Image recognition method for birds and types

Abstract

The invention discloses an image recognition method for birds and types, and particularly relates to the technical field of image recognition. The method comprises the following steps: performing multi-region segmentation processing on a target bird image, respectively extracting multi-level visual features of each segmented region, generating a hierarchical feature set, performing cross-region similarity matching analysis, and dynamically adjusting the weight of each segmented region according to a matching result; collecting multiple frames of behavior images of the same target bird under continuous time nodes, and carrying out space-time correlation analysis in combination with the environmental background features of the corresponding time nodes to generate a dynamic behavior feature sequence; and fusing the adjusted weight of each segmented region with the dynamic behavior feature sequence to construct a bird feature recognition matrix, and outputting a category judgment result of the target bird according to a preset spatial distribution matching degree between a bird feature database and the bird feature recognition matrix, thereby realizing efficient and automatic recognition of the bird category. The method is suitable for wild bird monitoring and ecological environment evaluation.

Description

Technical Field

[0001] The present invention belongs to the technical field of image recognition, and in particular relates to an image recognition method for birds and their species. Background Art

[0002] Current wild bird monitoring and ecological conservation efforts are placing increasingly high demands on bird species identification. Traditional methods, which rely on manual observation and comparison, suffer from low efficiency, significant subjective interference, and low recognition accuracy. In recent years, computer vision and big data analysis technologies have rapidly advanced, and automatic recognition techniques based on image processing have gradually become a research hotspot. However, existing technologies primarily rely on single image features or simple feature fusion methods, making it difficult to fully capture a bird's local color distribution, texture details, morphological structure, and dynamic behavioral characteristics. Furthermore, they lack adaptability to environmental factors such as lighting and background, which affects recognition accuracy. Summary of the Invention

[0003] The present invention aims to provide a method for identifying bird species and their images to solve the problems mentioned in the background art.

[0004] To achieve the purpose of the present invention, the present invention discloses a method for image recognition of birds and their species, comprising the following steps:

[0005] S1. Acquire the target bird image and perform multi-region segmentation processing, perform multi-level visual feature extraction on each segmented region, and generate a hierarchical feature set containing local morphological features and global contour features;

[0006] S2. Perform cross-region similarity matching analysis on the feature data of different segmented regions in the hierarchical feature set, and adjust the weight of each segmented region according to the matching results;

[0007] S3. Obtain multiple frames of behavioral images of the same target bird at consecutive time points, extract the posture change features of each frame image, and perform spatiotemporal correlation analysis with the environmental background features of the corresponding time points to generate a dynamic behavioral feature sequence;

[0008] S4, fusing the adjusted weights of each segmented region with the dynamic behavior feature sequence to construct a bird feature recognition matrix;

[0009] S5. Output the determination result of the target bird species and behavior pattern type according to the spatial distribution matching degree between the preset bird behavior pattern database and the bird feature recognition matrix.

[0010] Furthermore, S1 is specifically:

[0011] Divide the target bird image into head region, trunk region, wing region and tail region;

[0012] The color distribution histogram, texture gradient features and geometric contour features are extracted for each segmented area;

[0013] The color distribution histogram and texture gradient features of each segmented area are merged into local morphological features, and the geometric contour features of each segmented area are superimposed to generate global contour features;

[0014] The local morphological features and global contour features of each segmented region are integrated according to the region division order to obtain a hierarchical feature set of the target bird image.

[0015] Furthermore, S2 is specifically:

[0016] Calculate the cosine similarity between the local morphological features of the head region and the wing region, and the Euclidean distance between the geometric contour features of the trunk region and the tail region respectively;

[0017] According to the numerical range of cosine similarity and Euclidean distance, the correlation strength value between the head region and the wing region, and the morphological difference value between the trunk region and the tail region are determined;

[0018] Based on the association strength value and the morphological difference value, the weights of the head region, the trunk region, the wing region, and the tail region are adjusted.

