Bird identification method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of image recognition, in particular to a bird recognition method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a to-be-recognized image; inputting the to-be-recognized image into a pre-trained bird recognition model, and outputting a bird recognition result; taking the bird recognition result as a target bird recognition result corresponding to the to-be-recognized image; wherein the bird identification model comprises an illumination transformation network based on a U-Net structure and a bird identification network based on a ResNet-50 model, which are connected in sequence. The model in the bird recognition method provided by the invention comprises the illumination transformation network for converting the to-be-recognized image into the specified illumination recognition image, so that image processing is performed before image recognition, the to-be-recognized image has the specified illumination condition, the influence of the illumination condition of the image on image recognition is reduced, and the recognition efficiency is improved. The phenomenon of low image recognition precision during low image illuminance or overexposure is avoided, and the accuracy of bird recognition is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular to a bird recognition method, device, electronic equipment and storage medium. Background Art

[0002] Nowadays, pet feeders with AI functions have become mainstream. Bird recognition is a basic computer vision task for pet AI feeders. The detection and recognition of images captured under daily light through algorithms has gradually matured. However, it is still a difficult problem to recognize images captured under low light conditions or overexposure. Summary of the invention

[0003] In view of this, an object of the present invention is to provide a bird identification method, device, electronic device, and storage medium.

[0004] In a first aspect, an embodiment of the present invention provides a bird identification method, the method comprising: Obtain an image to be recognized; Input the image to be identified into the pre-trained bird identification model and output the bird identification result; The bird recognition result is used as the target bird recognition result corresponding to the image to be recognized; Among them, the bird recognition model includes an illumination transformation network based on a U-Net structure and a bird recognition network based on a ResNet-50 model, which are connected in sequence; the illumination transformation network based on the U-Net structure is used to convert the image to be recognized into a specified illumination recognition image with specified illumination conditions.

[0005] In conjunction with the first aspect, before the step of obtaining the image to be recognized, the method includes: Constructing an initial recognition model, which includes an illumination transformation network based on a U-Net structure and a bird recognition network based on a ResNet-50 model connected in sequence; Obtain an initial sample set; The initial sample set is used to train and optimize the illumination transformation network based on the U-Net structure, and the training recognition image with the specified illumination intensity is output; Use the training recognition images to train the bird recognition network based on the ResNet-50 model and output the bird prediction results; The trained bird recognition model is obtained by combining the trained illumination transformation network based on the U-Net structure and the trained bird recognition network based on the ResNet-50 model.

[0006] Combined with the first aspect, the illumination transformation network based on the U-Net structure includes an HSV color model and a reflectance estimation network based on the U-Net structure; The steps of training and optimizing the illumination transformation network based on the U-Net structure by using the initial sample set and outputting the training recognition image with the specified illumination intensity include: Inputting the initial sample set into the HSV color model for format conversion, so as to convert each initial sample in the initial sample set into an HSV image; Add brightness disturbance to the V channel parameters in the HSV image to obtain updated V channel parameters; Inputting the V channel parameter and the updated V channel parameter into the reflectivity estimation network at the same time, and outputting a first reflectivity corresponding to the V channel parameter and a second reflectivity corresponding to the updated V channel parameter; Calculate a first loss function by combining the V channel parameter, the first reflectivity, the second reflectivity, and the preset hyperparameters; Based on the first loss function, the reflectivity estimation network based on the U-Net structure is optimized to output the target V channel parameters; Combine the target V channel parameters, the H channel parameters and the S channel parameters in the HSV image to obtain the target HSV image; The target HSV image is input into the HSV color model for inverse format conversion, and a training recognition image with specified light intensity is output.

[0007] In combination with the first aspect, the step of calculating the first loss function in combination with the V channel parameter, the first reflectivity, the second reflectivity and the preset hyperparameters includes: ; ; ; ; ; in, is the first loss function, is the V channel loss function, is the reflectivity loss function, is the preset enhanced constraint loss; is the intensity of the incident light; is the first reflectivity; is the second reflectivity; For the general and Average reflectance after pooling; are the preset hyperparameters.

