A machine learning based protected area bird monitoring protection system

Abstract

The application relates to the technical field of bird monitoring and protection, and particularly relates to a protected area bird monitoring and protection system based on machine learning, which acquires a monitoring image through a collection module, obtains a first bird image and corresponding first bird data based on a preset machine learning algorithm through a default monitoring module, identifies the monitoring image based on pixel comparison with the first bird image through a rapid monitoring module, saves monitoring data through a cache module, and judges whether the monitoring data in the cache module meets a preset trigger condition through the cache monitoring module; if yes, the rapid monitoring module is called to perform monitoring; and if no, the default monitoring module is called to perform monitoring. The application realizes preliminary analysis of the monitoring data in a temporary storage mode through the cache module, so that the rapid monitoring module is called to perform rapid monitoring through pixel comparison according to the number of currently monitored birds, and the problem that a large number of birds cannot be identified at the same time in a short time in the prior art is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bird monitoring and protection, and particularly relates to a protected area bird monitoring and protection system based on machine learning. BACKGROUND

[0002] Birds, as key species in the ecosystem, their number and species changes can reflect the environmental quality and the health of the ecosystem, and are an important indicator of biodiversity monitoring. Through the monitoring of birds, not only can endangered birds and their habitats be effectively protected, but also can early warning and prevention and control of the spread of infectious diseases, reduce the impact of birds on power facilities and aviation safety, etc. benefits.

[0003] The bird monitoring method based on machine learning is the current mainstream research direction, especially the image recognition based method, which has been widely used in bird monitoring due to its efficient real-time detection capability. For example, the existing bird monitoring technology based on image recognition includes bird recognition classification technology using convolutional neural network (CNN) in deep learning, bird recognition technology based on YOLO series algorithm, etc.

[0004] However, the existing bird monitoring method based on machine learning image recognition often has poor effect when facing the scene of a large number of birds appearing in a short time. For example, in the scene of birds wandering back and forth or gathering in wetlands, due to the large number of individual birds, large changes in posture, and fast speed, the existing algorithm often has difficulty in accurately identifying and distinguishing each individual in the picture, and is prone to problems such as target loss or false association, thereby affecting the accuracy of the monitoring result. Therefore, people need a monitoring method that can identify a large number of birds at the same time in a short time. SUMMARY

[0005] Therefore, the present application provides a protected area bird monitoring and protection system based on machine learning to solve the problem that it is difficult to identify a large number of birds at the same time in a short time in the prior art.

[0006] The present application provides a protected area bird monitoring and protection system based on machine learning, comprising a cache module, a collection module, a default monitoring module, a rapid monitoring module and a cache monitoring module, wherein:

[0007] The collection module is used to obtain a monitoring image;

[0008] The default monitoring module is used to identify the monitoring image based on a preset machine learning algorithm, obtain a first bird image and corresponding first bird data, and save the first bird image and the first bird data as monitoring data in the cache module;

[0009] The rapid monitoring module is used to identify the monitoring image based on pixel comparison with the first bird image, obtain the second bird image and the corresponding second bird data, and save the second bird image and the second bird data as monitoring data in the cache module;

[0010] The cached monitoring module is used to determine whether the monitoring data in the cached module meets the preset trigger conditions. If yes, the fast monitoring module is called to perform monitoring; otherwise, the default monitoring module is called to perform monitoring. The preset trigger conditions are used to determine the current number of birds to be monitored based on the monitoring data.

[0011] In a preferred embodiment, the monitoring image is identified by comparing it pixel-wise with a first bird image to obtain a second bird image and corresponding second bird data, including:

[0012] Multiple first bird images are normalized to obtain multiple sample images of the same size;

[0013] Based on the size distribution of multiple first bird images, the size of the monitoring images was adjusted multiple times to obtain target images of multiple size specifications;

[0014] A detection window of the same size as the sample image is set in the target image. The detection window is moved and the content in the detection window is compared pixel by pixel with the sample image to obtain the second bird image and the corresponding second bird data.

[0015] In a preferred embodiment, a detection window of the same size as the sample image is set in the target image. The detection window is moved, and the content within the detection window is compared pixel-by-pixel with the sample image to obtain a second bird image and corresponding second bird data, including:

[0016] Obtain a preset mapping model, whereby the preset mapping model is used to map different images to unique feature values;

[0017] The sample images are input into a preset mapping model to obtain the sample label value corresponding to each sample image;

[0018] Set a detection window with the same size as the sample image in the target image, move the detection window and input the contents of the detection window into the preset mapping model each time it is moved to obtain the set of output feature values;

[0019] The sample label value is matched with the feature value in the output feature value set, and the position information of the detection window corresponding to the successfully matched feature value is obtained.

[0020] Based on the location information, a portion of the monitored image content is extracted and used as the initial bird image corresponding to the sample label value;

[0021] Based on the correspondence between the initial bird images, sample label values, sample images, the first bird image, and the first bird data, all initial bird images are summarized to obtain the second bird image and the corresponding second bird data.

[0022] In a preferred embodiment, multiple first bird images are normalized to obtain multiple sample images of the same size, including:

[0023] The first bird image is resized while preserving its features to obtain a standardized image.

[0024] The standardized image is transformed by pixel value transformation to obtain the sample image.

[0025] In a preferred embodiment, the first bird image is subjected to feature-preserving size adjustment to obtain a standardized image, including:

[0026] Adjust the size based on the target to obtain the first pixel and its neighboring pixels;

[0027] The pixel values ​​of the first pixel and its neighboring pixels are weighted and averaged to obtain the standardized pixels, and multiple standardized pixels constitute the standardized image.

