Bird species identification method, device and equipment and readable storage medium

Abstract

The invention belongs to the technical field of bird species identification, and provides a bird species identification method, device and equipment and a readable storage medium, and the method comprises the steps: obtaining bird data; wherein the bird data comprises bird picture data and / or bird sound data; processing the acquired bird data; inputting the processed bird data into the constructed recognition model for classification to obtain a classification result; wherein the processing of the acquired bird data type comprises the following steps: if the acquired bird data is bird picture data, setting bird sound data as null values; if the obtained bird data is bird sound data, bird picture data are set to be null values; according to the method, independent bird picture data and bird sound data are supported to serve as input of the recognition model, the bird picture data and the bird sound data are allowed to be input at the same time, strict requirements for whether the input picture and sound correspond to the same bird or not are not made, the feature extraction range can be expanded, and therefore the recognition accuracy is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of bird species identification, and in particular relates to a bird species identification method, device, equipment and readable storage medium. Background Art

[0002] Birds are an indispensable ecological element in nature, and the maintenance of their species diversity is of vital importance to ecological balance. Therefore, it is necessary to monitor the dynamics of bird populations.

[0003] Traditional bird monitoring methods, such as radar tracking, photoelectric monitoring, and manual observation, can provide information on bird activities to a certain extent, but they are obviously insufficient in terms of efficiency and comprehensiveness. Although the radar tracking method can accurately record the activity paths of birds, it is costly and complex to implement, making it difficult to apply on a large scale. Photoelectric monitoring technology is limited by lighting conditions and cannot fully capture the sound characteristics of birds. Although manual observation can intuitively identify birds, it requires extremely high professionalism and time investment from the observer, and is easily interfered by environmental factors, such as day and night changes and weather conditions.

[0004] With the rapid development of video surveillance technology, voiceprint recognition technology and neural network models, new solutions have been provided for bird monitoring. The type of bird can be obtained by identifying the input image or sound through the neural network model. However, most existing neural network models can only accept a single type of data input (i.e. only images or only sounds), which not only limits the feature extraction range of the model, but also fails to meet the needs of certain specific application scenarios. Summary of the invention

[0005] In view of the problems in the background technology, the present invention proposes a method for identifying bird species.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a method for identifying bird species, characterized by comprising:

[0008] Acquire bird data; wherein the bird data includes bird picture data and / or bird sound data;

[0009] processing the acquired bird data;

[0010] Inputting the processed bird data into the constructed recognition model for classification to obtain classification results;

[0011] The step of processing the acquired bird data type includes:

[0012] If the acquired bird data is bird picture data, the bird sound data is set to a null value;

[0013] If the acquired bird data is bird sound data, the bird picture data is set to a null value.

[0014] Furthermore, the processed bird data is input into the constructed recognition model for classification to obtain the classification result, including:

[0015] Using sliding window technology, the bird sound data is divided into multiple overlapping time frames, and the Mel-spectrogram of the bird sound data of each frame is calculated;

[0016] The sliding window technology is used to segment the bird image data into multiple local images;

[0017] A convolutional neural network is used to extract local spectrum feature vectors in the mel-spectrogram and image feature vectors in the local image respectively;

[0018] Fusing the local spectrum feature vector and the image feature vector to obtain a fused feature vector;

[0019] The fused feature vector is processed by a recurrent neural network to obtain a sequence feature vector;

[0020] A multi-layer perceptron is used to classify the sequence feature vectors to obtain the classification results.

[0021] Furthermore, the fusing of the local spectrum feature vector and the image feature vector includes:

[0022] The local spectrum feature vector and the image feature vector are weighted and fused, and the expression is:

[0023] S=ω 1 *s 1 +ω 2 *s 2

[0024] In the formula, S is the fusion result, ω 1 is the weight coefficient of the local spectrum feature vector, s 1 is the local spectrum feature vector, ω 2 is the image feature vector weight coefficient, s 2 is the image feature vector;

[0025] Among them, the weight coefficient of the image feature vector is greater than the weight coefficient of the local spectrum feature vector.