[0008] In combination with the first aspect, the steps of training the bird recognition network based on the ResNet-50 model using the training recognition images and outputting the bird prediction results include: Label the training recognition images for birds to obtain a labeled training image set; Divide the training image set into a training set and a validation set; The training image is input into the bird recognition network based on the ResNet-50 model, and the convolutional neural network in the bird recognition network based on the ResNet-50 model generates a feature map; Pass the feature map to the fully connected layer and then to the Softmax layer to output the bird prediction result; Use the validation set to verify the bird prediction results; If the verification fails, the bird recognition network based on the ResNet-50 model is optimized based on the preset loss function to obtain a trained bird recognition network based on the ResNet-50 model.

[0009] In combination with the first aspect, after the step of verifying the bird prediction results using the verification set, it also includes: If the verification is successful, the current bird recognition network based on the ResNet-50 model is determined as the trained bird recognition network based on the ResNet-50 model.

[0010] Combined with the first aspect, the preset loss function is: ; Among them, L is the preset loss function, N is the number of samples, The i-th sample belongs to The predicted probability of the class.

[0011] In a second aspect, the present invention provides a bird identification device, the device comprising: An acquisition module, used for acquiring an image to be recognized; The recognition module is used to input the image to be recognized into the pre-trained bird recognition model and output the bird recognition result; A determination module, used for taking the bird recognition result as the target bird recognition result corresponding to the image to be recognized; Among them, the bird recognition model includes an illumination transformation network based on a U-Net structure and a bird recognition network based on a ResNet-50 model, which are connected in sequence; the illumination transformation network based on the U-Net structure is used to convert the image to be recognized into a specified illumination recognition image with specified illumination conditions.

[0012] In a third aspect, the present invention provides an electronic device, the electronic device comprising a memory and a processor, the memory being used to store a computer program, and the processor running the computer program to enable the electronic device to execute the above method.

[0013] In a fourth aspect, the present invention provides a readable storage medium, in which computer program instructions are stored. When the computer program instructions are read and executed by a processor, the above method is executed.

[0014] The embodiments of the present invention bring the following beneficial effects: the bird identification method, device, electronic device, and storage medium provided by the present invention, the method comprising: obtaining an image to be identified; inputting the image to be identified into a pre-trained bird identification model, and outputting a bird identification result; using the bird identification result as a target bird identification result corresponding to the image to be identified; wherein the bird identification model comprises an illumination transformation network based on a U-Net structure and a bird identification network based on a ResNet-50 model connected in sequence; the illumination transformation network based on the U-Net structure is used to convert the image to be identified into a specified illumination identification image with specified illumination conditions.

[0015] The model in the bird identification method provided by the present invention includes an illumination transformation network for converting an image to be identified into an image identified under specified illumination, so as to perform image processing before image recognition so that the image to be identified has specified illumination conditions, thereby reducing the influence of the illumination conditions of the image on image recognition, avoiding the phenomenon of low image recognition accuracy when the image illumination is low or overexposed, and improving the accuracy of bird identification.

[0016] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0017] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 A flow chart of a bird identification method provided by an embodiment of the present invention; Figure 2 A schematic diagram of an illumination transformation network structure based on a U-Net structure in a bird recognition method provided by an embodiment of the present invention; Figure 3A schematic diagram of the structure of a bird recognition network based on the ResNet-50 model in the bird recognition method provided in an embodiment of the present invention; Figure 4 A schematic diagram of the structure of a bird identification device provided by an embodiment of the present invention; Figure 5 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention.

[0020] Reference numerals: 10 - acquisition module, 20 - identification module, 30 - determination module.

[0021] 130 - processor, 131 - memory, 132 - bus, 133 - communication interface. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0023] In order to facilitate understanding of this embodiment, the technical terms used in the design of the present invention are briefly introduced below.