[0019] Furthermore, S3 is specifically:

[0020] Collect multiple frames of behavioral images of the same target bird at multiple consecutive time points;

[0021] Extract wing flapping frequency, head rotation angle and tail swing trajectory as posture change features from multi-frame behavioral images;

[0022] Synchronously collect the ambient light intensity and vegetation coverage density corresponding to each frame of behavioral image as environmental background features;

[0023] The posture change features are aligned with the environmental background features in time series to generate a dynamic behavior feature sequence.

[0024] Furthermore, S4 is specifically:

[0025] Based on the adjusted weights of the head region, wing region, and tail region, the wing flapping frequency, head rotation angle, and tail swing trajectory in the dynamic behavior feature sequence are weighted respectively.

[0026] The weighted wing flapping frequency and head rotation angle are concatenated according to time nodes to generate a joint time series feature vector;

[0027] The weighted tail swing trajectory is spatially mapped with the vegetation cover density to generate the associated spatial feature vector;

[0028] The joint temporal feature vectors and associated spatial feature vectors are arranged according to time nodes to construct a bird feature recognition matrix.

[0029] Furthermore, S5 is specifically:

[0030] Extract standard flight posture templates and typical foraging behavior waveforms from a preset bird behavior pattern database;

[0031] The wing flapping frequency in the bird feature recognition matrix is ​​matched with the standard flight posture template by dynamic time warping to determine the similarity of the flight pattern;

[0032] The time series of head turning angles was cross-correlated with the typical foraging behavior waveform to generate a foraging behavior confidence value.

[0033] When the flight pattern similarity exceeds a first threshold and the foraging behavior confidence value exceeds a second threshold, a determination result of the target bird species and a type label including migration behavior, foraging behavior, or social behavior is output.

[0034] Compared with the existing technology, the significant progress of the present invention is: 1) by performing multi-region segmentation and multi-level visual feature extraction on the target bird image, the local morphological features and global contour features are fully acquired; cross-region similarity matching analysis is adopted and the weights of each segmented region are dynamically adjusted to ensure that the features of each region in the image are fully integrated; the posture change features and environmental background information extracted from multi-frame behavioral images at continuous time nodes are combined to perform spatiotemporal correlation analysis to construct an efficient and accurate bird feature recognition matrix, thereby realizing automatic determination of the target bird species; 2) it overcomes the shortcomings of traditional recognition that relies on a single feature, has low recognition accuracy, and is greatly affected by environmental interference, and greatly improves recognition efficiency and robustness, making it suitable for bird monitoring and ecological evaluation in complex wild environments; 3) by making full use of multi-dimensional information such as images, behaviors and environments, it provides a more comprehensive, more rigorous and more adaptable solution for automatic identification of bird species.

[0035] In order to more clearly illustrate the functional characteristics and structural parameters of the present invention, further description is given below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0037] Figure 1 This is a schematic diagram of a method for image recognition of birds and species according to the present invention. DETAILED DESCRIPTION

[0038] The technical solutions 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 only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0039] A method for identifying bird species and images, the specific implementation of which is as follows:

[0040] S1. Acquire the target bird image and perform multi-region segmentation processing, perform multi-level visual feature extraction on each segmented region, and generate a hierarchical feature set containing local morphological features and global contour features;

[0041] S2. Perform cross-region similarity matching analysis on the feature data of different segmented regions in the hierarchical feature set, and adjust the weight of each segmented region according to the matching results;

[0042] S3. Obtain multiple frames of behavioral images of the same target bird at consecutive time points, extract the posture change features of each frame image, and perform spatiotemporal correlation analysis with the environmental background features of the corresponding time points to generate a dynamic behavioral feature sequence;

[0043] S4, fusing the adjusted weights of each segmented region with the dynamic behavior feature sequence to construct a bird feature recognition matrix;

[0044] S5. Output the determination result of the target bird species and behavior pattern type according to the spatial distribution matching degree between the preset bird behavior pattern database and the bird feature recognition matrix.

