the specific unit; for example, an acquisition unit may also be described as "a unit that acquires real-time monitoring video corresponding to a feeding area in response to detecting bird activity within that feeding area."

[0137] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0138] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A bird recognition method based on bird posture and image quality, comprising: In response to the detection of bird activity in the feeding area, real-time monitoring video of the feeding area is collected; The real-time monitoring video is sampled to obtain a sequence of images to be detected; For each image in the image sequence to be detected, the following processing steps are performed: Bird pose recognition is performed on the image to be detected to generate bird pose recognition results, wherein the bird pose recognition results represent the confidence level that the image to be detected contains birds in different poses; Determine the image quality result corresponding to the image to be detected; Bird identification is performed on the image to be detected to determine the bird identification result; Based on the bird posture recognition results and the image quality results, the bird recognition results are optimized to obtain optimized bird recognition results. The maximum confidence score contained in the bird posture recognition results and the image quality results are used as weights to update the confidence scores for N bird categories in the bird recognition results. Based on the obtained optimized bird recognition result sequence, a global bird recognition result is generated for the real-time monitoring video.

2. The method according to claim 1, wherein, Before acquiring real-time monitoring video of the feeding area in response to detecting bird activity within the feeding area, the method further includes: Changes in heat sources and / or pressure in the feeding area are detected using infrared sensors and / or pressure sensors to determine whether bird activity exists in the feeding area.

3. The method according to claim 1, wherein, The step of sampling the real-time monitoring video to obtain the image sequence to be detected includes: The real-time monitoring video is downsampled at preset time intervals to obtain a sequence of images to be detected.

4. The method according to claim 1, wherein, The step of sampling the real-time monitoring video to obtain the image sequence to be detected includes: At preset time intervals, the real-time monitoring video is downsampled to obtain a candidate image sequence; For each candidate image in the candidate image sequence, perform the following filtering steps: The candidate image is subjected to image binarization processing to obtain the binarized image; The binarized image is then dilated to obtain the dilated image. Determine the image difference between the dilated image and the background image to obtain at least one difference set region, wherein the background image is the image when there are no obstructions in the feeding area; In response to the existence of a difference region with a corresponding area greater than a preset area in the at least one difference region, the candidate image is determined as the image to be detected.

5. The method according to claim 1, wherein, The step of performing bird pose recognition on the image to be detected to generate bird pose recognition results includes: By using a pre-trained bird pose recognition model, the probability distribution of bird poses corresponding to the image to be detected is determined, wherein the probability distribution of bird poses represents the confidence that the image to be detected contains birds corresponding to different bird poses. The bird posture recognition result is determined based on the bird posture distribution.

6. The method according to claim 1, wherein, Determining the image quality result corresponding to the image to be detected includes: The image quality of the image to be detected is scored using a pre-trained image quality scoring model to obtain the image quality result.

7. The method according to claim 1, wherein, The step of performing bird identification on the image to be detected to determine the bird identification result includes: The bird identification result is determined by using a pre-trained fine-grained bird recognition model to identify birds in the image to be detected.

8. The method according to claim 1, wherein, The step of optimizing the bird recognition result based on the bird posture recognition result and the image quality result to obtain the optimized bird recognition result includes: Using the bird pose recognition result and the image quality result as optimization weights, the bird recognition result is weighted and fused with confidence to obtain the optimized bird recognition result.

9. The method according to claim 8, wherein, The optimized bird recognition result is obtained by performing a confidence-weighted fusion of the bird recognition result and the image quality result as optimization weights, including: Using the bird pose recognition result and the image quality result as optimization weights, the bird recognition result is weighted and fused with confidence to obtain the weighted fused bird recognition result. By using a balancing factor, the distribution of the bird identification results and the weighted bird identification results is balanced to obtain the optimized bird identification results.

10. The method according to claim 8, wherein, The optimized bird recognition result is obtained by performing a confidence-weighted fusion of the bird recognition result and the image quality result as optimization weights, including: Using the bird pose recognition result and the image quality result as optimization weights, the bird recognition result is weighted and fused with confidence to obtain the weighted fused bird recognition result. The bird recognition result after weighted fusion is normalized by activating the layer to obtain the optimized bird recognition result.

11. A bird recognition device based on bird posture and image quality, comprising: The acquisition unit is configured to acquire real-time monitoring video of the feeding area in response to the detection of bird activity in the feeding area; The video sampling unit is configured to perform video sampling on the real-time monitoring video to obtain a sequence of images to be detected; The execution unit is configured to perform the following processing steps for each image to be detected in the image sequence: perform bird pose recognition on the image to be detected to generate a bird pose recognition result, wherein the bird pose recognition result characterizes the confidence that the image to be detected contains a bird in different poses; and determine the image quality result corresponding to the image to be detected. Bird identification is performed on the image to be detected to determine the bird identification result; based on the bird posture identification result and the image quality result, the bird identification result is optimized to obtain the optimized bird identification result, wherein the maximum confidence score contained in the bird posture identification result and the image quality result are used as weights to update the confidence scores for N bird categories contained in the bird identification result; The generation unit is configured to generate global bird recognition results for the real-time monitoring video based on the obtained optimized bird recognition result sequence.

12. An electronic device, comprising: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 10.

13. A computer-readable medium having a computer program stored thereon, wherein, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 10.

Images