[0024] The HSV (Hue, Saturation, Value) color model is an intuitive representation method based on human color perception. It divides color information into three independent components: hue (Hue, H), saturation (Saturation, S) and brightness / brightness (Value, V).

[0025] Hue (H) represents the basic attributes of color, namely red, orange, yellow, green, cyan, blue, purple, etc. In the HSV color model, hue is usually expressed as an angle value ranging from 0 to 360 degrees, where red corresponds to 0 or 360 degrees, green corresponds to 120 degrees, and blue corresponds to 240 degrees.

[0026] Saturation (S) is used to describe the purity or intensity of a color, that is, how much white is mixed in with the color. A saturation of 0% means there is no color at all, and the image appears grayscale; while 100% represents the purest color with no white light mixed in.

[0027] Brightness / Volume (Value, V) is used to reflect the brightness of a color, or the proportion of black in a color. When the V value is 0, the color of that point is black regardless of the values ​​of H and S; the higher the V value, the brighter the color, until it reaches the maximum value of 100%, at which point the color is the brightest.

[0028] After introducing the technical terms involved in the present invention, the application scenarios and design concepts of the embodiments of the present invention are briefly introduced.

[0029] The existing bird identification work is affected by light intensity, resulting in low identification accuracy.

[0030] Based on this, an embodiment of the present invention provides a bird identification method, device, electronic device, and storage medium.

[0031] Example 1 The present invention provides a bird identification method, combining Figure 1 As shown, the method includes: S110, obtaining an image to be recognized.

[0032] S120, inputting the image to be identified into a pre-trained bird identification model, and outputting a bird identification result.

[0033] S130: Using the bird recognition result as the target bird recognition result corresponding to the image to be recognized.

[0034] Among them, the bird recognition model includes an illumination transformation network based on a U-Net structure and a bird recognition network based on a ResNet-50 model, which are connected in sequence. The illumination transformation network based on the U-Net structure is used to convert the image to be recognized into a specified illumination recognition image with specified illumination conditions.

[0035] In this embodiment, during the bird recognition process, the pre-trained bird recognition model first performs illumination transformation processing on the image to be recognized and then performs category recognition under specified lighting conditions to obtain the target bird recognition result. In this way, the influence of low light or overexposure conditions on image recognition can be avoided, thereby improving the image recognition capability of the model and improving recognition efficiency.

[0036] Among them, the data sources of the image to be identified in step S110 can be multiple. As an implementable method, the image to be identified is the original image acquired by the acquisition end, or the RGB format image after preprocessing the acquired original image; as another implementable method, the identification image is an image transmitted by other mobile terminals through wireless radio frequency, Bluetooth, wired communication, 4G or 5G mobile communication, etc.; as another implementable method, it is an image downloaded based on the Internet of Things; as another implementable method, the image to be identified can also be obtained by intercepting the video frame based on the acquired video. The above examples are all implementable and are not limited here.

[0037] In combination with the first aspect, before step S110, the method includes: S010, constructing an initial recognition model, which includes an illumination transformation network based on a U-Net structure and a bird recognition network based on a ResNet-50 model, which are connected in sequence.

[0038] S020, obtaining an initial sample set.

[0039] S030, using the initial sample set to train and optimize the illumination transformation network based on the U-Net structure, and output a training recognition image with a specified illumination intensity.

[0040] S040, using the training recognition images to train the bird recognition network based on the ResNet-50 model, and outputting the bird prediction results.

[0041] S050, optimizing the bird recognition network based on the ResNet-50 model based on a preset loss function, and obtaining a trained bird recognition network based on the ResNet-50 model.

[0042] S060, combining the trained U-Net-based illumination transformation network and the trained bird recognition network to obtain a trained bird recognition model.

[0043] It is understandable that before actual application, an initial bird recognition model needs to be constructed and trained until a trained bird recognition model that meets preset requirements is obtained. In this way, in actual application, the image to be recognized can be input into the trained bird recognition model to perform bird recognition and output the recognition result.

[0044] The initial sample set in step S020 is obtained in a similar way to the image to be identified in step S110 , and may be obtained by any one or more of the above-mentioned methods, which are not limited here.