[0026] Furthermore, constructing the recognition model includes:

[0027] Get bird image dataset and bird sound dataset;

[0028] Manually annotating the bird image data and the bird sound data to obtain manually annotated information;

[0029] Dividing the manually annotated bird image data and the bird sound data into a training set and a test set respectively;

[0030] According to the manually labeled information, the corresponding training set of the bird picture data and the bird sound data is input into the recognition model for training to obtain a trained recognition model;

[0031] The corresponding test sets of the bird picture data and the bird sound data are input into the trained recognition model for testing to obtain a constructed recognition model.

[0032] Furthermore, before manually labeling the bird picture data and the bird sound data, the method further includes: preprocessing the bird picture data and the bird sound data.

[0033] Further, the bird image data is preprocessed, including:

[0034] Filter bird posture image data;

[0035] Enhance image data by flipping, rotating, translating, and cropping;

[0036] Enhance image data by adding noise, color transformation, and pixel erasing;

[0037] Preprocessing the bird sound data includes:

[0038] normalizing the bird sound data;

[0039] The bird sound data is processed by pre-emphasis, high-pass filtering, frame windowing, and endpoint detection.

[0040] Furthermore, if the acquired bird picture data or bird sound data is of a newly added bird, the constructed recognition model is updated according to the bird picture data or bird sound data of the newly added bird to obtain an updated recognition model.

[0041] In a second aspect, the present invention provides a bird species identification device, characterized by comprising:

[0042] A data acquisition module, used to acquire bird data; wherein the bird data includes bird picture data and / or bird sound data;

[0043] A data processing module, used for processing the acquired bird data type;

[0044] A classification module, used for inputting the processed bird data into the constructed recognition model for classification to obtain a classification result;

[0045] The step of processing the acquired bird data type includes:

[0046] If the acquired bird data is bird picture data, the bird sound data is set to a null value;

[0047] If the acquired bird data is bird sound data, the bird picture data is set to a null value.

[0048] In a third aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, which can implement the above-mentioned bird species identification method when the computer instructions are executed.

[0049] In a fourth aspect, the present invention proposes a device, comprising a memory and a processor, wherein the memory stores computer instructions that can be executed on the processor, and the processor executes the above-mentioned bird species identification method when executing the computer instructions.

[0050] Beneficial effects of the present invention:

[0051] 1. The method of the present invention not only supports separate bird picture data and bird sound data as input of the recognition model, but also allows the two to be input at the same time, and does not make strict requirements on whether the input pictures and sounds correspond to the same bird. It can expand the scope of feature extraction, thereby improving the accuracy of recognition and solving the problem that traditional models can only accept a single type of data input.

[0052] 2. The method of the present invention identifies birds by constructing a recognition model. The recognition model combines the advantages of convolutional neural network (CNN), recurrent neural network (RNN) and multi-layer perceptron (MLP) to construct a multimodal recognition system. Among them, CNN is good at processing image data and can effectively extract spatial structural information in images, such as texture, edge, shape, etc.; RNN is good at processing sequence data and can capture the temporal dependency in sound or image, such as the pitch change of bird song and the dynamic change in image; MLP can make accurate classification judgment based on the rich features extracted by CNN and RNN, which not only improves the accuracy of recognition, but also enhances the generalization ability of the model.

[0053] Other features and advantages of the present invention will be described in the following description, and partly become obvious from the description, or be understood by implementing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0055] Figure 1 A flow chart showing the bird species identification method of the present invention;

[0056] Figure 2 A flow chart of a method for processing bird picture data and bird sound data of the present invention is shown;

[0057] Figure 3 The framework diagram of the bird species identification device of the present invention is shown;

[0058] Figure 4 A schematic diagram of the device structure of the present invention is shown. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. 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 ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] Refer to the instruction manual Figure 1 and attached Figure 2 As shown, attached Figure 1 A flowchart showing the bird species identification method is shown in the Appendix. Figure 2 A method flow chart of processing bird picture data and bird sound data is shown, a method for identifying bird species, specifically comprising the following steps:

[0061] S1. Acquire bird data; wherein the bird data includes bird picture data and / or bird sound data;

[0062] To obtain bird image data, high-definition monitoring equipment is installed in the bird distribution and gathering areas, and real-time monitoring and operation are carried out, and the captured bird photos and videos are transmitted back to the backend system for analysis. Among them, the monitoring equipment can be of the following types:

[0063] 1. Telephoto dome camera: Deploy telephoto dome cameras in areas where birds are active at close range, such as wetland edges and around lakes. These devices have high-magnification optical zoom lenses that can clearly capture the subtle movements and behavioral characteristics of birds, such as foraging, nesting, and reproduction.

[0064] 2. Medium-sized PTZ: Medium-sized PTZ monitoring equipment is installed in areas that require wider field of view monitoring, such as forest edges and grasslands. These PTZs support flexible horizontal and vertical rotation, and combined with high-definition cameras, they can achieve dynamic tracking and recording of bird activities in a larger range.

[0065] Bird sound data is obtained by installing voiceprint monitoring terminals in bird-gathering areas such as nature reserves, collecting bird sounds in the surrounding area, and transmitting the collected sounds back to the backend system for processing, analysis and identification.

[0066] In order to ensure that the acquired bird data and sound data are of the same bird as much as possible, the monitoring equipment and voiceprint monitoring terminal are installed at the same or similar geographical location.

[0067] S2, processing the acquired bird data;

[0068] Specifically, if the acquired bird data is bird picture data, the bird sound data is set to a null value;

[0069] If the acquired bird data is bird sound data, the bird picture data is set to a null value;

[0070] If the acquired bird data is bird sound data and bird picture data, no operation is performed.

[0071] S3, inputting the processed bird data into the constructed recognition model for classification to obtain the classification result;

[0072] Specifically, the constructed recognition model includes: a parallel convolutional neural network module, a recurrent neural network and a multi-layer perceptron connected in series in sequence; wherein the parallel convolutional neural network module includes two parallel convolutional neural networks, and the two convolutional neural networks are used to process bird picture data and bird sound data respectively.

[0073] In some embodiments, the acquired bird data is input into the constructed recognition model for classification, which specifically includes the following steps:

[0074] S31, using a sliding window technique to segment the bird sound data into a plurality of overlapping time frames, and calculating a mel-spectrogram of the bird sound data of each frame;

[0075] Using sliding window technology, bird sound data is divided into multiple overlapping time frames, each of which can independently reflect the characteristics of the audio signal in a short period of time. Calculating the Mel-spectrogram can convert the audio signal from the time domain to the frequency domain, and can capture the key frequency components in the audio, especially the frequency modulation and pitch changes in the bird's song, providing rich information for subsequent feature extraction and classification.

[0076] S32, using sliding window technology to divide the bird image data into multiple local images;

[0077] By dividing the entire image into multiple local images, we can focus on different areas and details in the image, such as the texture of a bird's feathers, the shape of its beak, etc., which helps capture subtle features of the image and improve classification accuracy.

[0078] In step S31 and step S32, a sliding window technique is used to allow overlap between the time frame and the local image, thereby maintaining the integrity and continuity of the image and reducing information loss due to segmentation.

[0079] S33, using a convolutional neural network to extract local spectrum feature vectors in the mel-spectrogram and image feature vectors in the local image;

[0080] Convolutional neural networks (CNNs) can capture spatial structural information in sound data (frequency distribution and dynamic changes of audio signals, etc.) and spatial structural information in images (texture, edges, shapes, etc. of images), which helps to identify different bird species.

[0081] S34, fusing the local spectrum feature vector and the image feature vector to obtain a fused feature vector;

[0082] By fusing local spectral features and image features, we can make full use of the complementary information between sound and image. Local spectral features reflect the characteristics of bird sounds, while image features reflect the appearance of birds. Combining the two can more comprehensively and accurately describe the comprehensive characteristics of birds.