[0045] In a preferred embodiment, a target bird image is acquired and subjected to multi-region segmentation processing. Multi-level visual feature extraction is performed on each segmented region to generate a hierarchical feature set including local morphological features and global contour features, specifically:

[0046] Divide the target bird image into head region, trunk region, wing region and tail region;

[0047] The color distribution histogram, texture gradient features and geometric contour features are extracted for each segmented area;

[0048] The color distribution histogram and texture gradient features of each segmented area are merged into local morphological features, and the geometric contour features of each segmented area are superimposed to generate global contour features;

[0049] The local morphological features and global contour features of each segmented region are integrated according to the region division order to obtain a hierarchical feature set of the target bird image.

[0050] In a preferred embodiment, cross-region similarity matching analysis is performed on the feature data of different segmented regions in the hierarchical feature set, and the weight of each segmented region is adjusted according to the matching results, specifically:

[0051] Calculate the cosine similarity between the local morphological features of the head region and the wing region, and the Euclidean distance between the geometric contour features of the trunk region and the tail region respectively;

[0052] According to the numerical range of cosine similarity and Euclidean distance, the correlation strength value between the head region and the wing region, and the morphological difference value between the trunk region and the tail region are determined;

[0053] Based on the association strength value and the morphological difference value, the weights of the head region, the trunk region, the wing region, and the tail region are adjusted.

[0054] In a preferred embodiment, multiple frames of behavioral images of the same target bird at consecutive time nodes are obtained, the posture change characteristics of each frame image are extracted, and a spatiotemporal correlation analysis is performed with the environmental background characteristics of the corresponding time node to generate a dynamic behavioral feature sequence, specifically:

[0055] Collect multiple frames of behavioral images of the same target bird at multiple consecutive time points;

[0056] Extract wing flapping frequency, head rotation angle and tail swing trajectory as posture change features from multi-frame behavioral images;

[0057] Synchronously collect the ambient light intensity and vegetation coverage density corresponding to each frame of behavioral image as environmental background features;

[0058] The posture change features are aligned with the environmental background features in time series to generate a dynamic behavior feature sequence.

[0059] In a preferred embodiment, the adjusted weights of each segmented region are fused with the dynamic behavior feature sequence to construct a bird feature recognition matrix, specifically:

[0060] Based on the adjusted weights of the head region, wing region, and tail region, the wing flapping frequency, head rotation angle, and tail swing trajectory in the dynamic behavior feature sequence are weighted respectively.

[0061] The weighted wing flapping frequency and head rotation angle are concatenated according to time nodes to generate a joint time series feature vector;

[0062] The weighted tail swing trajectory is spatially mapped with the vegetation cover density to generate the associated spatial feature vector;

[0063] The joint temporal feature vectors and associated spatial feature vectors are arranged according to time nodes to construct a bird feature recognition matrix.

[0064] In a preferred embodiment, the target bird species determination result is output based on the spatial distribution matching degree between the preset bird feature database and the bird feature recognition matrix, specifically:

[0065] Extract standard flight posture templates and typical foraging behavior waveforms from a preset bird behavior pattern database;

[0066] The wing flapping frequency in the bird feature recognition matrix is ​​matched with the standard flight posture template by dynamic time warping to determine the similarity of the flight pattern;

[0067] The time series of head turning angles was cross-correlated with the typical foraging behavior waveform to generate a foraging behavior confidence value.

[0068] When the flight pattern similarity exceeds a first threshold and the foraging behavior confidence value exceeds a second threshold, a determination result of the target bird species and a type label including migration behavior, foraging behavior, or social behavior is output.

[0069] Example

[0070] Figure 1 The present invention provides a method for identifying bird species and their images, comprising the following steps:

[0071] S1. Acquire the target bird image and perform multi-region segmentation processing, perform multi-level visual feature extraction on each segmented region, and generate a hierarchical feature set containing local morphological features and global contour features;

[0072] S2. Perform cross-region similarity matching analysis on the feature data of different segmented regions in the hierarchical feature set, and adjust the weight of each segmented region according to the matching results;

[0073] S3. Obtain multiple frames of behavioral images of the same target bird at consecutive time points, extract the posture change features of each frame image, and perform spatiotemporal correlation analysis with the environmental background features of the corresponding time points to generate a dynamic behavioral feature sequence;

[0074] S4, fusing the adjusted weights of each segmented region with the dynamic behavior feature sequence to construct a bird feature recognition matrix;

[0075] S5. Output the determination result of the target bird species and behavior pattern type according to the spatial distribution matching degree between the preset bird behavior pattern database and the bird feature recognition matrix.