[0045] Specifically, the illumination transformation network based on the U-Net structure is first trained, and the first loss function is calculated based on the output transformed illumination recognition image, and the first loss function is used to optimize until the trained illumination transformation network is obtained. In this way, the illumination transformation of the image to be recognized before bird recognition is realized through the trained illumination transformation network, so as to obtain a training recognition image with specified illumination conditions, that is, a specified illumination image. Through the data processing capability of the model, it is possible to realize fast and accurate illumination transformation, which can effectively avoid the influence of insufficient illumination intensity and over-exposure of the image to be recognized on the recognition process, which is conducive to improving the recognition accuracy.

[0046] Afterwards, the bird recognition network based on the ResNet-50 model is trained and optimized using training recognition images with specified light intensity until a trained bird recognition network based on the ResNet-50 model that meets the requirements is obtained, thereby obtaining a fully trained bird recognition model.

[0047] In combination with the first aspect, step S030 includes: S031, inputting the initial sample set into the HSV color model for format conversion, so as to convert each initial sample in the initial sample set into an HSV image.

[0048] S032, adding brightness disturbance to the V channel parameters in the HSV image to obtain updated V channel parameters.

[0049] S033, inputting the V channel parameter and the updated V channel parameter into the reflectivity estimation network at the same time, and outputting a first reflectivity corresponding to the V channel parameter and a second reflectivity corresponding to the updated V channel parameter.

[0050] S034, calculating a first loss function by combining the V channel parameter, the first reflectivity, the second reflectivity, and the preset hyperparameters.

[0051] S035, optimizing the reflectivity estimation network based on the U-Net structure based on the first loss function, and outputting the target V channel parameters.

[0052] S036, combining the target V channel parameters, the H channel parameters and the S channel parameters in the HSV image to obtain the target HSV image.

[0053] S037, input the target HSV image into the HSV color model for format inverse conversion, and output a training recognition image with a specified light intensity.

[0054] Among them, the HSV color model is used to convert an image in RGB format to an HSV image, or to inversely convert an HSV image to an image in RGB format. In this embodiment, converting the initial sample in RGB format to an HSV image through the HSV color model helps to more naturally process certain types of image operations, such as color adjustment, object detection, and feature extraction in computer vision tasks. For example, when performing a hue transformation, only the H channel needs to be modified without changing the other two channels at the same time; similarly, the overall contrast and brightness of the image can be enhanced or weakened by adjusting the S or V channel.

[0055] It can be understood that according to the Retinex theory, the brightness of the image is determined by the intensity of the incident light and the reflectivity of the object, and the reflectivity of the same object is the same. Therefore, when the training samples to be identified are adjusted to a specified lighting parameter, the robustness of bird identification is optimal. Based on this, in order to avoid the image brightness being too low or overexposed in the present invention, the parameters of the V channel need to be adjusted, so the image is first converted into the HSV format for easy operation.

[0056] After converting the initial sample to HSV format, the information of the hue H channel, saturation S channel and brightness V channel is obtained, and random brightness disturbance is added to the V channel parameters to update the brightness value of the image and obtain the updated V channel parameters. Among them, the value range of brightness perturbation γ is .

[0057] Construct a reflectivity estimation network based on the U-Net structure, where U-Net is a classic encoder-decoder structure. The encoder part uses convolutional layers and maximum pooling layers to gradually reduce the spatial resolution and extract features; the decoder part uses convolutional layers and maximum pooling layers to gradually reduce the spatial resolution and extract features. Furthermore, skip connections are added at each decoder stage to connect the feature map in the encoder with the feature map in the decoder to retain more detailed information.