[0083] Specifically, the fusion of the local spectrum feature vector and the image feature vector includes:

[0084] The local spectrum feature vector and the image feature vector are weighted and fused, and the expression is:

[0085] S=ω 1 *s 1 +ω 2 *s 2

[0086] In the formula, S is the fusion result, ω 1 is the weight coefficient of the local spectrum feature vector, s 1 is the local spectrum feature vector, ω 2 is the image feature vector weight coefficient, s 2 is the image feature vector, ω 1 and ω 2 are greater than or equal to 0, and ω 1 +ω 2=1. The image feature vector weight coefficient is greater than the local spectrum feature vector weight coefficient. As a preferred embodiment, the local spectrum feature vector weight coefficient ω 1 The value is 0.3, and the image feature vector weight coefficient ω 2 The value of is 0.7.

[0087] It is worth noting that in this recognition model, the image feature vector is given a higher weight coefficient (such as ω 2 =0.7), which reflects that when the bird picture data and bird sound data are combined for judgment, the image feature has a higher priority than the sound feature. This ensures that the possible mismatch between the bird picture data and the bird sound data can be handled, that is, due to the reasons for the monitoring equipment and the voiceprint monitoring terminal, the bird picture data and the bird sound data are not from the same bird. By increasing the weight of the image feature, the recognition model will be more inclined to use the image feature for recognition.

[0088] By making the image feature vector weight coefficient greater than the local spectrum feature vector weight coefficient, it can be ensured that even when the bird picture data and the bird sound data do not correspond, a relatively accurate classification result can be given based on the image features.

[0089] S35, using a recurrent neural network to process the fused feature vector to obtain a sequence feature vector;

[0090] Recurrent neural networks (RNNs) have powerful sequence modeling capabilities. They can mine the temporal dependencies between adjacent frames and the corresponding structural relationships between pixels in an image. They can capture the temporal changes in bird calls and the dynamic changes in images, such as frequency modulation, pitch changes, image color and geometric changes, which are key clues for identifying different bird calls and bird images.

[0091] S36. Use a multi-layer perceptron (MLP) to classify the sequence feature vector to obtain a classification result.

[0092] Based on the rich features learned jointly from CNN and RNN, MLP accurately classifies the input bird song signals and bird images.

[0093] In step S31 and step S32, if only bird picture data or bird sound data is included, it is set to a null value in step S2, so the corresponding type of data will be regarded as a null value (zero vector) input. Exemplarily, if only bird sound data is input into the constructed recognition model, then in the step of processing bird picture data, the relevant input will be regarded as a null value. It is understandable that in this case, when the convolutional neural network receives a null value as input, its output will naturally be a null value. Therefore, in step S34, the fusion result will be obtained only based on the feature vector of non-null value.

[0094] By classifying bird image data and bird sound data through the recognition model, not only the time dimension information of the spectrogram is retained, but also the respective advantages of CNN and RNN in feature extraction and time series modeling are fully utilized, which brings significant performance improvements to bird song recognition and image recognition.

[0095] In some embodiments, building a recognition model specifically includes the following steps:

[0096] S311, obtaining a bird picture dataset and a bird sound dataset;

[0097] The bird image dataset is collected by using a crawler program to collect bird images on the website. The bird sound data is collected by installing sound collection devices in nature reserves and downloading them from open source websites.

[0098] S312, preprocessing the bird picture data and the bird sound data;

[0099] Preprocess the bird image data, including:

[0100] S3121. Filter bird posture image data; specifically, filter images with clear sample quality, clear features, appropriate bright spots, and low noise points, and select images of birds in various postures as collection objects, including bird flying, feeding, and running postures.

[0101] S3122, enhancing the image data by flipping, rotating, translating, and cropping;

[0102] S3123, enhancing the image data by adding noise, color conversion, and pixel erasing.

[0103] Preprocessing of bird sound data includes the following steps:

[0104] S3124. Normalize bird sound data. Specifically, bird song sound data in a natural environment is usually stored in a dual-channel format, and bird song sounds are complex and diverse. Therefore, it is necessary to merge and store bird song data signals from two different sources (one is collected directly from a nature reserve, and the other is downloaded from a website) in a WAV format, and in the process, normalize the signal amplitude.