[0076] Specifically, the target bird image is acquired and multi-region segmentation is performed. Multi-level visual feature extraction is performed on each segmented region to generate a hierarchical feature set containing local morphological features and global contour features, including:

[0077] Divide the target bird image into head region, trunk region, wing region and tail region;

[0078] Specifically, a captured bird image is preprocessed and divided into four basic regions based on the bird's distinct structural features: the head, torso, wings, and tail. For example, a flying swallow can be segmented based on the visual boundaries between its head and torso, and its wings and tail, with each region corresponding to a different part of the bird's anatomy.

[0079] For example, when the head of a bird in an image has a relatively concentrated color, a relatively small shape, and is located at the top of the image, the upper part of the image can be set as the head area; the middle part as the torso area; the areas extending on both sides as wings; and the lower part of the image as the tail area.

[0080] The color distribution histogram, texture gradient features and geometric contour features are extracted for each segmented area;

[0081] Specifically, feature extraction is performed for each segmented area, including:

[0082] Color distribution histogram: Statistics the distribution of pixels of each color channel (such as red, green, and blue) in the segmented area.

[0083] Texture gradient features: Use operators such as the Sobel operator or gradient operator to detect local texture changes and reflect changes in the details of the segmented area.

[0084] Geometric contour features: Use edge detection algorithms to obtain region edges and then describe the shape contours of the segmented regions.

[0085] For example, for the wing region, the extracted color histogram can reflect the dominant distribution of colors on the wings; the texture gradient can capture the subtle texture changes of the feathers; and the geometric contour describes the shape of the wings, such as the degree of curvature or the smoothness of the edges.

[0086] Assuming that the number of pixels of the i-th color component in the segmented area is n(i) and the total number of pixels is N, the histogram normalization value is Among them, H(i) represents the normalized value of the i-th color interval in the segmented area, reflecting the proportion of the color interval; n(i) represents the number of pixels falling in the i-th color interval in the segmented area, and counts the frequency of occurrence of color components in the segmented area; N represents the total number of pixels in the segmented interval; i represents the index of each color interval in the color histogram.

[0087] The color distribution histogram and texture gradient features of each segmented area are merged into local morphological features, and the geometric contour features of each segmented area are superimposed to generate global contour features;

[0088] Specifically, for each segmented region, the color distribution histogram and texture gradient features are combined to form a local morphological feature vector. This vector reflects both the color and texture information of the segmented region. Simultaneously, the geometric contour features extracted from each segmented region are accumulated or superimposed to form a feature that reflects the global contour of the entire image.

[0089] For example, for the head area, the color histogram and texture gradient in the head area are synthesized into a vector describing the head details according to fixed rules (such as vector splicing); the edge contour features of the head, torso, wings, and tail are added through mathematical operations (such as vector summation) to obtain a global description vector reflecting the overall shape of the bird.

[0090] The local morphological features and global contour features of each segmented region are integrated according to the region division order to obtain the hierarchical feature set of the target bird image;

[0091] Specifically, according to a predetermined region division order (such as head, trunk, wings, and tail), the local morphological features of each segmented region are integrated with the global contour features in a fixed format to form a set of hierarchical feature sets to comprehensively describe the target bird image.

[0092] For example, if the local feature of the head is vector F t , the local feature of the trunk is vector F q , the local feature of the wing is vector F c , the local feature of the tail is vector F w , the global contour feature is recorded as F qj , then the hierarchical feature set can be expressed as: F fj ={F t ,F q ,F c ,F w ,F qj The above integration method can ensure that the characteristics of each region have a clear correspondence in the matching analysis.