[0058] Combination Figure 2 As shown, the input layer of the reflectivity estimation network includes two channels, corresponding to the V channel parameters (combined with Figure 2 shown in the example, represented by "V" in this embodiment) and the updated V channel parameters (combined with Figure 2 As shown, in this embodiment, “ Represented); The output layer of the reflectivity estimation network corresponds to two channels (combined Figure 2 As shown, in this embodiment, R and Indicated by ). Combined Figure 2As shown in the figure, the green rectangle on the left represents the initial sample, the blue rectangle on the left represents the V channel parameters extracted after the HSV conversion of the initial sample; the yellow rectangle on the left represents the updated V channel parameters obtained by adding brightness perturbation to the blue rectangle. The red rectangle on the right represents the output of the R channel, and the orange rectangle represents The purple rectangle represents the output of the R channel combined with The final output is obtained after the output of the channel.

[0059] Combination Figure 2 As shown in the figure, the reflectivity estimation network of the U-Net structure includes 4 upsampling blocks and 4 downsampling blocks. Each sampling block includes 2 convolutional layers with a kernel size of 3x3. Each convolutional layer is followed by batch normalization (BN) and ReLU activation function. These convolutional layers are responsible for extracting features from the input image; combined with the maximum pooling layer with a pool size of 2x2, the computational load is reduced and overfitting is controlled by reducing the spatial dimension of the feature map; the bilinear interpolation layer is used to upsample the feature map and increase its spatial dimension; plus the two convolutional layers of the bottleneck layer and one convolutional layer of the fully connected layer, there are a total of 19 convolutional layers.

[0060] In combination with the first aspect, step S034 includes: ; ; ; ; ; in, is the first loss function, is the V channel loss function, is the reflectivity loss function, is the preset enhanced constraint loss; is the intensity of the incident light; is the first reflectivity; is the second reflectivity; For the general and Average pooled reflectance; are the preset hyperparameters.

[0061] Among them, the intensity of incident light I = brightness value V ÷ reflectivity R of the object. In fact, the reflectivity of the same object is constant. Therefore, the brightness value V is positively correlated with the intensity of incident light I. The intensity of incident light I can be adjusted by adjusting the brightness value V (i.e., the V channel parameter), thereby improving the robustness of bird recognition when the image illumination is within a certain range.

[0062] Based on the first loss function obtained in step S034, the reflectivity estimation network based on the U-Net structure is optimized so that the reflectivity estimation network can accurately and effectively perform image light transformation processing and output the target V channel parameters with the specified illumination intensity. Then, the target HSV image format composed of the target V channel parameters, S channel parameters and H channel parameters is converted into a training recognition image with the specified illumination intensity in RGB format through the HSV color model. In this way, the initial sample is processed by the HSV color model and the reflectivity estimation network based on the U-Net structure, reducing the adverse effects of image illumination on bird identification, which is conducive to the smooth progress of subsequent identification operations.

[0063] In combination with the first aspect, the value range of E is 0.2-0.3, and in this embodiment, the value of E is 0.25.

[0064] In combination with the first aspect, step S040 includes: S041, annotate the training recognition images for birds to obtain an annotated training image set.

[0065] S042, dividing the training image set into a training set and a validation set.

[0066] S043, inputting the training image into a bird recognition network based on the ResNet-50 model, and generating a feature map using a convolutional neural network in the bird recognition network based on the ResNet-50 model.

[0067] S044, passes the feature map to the fully connected layer and then to the Softmax layer to output the predicted bird prediction results.

[0068] S045, use the validation set to verify the bird prediction results.

[0069] S046: If the verification fails, the bird recognition network based on the ResNet-50 model is optimized based on a preset loss function to obtain a trained bird recognition network based on the ResNet-50 model.

[0070] S047, if the verification is successful, the current ResNet-50-based bird recognition network is determined as the trained ResNet-50-based bird recognition network.

[0071] Combination Figure 3 As shown in Figure 2, the bird recognition network based on ResNet-50 includes an output layer, a convolutional neural network (CNN), a fully connected layer (FC), a Softmax layer and an output layer. Figure 3The multiple rectangles of different colors in the figure represent multiple images. Multiple images are input into CNN, which processes these images through multiple layers to extract features and generates feature maps. These feature maps represent the features learned from the input images. The feature maps are then passed to the fully connected layer, which helps map the learned features to the output categories, and the output is normalized to probability through the Softmax layer, indicating the possibility of each bird. The final output is the bird prediction result based on the input image, such as North American cardinal, house sparrow, blue jay, mourning dove, etc.