[0105] Specifically, the normalization method used is maximum normalization, and the formula is:

[0106]

[0107] Where x(i) is the element with index i in the sequence x, and N is the total number of elements in the sequence.

[0108] S3125, pre-emphasis, high-pass filtering, frame splitting and windowing, endpoint detection of bird sound data;

[0109] S313, manually annotating bird image data and bird sound data to obtain manually annotated information;

[0110] By manually classifying and labeling the collected bird images and sound data, it is possible to clearly indicate which type of bird each image and each piece of sound data belongs to. Manual labelers use labeling tools (X-Anylabeling, Labeling, Labelme) to label the bird images in the training set, marking the specific type of bird in the image and the specific location of the bird in the image. This is achieved by drawing a bounding box that should tightly surround the bird in the image but not contain too much background.

[0111] S314, dividing the manually annotated bird image data and bird sound data into a training set and a test set respectively;

[0112] The training set is used to train the model so that it can learn how to distinguish different types of birds; the test set is used to evaluate the performance of the constructed model.

[0113] S315, according to the manual annotation information, inputting the corresponding training set of bird picture data and bird sound data into the recognition model for training to obtain a trained recognition model;

[0114] The recognition model learns how to distinguish different types of birds by studying the images and sound data in the training set and their corresponding manual annotation information.

[0115] S316, inputting the corresponding test set of bird picture data and bird sound data into the trained recognition model for testing to obtain a constructed recognition model.

[0116] Use the test set to test the constructed recognition model to evaluate its performance on unseen data.

[0117] Although the corresponding bird image data and sound data are used in the training process, this multimodal model does not affect its flexibility in actual use. Specifically, for inputs with only bird image data or only bird sound data, the recognition model can still use its feature extraction and recognition capabilities learned during the training process to accurately identify birds on the input data. This is because the model has learned the feature representations of image and sound data separately during training, so it can process these two data types independently.

[0118] In some embodiments, if the acquired bird picture data or bird sound data is of a newly added bird type, the constructed recognition model is updated according to the bird picture data or bird sound data of the newly added bird to obtain an updated recognition model.

[0119] In the process of actually applying the recognition model to identify birds, the recognition model may face bird species that were not included in the original training set. Since the recognition ability of the model is built based on the data set it has learned, if the input data contains birds that the recognition model has never seen, the model may not be able to accurately identify the species of the bird. Therefore, in order to cope with this situation and ensure that the model can continue to effectively identify birds, it is necessary to continuously update the recognition model, that is, to use the pictures or sound data of the newly added birds as the training set to train the recognition model. Specifically, when encountering birds that the model cannot recognize, that is, newly added birds, first collect the pictures and sound data of the newly added birds and manually annotate them. Then, add these newly collected data to the original training set to form an updated training set. Then, use the updated training set to retrain the recognition model to obtain an updated recognition model.

[0120] The updated recognition model not only retains the original recognition capabilities for known birds, but also adds the ability to recognize new birds.

[0121] Refer to the instruction manual Figure 3 As shown, it shows a framework diagram of a bird species identification device. Based on the same inventive concept, the present invention also proposes a bird species identification device, including:

[0122] The data acquisition module 110 is used to acquire bird data; wherein the bird data includes bird picture data and / or bird sound data;

[0123] A data processing module 120, used for processing the acquired bird data type;

[0124] The classification module 130 is used to input the processed bird data into the constructed recognition model for classification to obtain the classification result;

[0125] The step of processing the acquired bird data type includes:

[0126] If the acquired bird data is bird picture data, the bird sound data is set to a null value;

[0127] If the acquired bird data is bird sound data, the bird picture data is set to a null value.

[0128] Based on the same inventive concept, the present invention also proposes a computer-readable storage medium on which computer instructions are stored. When the computer instructions are executed, the above-mentioned bird species identification method can be implemented.