[0093] Specifically, a cross-region similarity matching analysis is performed on the feature data of different segmented regions in the hierarchical feature set, and the weight of each segmented region is adjusted according to the matching results, including:

[0094] Calculate the cosine similarity between the local morphological features of the head region and the wing region, and the Euclidean distance between the geometric contour features of the trunk region and the tail region respectively;

[0095] Specifically, the feature data of each segmented region in the hierarchical feature set are compared.

[0096] For the head region and wing region, extract their respective local morphological feature vectors (such as F t and F c ), cosine similarity is used to measure the similarity, specifically:

[0097] Among them, CSS represents the cosine similarity between the local morphological features of the head region and the wing region; F t (j) represents the jth component of the local morphological feature vector of the head region, reflecting the value of the head region in the feature dimension; F c (j) represents the jth component in the local morphological feature vector of the wing region, reflecting the value of the wing region on the same feature dimension; m represents the dimension of the feature vector, that is, the total number of components contained in the local morphological feature vector.

[0098] For the trunk and tail regions, extract their respective geometric contour feature vectors (e.g. E q and E c ), the Euclidean distance is used to measure the difference in shape, specifically:

[0099] Among them, ECD represents the Euclidean distance between the geometric contour features of the trunk region and the tail region; E q (j) represents the jth component in the geometric contour feature vector of the trunk region, which is used to describe the contour information of the trunk region in the feature dimension; E c (j). The jth component of the geometric contour feature vector of the tail region is used to describe the contour information of the tail region in the corresponding feature dimension; M represents the dimension of the geometric contour feature vector, that is, the total number of components.

[0100] According to the numerical range of cosine similarity and Euclidean distance, the correlation strength value between the head region and the wing region, and the morphological difference value between the trunk region and the tail region are determined;

[0101] Specifically, based on the calculation result of cosine similarity, the numerical value is mapped to the association strength value between the head and the wings; similarly, based on the calculation result of Euclidean distance, the numerical value is mapped to the morphological difference value between the trunk and the tail.

[0102] When the cosine similarity is close to 1, it means that the head and wings have high similarity and the correlation strength value is large;

[0103] When the Euclidean distance is small, it means that the trunk and tail shapes are similar and the morphological difference value is low.

[0104] For example, assuming that the cosine similarity between the head region and the wing region is 0.92, the association strength value can be set to 0.92; if the Euclidean distance between the trunk region and the tail region is 0.15, the morphological difference value can be set to 0.15.

[0105] Based on the association strength value and the morphological difference value, the weights of the head region, the trunk region, the wing region, and the tail region are adjusted;

[0106] Specifically, the weights of each segmented region in the target bird image are dynamically adjusted using the association strength value and the morphological difference value.

[0107] If the correlation strength value between the head and wings is high, it means that the information of the head region and the wing region is relatively consistent during recognition, and a higher weight can be given;

[0108] If the morphological difference between the torso and tail is large, it means that the contours of the torso and tail regions have changed significantly, and the weight needs to be lowered or increased according to the specific situation.

[0109] Specifically, we acquire multiple frames of behavioral images of the same target bird at consecutive time points, extract the posture change features of each frame, and perform spatiotemporal correlation analysis with the environmental background features of the corresponding time points to generate a dynamic behavioral feature sequence, including:

[0110] Collect multiple frames of behavioral images of the same target bird at multiple consecutive time points;

[0111] Specifically, behavioral images of the target bird are collected at different consecutive time points. The behavioral images can be from video recordings or time-lapse photography. Each frame of the behavioral image records the dynamic state of the bird at different time points.

[0112] For example, during the bird's take-off process, one frame is captured every 0.1 seconds, and a total of 50 frames of images are obtained to record the bird's change process from stillness to flight.

[0113] Extract wing flapping frequency, head rotation angle and tail swing trajectory as posture change features from multi-frame behavioral images;

[0114] Specifically, the following features are extracted from each frame of the behavior image:

[0115] Wing flapping frequency: Analyze the changes in wings in consecutive frames and calculate the number of times the wings complete flapping per unit time.

[0116] Head rotation angle: Measures the rotation angle of the head in each frame relative to the previous frame or reference orientation.