[0072] Subsequently, the output bird prediction results are verified based on the labeled verification set. If the verification fails, step S046 is executed to optimize the model based on the preset loss function until a bird recognition network based on the ResNet-50 model that meets the requirements is obtained; if the verification is successful, it means that the bird recognition network based on the ResNet-50 model at this time can meet the application recognition accuracy requirements, and no additional training is required, and a trained bird recognition network based on the ResNet-50 model has been obtained.

[0073] Assume there are N samples, each of which belongs to one of C categories. samples, its pseudo label is the real label , the output of the fully connected layer is ,in Indicates The output scores of samples in C categories are converted into probability distribution by Softmax operation:

[0074] in, It represents the predicted probability that the i-th sample belongs to the C-th category.

[0075] Combined with the first aspect, the preset loss function is: ; Among them, L is the preset loss function, N is the number of samples, The i-th sample belongs to The predicted probability of the class.

[0076] In a second aspect, the present invention provides a bird identification device, Figure 4 As shown, the device includes: an acquisition module 10, an identification module 20, and a determination module 30.

[0077] The acquisition module 10 is used to acquire the image to be recognized.

[0078] The recognition module 20 is used to input the image to be recognized into a pre-trained bird recognition model and output a bird recognition result.

[0079] The determination module 30 is used to use the bird recognition result as the target bird recognition result corresponding to the image to be recognized; Among them, the bird recognition model includes an illumination transformation network based on a U-Net structure and a bird recognition network based on a ResNet-50 model, which are connected in sequence; the illumination transformation network based on the U-Net structure is used to convert the image to be recognized into a specified illumination recognition image with specified illumination conditions.

[0080] In a third aspect, an embodiment of the present invention provides an electronic device, Figure 5 As shown, the electronic device includes a memory 131 and a processor 130. The memory 131 is used to store computer programs, and the processor 130 runs the computer programs to enable the electronic device to perform the above method.

[0081] Furthermore, combined with Figure 5 The electronic device shown further includes a bus 132 and a communication interface 133 , and the processor 130 , the communication interface 133 and the memory 131 are connected via the bus 132 .

[0082] The memory 131 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk storage. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 133 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. may be used. The bus 132 may be an ISA bus, a PCI bus, or an EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0083] The processor 130 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 130. The above processor 130 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiment of the present invention can be directly embodied as a hardware decoding processor for execution, or a combination of hardware and software modules in the decoding processor for execution. The software module may be located in a storage medium mature in the art, such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 131, and the processor 130 reads the information in the memory 131 and completes the steps of the method of the above embodiment in combination with its hardware.

[0084] In a fourth aspect, an embodiment of the present invention provides a readable storage medium, in which computer program instructions are stored. When the computer program instructions are read and executed by a processor, the above method is executed.

[0085] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system and device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0086] In addition, in the description of the embodiments of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, and it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0087] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0088] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0089] Finally, it should be noted that the above embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above embodiments, those skilled in the art should understand that any person skilled in the art can still modify the technical solutions recorded in the above embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for identifying birds, characterized in that: The method comprises: Obtain an image to be recognized; Inputting the image to be identified into a pre-trained bird identification model, and outputting a bird identification result; Using the bird recognition result as the target bird recognition result corresponding to the image to be recognized; Among them, the bird recognition model includes an illumination transformation network based on a U-Net structure and a bird recognition network based on a ResNet-50 model connected in sequence; the illumination transformation network based on the U-Net structure is used to convert the image to be recognized into a specified illumination recognition image with specified illumination conditions.