[0129] Refer to the instruction manual Figure 4 As shown, it shows a schematic diagram of the device structure. Based on the same inventive concept, the present invention also proposes a device, including a memory and a processor. The memory stores computer instructions that can be run on the processor. When the processor runs the computer instructions, the above-mentioned bird species identification method is executed.

[0130] Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory.

[0131] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0132] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions 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.

Claims

1. A method for identifying bird species, characterized in that: include: Acquire bird data; wherein the bird data includes bird picture data and / or bird sound data; processing the acquired bird data; Inputting the processed bird data into the constructed recognition model for classification to obtain classification results; The step of processing the acquired bird data type includes: If the acquired bird data is bird picture data, the bird sound data is set to a null value; If the acquired bird data is bird sound data, the bird picture data is set to a null value.

2. The bird species identification method according to claim 1, characterized in that: The step of inputting the processed bird data into the constructed recognition model for classification to obtain the classification result comprises: Using sliding window technology, the bird sound data is divided into multiple overlapping time frames, and the Mel-spectrogram of the bird sound data of each frame is calculated; The sliding window technology is used to segment the bird image data into multiple local images; A convolutional neural network is used to extract local spectrum feature vectors in the mel-spectrogram and image feature vectors in the local image respectively; Fusing the local spectrum feature vector and the image feature vector to obtain a fused feature vector; The fused feature vector is processed by a recurrent neural network to obtain a sequence feature vector; A multi-layer perceptron is used to classify the sequence feature vectors to obtain the classification results.

3. The bird species identification method according to claim 2, characterized in that: The fusing of the local spectrum feature vector and the image feature vector comprises: The local spectrum feature vector and the image feature vector are weighted and fused, and the expression is: S=ω1*s1+ω2*s2 Where S is the fusion result, ω1 is the weight coefficient of the local spectrum feature vector, s1 is the local spectrum feature vector, ω2 is the weight coefficient of the image feature vector, and s2 is the image feature vector; Among them, the weight coefficient of the image feature vector is greater than the weight coefficient of the local spectrum feature vector.

4. The bird species identification method according to claim 1, characterized in that: Constructing the recognition model includes: Get bird image dataset and bird sound dataset; Manually annotating the bird image data and the bird sound data to obtain manually annotated information; Dividing the manually annotated bird image data and the bird sound data into a training set and a test set respectively; According to the manually labeled information, the corresponding training set of the bird picture data and the bird sound data is input into the recognition model for training to obtain a trained recognition model; The corresponding test sets of the bird picture data and the bird sound data are input into the trained recognition model for testing to obtain a constructed recognition model.

5. The bird species identification method according to claim 4, characterized in that: Before manually labeling the bird picture data and the bird sound data, the method further includes: preprocessing the bird picture data and the bird sound data.

6. The bird species identification method according to claim 5, characterized in that: Preprocessing the bird image data includes: Filter bird posture image data; Enhance image data by flipping, rotating, translating, and cropping; Enhance image data by adding noise, color transformation, and pixel erasing; Preprocessing the bird sound data includes: normalizing the bird sound data; The bird sound data is processed by pre-emphasis, high-pass filtering, frame windowing, and endpoint detection.

7. The bird species identification method according to claim 3, characterized in that: If the acquired bird picture data or bird sound data is of a newly added bird, the constructed recognition model is updated according to the bird picture data or bird sound data of the newly added bird to obtain an updated recognition model.

8. A bird species identification device, characterized in that: include: A data acquisition module, used to acquire bird data; wherein the bird data includes bird picture data and / or bird sound data; A data processing module, used for processing the acquired bird data type; The classification module is used to input the processed bird data into the constructed recognition model for classification to obtain classification results.

9. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed, the bird species identification method according to any one of claims 1 to 7 can be implemented.

10. A device comprising a memory and a processor, wherein the memory stores computer instructions that can be executed on the processor, characterized in that: When the processor runs the computer instructions, the method for identifying bird species according to any one of claims 1 to 7 is performed.

Images