[0117] Tail swing trajectory: Obtain the motion trajectory of the tail in the image, which can be obtained by tracking the position changes of the tail key points.

[0118] For example, if the wings are observed to complete one complete flapping within the time interval Δt, the wing flapping frequency is: Where f represents the wing flapping frequency, usually measured in beats per second, describing the number of times the wings complete a complete flapping cycle per unit time; Δt represents the time interval within a complete flapping cycle.

[0119] For the head rotation angle, θ can be defined z =|θ t -θ t-Δt |; where θ z represents the rotation angle of the head between two consecutive time nodes, reflecting the amplitude of the bird's head movement; θ t Indicates the head angle value measured at the current time t; θ t-Δt Indicates the head angle value measured at time t-Δt.

[0120] Synchronously collect the ambient light intensity and vegetation coverage density corresponding to each frame of behavioral image as environmental background features;

[0121] Specifically, while capturing each frame of behavioral image, the corresponding environmental background information of the frame is recorded, including:

[0122] Ambient light intensity: can be obtained using the overall brightness of the image or dedicated sensor data.

[0123] Vegetation cover density: Detects background vegetation by analyzing the ratio of green pixels in the background of the image or using specialized algorithms.

[0124] For example, in a frame of image, if the measured average brightness is 120 (unit is self-defined) and the proportion of green pixels in the background is 40%, these two data will be saved as the environmental background features of the frame.

[0125] Align the posture change features with the environmental background features in time series to generate a dynamic behavior feature sequence;

[0126] Specifically, the posture change features extracted from each frame of the image (wing flapping frequency, head rotation angle, tail swing trajectory) are aligned with the environmental background features of the corresponding time nodes (light intensity, vegetation cover density) in chronological order to form a continuous dynamic behavior feature sequence.

[0127] Specifically, the adjusted weights of each segmented region are fused with the dynamic behavior feature sequence to construct a bird feature recognition matrix, including:

[0128] Based on the adjusted weights of the head region, wing region, and tail region, the wing flapping frequency, head rotation angle, and tail swing trajectory in the dynamic behavior feature sequence are weighted respectively.

[0129] Specifically, the corresponding dynamic behavior features are weighted according to the weights of the head region, wing region, and tail region.

[0130] For wing flapping frequency, multiply by the weight adjusted for wing area;

[0131] For the head rotation angle, multiply it by the head area weight;

[0132] For tail swing trajectories, multiply by the tail area weight.

[0133] The weighted wing flapping frequency and head rotation angle are concatenated according to time nodes to generate a joint time series feature vector;

[0134] Specifically, the weighted wing flapping frequency and the head rotation angle are vector-concatenated at each time node, that is, the weighted wing flapping frequency and the head rotation angle are merged into a joint time series feature vector in a horizontal connection manner.

[0135] The weighted tail swing trajectory is spatially mapped with the vegetation cover density to generate the associated spatial feature vector;

[0136] Specifically, the weighted tail swing trajectory is mapped to the vegetation coverage density corresponding to the time node to form a vector reflecting the spatial distribution characteristics.

[0137] A linear combination method can be used for mapping, for example, the mapping formula is set as:

[0138] S(q)=α*T w,q (q)+(1-α)*V(q); where S(q) represents the associated spatial feature vector or value generated by mapping the tail swing characteristics with the vegetation cover density in the environment at time q; α represents the mapping weight coefficient, which ranges from 0 to 1 and is used to balance the contribution of the tail swing characteristics and vegetation cover density to the spatial characteristics; T w,q (q) represents the tail swing characteristics after weight adjustment at time q; V(q) represents the vegetation cover density measured at time q, which reflects the proportion or density of green vegetation in the background environment.

[0139] Arrange the joint temporal feature vectors and associated spatial feature vectors by time nodes to construct a bird feature recognition matrix;

[0140] Specifically, the joint temporal feature vector and the associated spatial feature vector are arranged in chronological order to form a two-dimensional bird feature recognition matrix, in which each row corresponds to the comprehensive feature at the same time node.