2. The method according to claim 1, characterized in that Before the step of obtaining the image to be recognized, it includes: Constructing an initial recognition model, wherein the initial recognition model includes an illumination transformation network based on a U-Net structure and a bird recognition network based on a ResNet-50 model connected in sequence; Obtain an initial sample set; Using the initial sample set to train and optimize the illumination transformation network based on the U-Net structure, and output a training recognition image with a specified illumination intensity; Using the training recognition images to train the bird recognition network based on the ResNet-50 model, and outputting bird prediction results; The trained bird recognition model is obtained by combining the trained illumination transformation network based on the U-Net structure and the trained bird recognition network based on the ResNet-50 model.

3. The method according to claim 2, characterized in that The illumination transformation network based on the U-Net structure includes an HSV color model and a reflectivity estimation network based on the U-Net structure; The step of training and optimizing the illumination transformation network based on the U-Net structure by using the initial sample set and outputting a training recognition image with a specified illumination intensity comprises: Inputting the initial sample set into the HSV color model for format conversion, so as to convert each initial sample in the initial sample set into an HSV image; Add brightness disturbance to the V channel parameter in the HSV image to obtain an updated V channel parameter; Inputting the V channel parameter and the updated V channel parameter into the reflectivity estimation network at the same time, and outputting a first reflectivity corresponding to the V channel parameter and a second reflectivity corresponding to the updated V channel parameter; Calculate a first loss function by combining the V channel parameter, the first reflectivity, the second reflectivity, and a preset hyperparameter; Optimizing the reflectivity estimation network based on the U-Net structure based on the first loss function, and outputting target V channel parameters; Combining the target V channel parameters, the H channel parameters and the S channel parameters in the HSV image, a target HSV image is obtained; The target HSV image is input into the HSV color model for format inverse conversion, and the training recognition image with the specified light intensity is output.

4. The method according to claim 3, characterized in that The step of calculating a first loss function by combining the V channel parameter, the first reflectivity, the second reflectivity and a preset hyperparameter includes: ； ； ； ； in, is the first loss function, is the V channel loss function, is the reflectivity loss function, is the preset enhanced constraint loss; is the intensity of the incident light; is the first reflectivity; is the second reflectivity; For the general and Average reflectance after pooling; are the preset hyperparameters.

5. The method according to claim 2, characterized in that: The step of training the ResNet-50-based bird recognition network using the training recognition image and outputting the bird prediction result comprises: Annotating the training recognition images for birds to obtain an annotated training image set; Dividing the training image set into a training set and a validation set; Inputting the training set into the bird recognition network based on the ResNet-50 model, the convolutional neural network in the bird recognition network based on the ResNet-50 model generates a feature map; Pass the feature map to the fully connected layer and then to the Softmax layer to output the bird prediction result; Using the verification set to verify the bird prediction result; If the verification fails, the bird recognition network based on the ResNet-50 model is optimized based on a preset loss function to obtain a trained bird recognition network based on the ResNet-50 model.

6. The method according to claim 5, characterized in that After the step of verifying the bird prediction result using the verification set, the method further includes: If the verification is successful, the current ResNet-50-based bird recognition network is determined as the trained bird recognition network.

7. The method according to claim 5, characterized in that The preset loss function is: ； Among them, L is the preset loss function, N is the number of samples, The i-th sample belongs to The predicted probability of the class.

8. A bird identification device, characterized in that: The device comprises: An acquisition module, used for acquiring an image to be recognized; A recognition module, used for inputting the image to be recognized into a pre-trained bird recognition model and outputting a bird recognition result; A determination module, configured to use the bird recognition result as a target bird recognition result corresponding to the image to be recognized; Among them, the bird recognition model includes an illumination transformation network based on a U-Net structure and a bird recognition network based on a ResNet-50 model connected in sequence; the illumination transformation network based on the U-Net structure is used to convert the image to be recognized into a specified illumination recognition image with specified illumination conditions.

9. An electronic device, characterized in that: The electronic device comprises a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the method according to any one of claims 1 to 7.

10. A storage medium, characterized in that: The storage medium stores computer program instructions, and when the computer program instructions are read and executed by a processor, the method according to any one of claims 1 to 7 is executed.

Images