[0141] Specifically, based on the spatial distribution matching degree between the preset bird behavior pattern database and the bird feature recognition matrix, the determination results of the target bird species and behavior pattern type are output, including:

[0142] Extract standard flight posture templates and typical foraging behavior waveforms from a preset bird behavior pattern database;

[0143] Specifically, the Bird Behavior Pattern Database stores standard characteristic data for various bird species in specific behavioral scenarios. Standard flight posture templates are idealized or averaged descriptions of key bird movements during flight (such as wing amplitude, frequency, and angle changes). Typical foraging behavior waveforms, on the other hand, are temporal fluctuations in a bird's head rotation, movement rhythm, and other movements during foraging.

[0144] For example, in the database, for a certain migratory bird, the standard flight posture template may include data such as wing flapping frequency of 1.0 to 1.2 times per second and flapping amplitude of 30° to 40°; the typical foraging behavior waveform may be manifested as the head swinging periodically to search for food, and its waveform curve shows obvious rising and falling cycles.

[0145] The wing flapping frequency in the bird feature recognition matrix is ​​matched with the standard flight posture template by dynamic time warping to determine the similarity of the flight pattern;

[0146] Specifically, the dynamic time warping (DTU) method is first used to calculate the distance between the actual wing flap frequency sequence and the standard flight attitude template sequence. Specifically, for each pair of elements in the wing flap frequency sequence and the standard flight attitude template sequence that are matched via the alignment path, the square of the difference between them is calculated. These squared values ​​are then summed across all matching pairs, and the square root of the sum is taken. Among all possible matching paths, the path that minimizes the square root is selected. The resulting minimum square root value is the dynamic time warping distance (DTU) between the two sequences. Next, the flight pattern similarity is calculated using a natural exponential function. This method uses the negative DTU distance as an exponent, raising the negative DTU distance to the natural logarithm base e. When the DTU distance is small, the exponent approaches 1, indicating high flight pattern similarity. Conversely, when the distance is large, the exponent approaches 0, indicating low flight pattern similarity.

[0147] The time series of head turning angles was cross-correlated with the typical foraging behavior waveform to generate a foraging behavior confidence value.

[0148] Specifically, for the time series of head rotation angles extracted from the behavioral images of the target birds and the typical foraging behavior waveforms stored in the database, first, under the condition of zero delay (i.e., without introducing time offset), the corresponding values ​​of the two sequences at the same time point are multiplied in pairs, and all the products are summed to obtain a cross-correlation value. Subsequently, the squares of the values ​​at each time point in the two sequences are calculated respectively, and all the square values ​​are summed, and then the two sums are multiplied and the square roots are taken. Finally, the zero-delay cross-correlation value obtained above is divided by the value of this square root, and the result is the normalized cross-correlation value, also called the foraging behavior confidence value. This value ranges from 0 to 1. The closer the value is to 1, the more similar the two sequences are in shape and trend, that is, the higher the degree of match between the target behavior and the typical foraging behavior.

[0149] When the flight pattern similarity exceeds a first threshold and the foraging behavior confidence value exceeds a second threshold, outputting a determination result of the target bird species and a type label including migration behavior, foraging behavior, or social behavior;

[0150] Specifically, two thresholds are pre-set: the first threshold is used to determine whether the flight pattern similarity is high enough, and the second threshold is used to determine whether the foraging behavior confidence value meets expectations.

[0151] When the flight pattern similarity reaches or exceeds a first threshold (e.g., 0.85), it indicates that the flight state of the target bird matches the standard flight template well;

[0152] Meanwhile, if the foraging behavior confidence value reaches or exceeds a second threshold value (eg, 0.80), it indicates that the bird's behavioral characteristics conform to the foraging pattern.

[0153] By combining these two indicators, the final target bird species and behavioral pattern determination results can be output. The specific type labels may be "migratory behavior", "foraging behavior" or "social behavior", etc., and they are distinguished according to preset rules.

[0154] For example, if the detected flight pattern similarity is 0.90 (exceeding the first threshold of 0.85) and the foraging behavior confidence value is 0.88 (exceeding the second threshold of 0.80), the target bird will be determined to be in a foraging behavior state and the "foraging behavior" type label will be output. If the detection result does not meet both thresholds, the output may be a label of another behavior state or an uncertain state.

[0155] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0156] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0157] Those skilled in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0158] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and modules described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0159] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0160] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, and may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected to achieve the purpose of this embodiment according to actual needs.

[0161] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0162] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, 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 and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0163] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0164] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for identifying bird species and images, characterized in that: The steps include: S1. Acquire the target bird image and perform multi-region segmentation processing, perform multi-level visual feature extraction on each segmented region, and generate a hierarchical feature set containing local morphological features and global contour features; S2. Perform cross-region similarity matching analysis on the feature data of different segmented regions in the hierarchical feature set, and adjust the weight of each segmented region according to the matching results; S3. Obtain multiple frames of behavioral images of the same target bird at consecutive time points, extract the posture change features of each frame image, and perform spatiotemporal correlation analysis with the environmental background features of the corresponding time points to generate a dynamic behavioral feature sequence; S4, fusing the adjusted weights of each segmented region with the dynamic behavior feature sequence to construct a bird feature recognition matrix; S5. Output the determination result of the target bird species and behavior pattern type according to the spatial distribution matching degree between the preset bird behavior pattern database and the bird feature recognition matrix.

2. The method for identifying bird species and their images according to claim 1, wherein: S1 is specifically: Divide the target bird image into head region, trunk region, wing region and tail region; The color distribution histogram, texture gradient features and geometric contour features are extracted for each segmented area; The color distribution histogram and texture gradient features of each segmented area are merged into local morphological features, and the geometric contour features of each segmented area are superimposed to generate global contour features; The local morphological features and global contour features of each segmented region are integrated according to the region division order to obtain a hierarchical feature set of the target bird image.

3. The method for identifying bird species and their images according to claim 1, wherein: S2 is specifically: Calculate the cosine similarity between the local morphological features of the head region and the wing region, and the Euclidean distance between the geometric contour features of the trunk region and the tail region respectively; According to the numerical range of cosine similarity and Euclidean distance, the correlation strength value between the head region and the wing region, and the morphological difference value between the trunk region and the tail region are determined; Based on the association strength value and the morphological difference value, the weights of the head region, the trunk region, the wing region, and the tail region are adjusted.

4. The method for identifying bird species and images according to claim 1, wherein: S3 specifically: Collect multiple frames of behavioral images of the same target bird at multiple consecutive time points; Extract wing flapping frequency, head rotation angle and tail swing trajectory as posture change features from multi-frame behavioral images; Synchronously collect the ambient light intensity and vegetation coverage density corresponding to each frame of behavioral image as environmental background features; The posture change features are aligned with the environmental background features in time series to generate a dynamic behavior feature sequence.

5. The method for image recognition of birds and species according to claim 1, characterized in that: S4 is specifically: Based on the adjusted weights of the head region, wing region, and tail region, the wing flapping frequency, head rotation angle, and tail swing trajectory in the dynamic behavior feature sequence are weighted respectively. The weighted wing flapping frequency and head rotation angle are concatenated according to time nodes to generate a joint time series feature vector; The weighted tail swing trajectory is spatially mapped with the vegetation cover density to generate the associated spatial feature vector; The joint temporal feature vectors and associated spatial feature vectors are arranged according to time nodes to construct a bird feature recognition matrix.

6. The method for identifying bird species and images according to claim 1, wherein: S5 is specifically: Extract standard flight posture templates and typical foraging behavior waveforms from a preset bird behavior pattern database; The wing flapping frequency in the bird feature recognition matrix is ​​matched with the standard flight posture template by dynamic time warping to determine the similarity of the flight pattern; The time series of head turning angles was cross-correlated with the typical foraging behavior waveform to generate a foraging behavior confidence value. When the flight pattern similarity exceeds a first threshold and the foraging behavior confidence value exceeds a second threshold, a determination result of the target bird species and a type label including migration behavior, foraging behavior, or social behavior is output.

Images