[0028] In a preferred embodiment, pixel value transformation is performed on the standardized image to obtain a sample image, including:

[0029] Get the preset mapped pixel range;

[0030] In the standardized image, the pixel values ​​located in the preset mapping pixel range are transformed to 1, and the remaining pixel values ​​are transformed to 0 to obtain the sample image.

[0031] In a preferred embodiment, the preset mapping model comprises:

[0032] The characteristic value is calculated using the following formula:

[0033] V = ∑c × p;

[0034] Where V represents the feature value corresponding to the input image of a preset mapping model, p represents a pixel value in the input image, and c represents the preset coefficient corresponding to each pixel value, wherein the preset coefficients are different powers of two.

[0035] In a preferred embodiment, based on the correspondence between initial bird images, sample label values, sample images, first bird images, and first bird data, all initial bird images are aggregated to obtain second bird images and corresponding second bird data, including:

[0036] Determine the coverage relationship among multiple initial bird images;

[0037] If an initial bird image covers more than a preset threshold of other initial bird images, then the covered initial bird images will be used as the second bird images.

[0038] If the number of other initial bird images covered by an initial bird image does not exceed a preset threshold, then the covered initial bird images are used as the second bird images.

[0039] Based on the correspondence between the second bird image, the initial bird image, the sample label value, the sample image, the first bird image, and the first bird data, the first bird data corresponding to the second bird image is taken as its corresponding second bird data.

[0040] In a preferred embodiment, each piece of monitoring data is temporarily stored in a cache module and then periodically deleted; the determination of whether the monitoring data in the cache module meets preset trigger conditions includes:

[0041] Determine if the cache module has reached its capacity limit;

[0042] If so, the rapid monitoring module will be invoked for monitoring;

[0043] If not, the default monitoring module will be invoked for monitoring.

[0044] In a preferred embodiment, determining whether the monitoring data in the cache module meets preset trigger conditions includes:

[0045] Determine whether the number of monitoring data of the same bird species saved to the cache module within a preset time period has reached a preset threshold.

[0046] If so, the rapid monitoring module will be invoked for monitoring;

[0047] If not, the default monitoring module will be invoked for monitoring.

[0048] The beneficial effects of using the above embodiments are:

[0049] This invention provides a machine learning-based bird monitoring and protection system for protected areas, comprising a cache module, an acquisition module, a default monitoring module, a fast monitoring module, and a cached monitoring module. The acquisition module acquires monitoring images. The default monitoring module identifies the monitoring images based on a preset machine learning algorithm, obtaining a first bird image and corresponding first bird data, and stores the first bird image and first bird data as monitoring data in the cache module. The fast monitoring module identifies the monitoring images by comparing pixels with the first bird image, obtaining a second bird image and corresponding second bird data, and stores the second bird image and second bird data as monitoring data in the cache module. The cached monitoring module determines whether the monitoring data in the cache module meets preset trigger conditions. If yes, the fast monitoring module is invoked for monitoring; otherwise, the default monitoring module is invoked. The preset trigger conditions are used to determine the quantity of birds currently to be monitored based on the monitoring data. This invention cleverly utilizes a small improvement by using a caching module to temporarily store data and achieves preliminary analysis of monitoring data through preset trigger conditions, thereby determining the current number of monitored birds. Under normal circumstances, the default monitoring module is used to accurately monitor the birds through machine learning, while when the number of birds is large, the fast monitoring module is called to quickly monitor the birds through pixel comparison, thus solving the problem of difficulty in identifying a large number of birds simultaneously in a short period of time in the prior art. Attached Figure Description

[0050] Figure 1 A system architecture diagram of an embodiment of the machine learning-based bird monitoring and protection system for protected areas provided by the present invention;

[0051] Figure 2 for Figure 1 A diagram illustrating the steps involved in executing the rapid monitoring module.

[0052] Figure 3 The diagram shows the specific steps of step S203 in section 2. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Combination Figure 1As shown, a specific embodiment of the present invention discloses a bird monitoring and protection system for protected areas based on machine learning, including a caching module 110, a data acquisition module 120, a default monitoring module 130, a rapid monitoring module 140, and a caching monitoring module 150, wherein:

[0055] Acquisition module 120 is used to acquire monitoring images;

[0056] The default monitoring module 130 is used to identify monitoring images based on a preset machine learning algorithm, obtain the first bird image and the corresponding first bird data, and save the first bird image and the first bird data as monitoring data in the cache module;

[0057] The rapid monitoring module 140 is used to identify the monitoring image based on pixel comparison with the first bird image, obtain the second bird image and the corresponding second bird data, and save the second bird image and the second bird data as monitoring data in the cache module;

[0058] The cache monitoring module 150 is used to determine whether the monitoring data in the cache module meets the preset trigger conditions. If yes, the fast monitoring module is called to perform monitoring; otherwise, the default monitoring module is called to perform monitoring. The preset trigger conditions are used to determine the current number of birds to be monitored based on the monitoring data.

[0059] In the above process, the monitoring image refers to the unprocessed image collected, while the monitoring data represents the data required for bird monitoring. The specific type of data can be flexibly set according to the actual situation, generally including the image and the data of the corresponding bird species. In this embodiment, it specifically refers to the first bird image, the second bird image, the first bird data, and the second bird data. The "first" and "second" are merely prefixes distinguishing different sources of the same type of data. The first bird image and the second bird data are monitoring data obtained by the default monitoring module, while the second bird image and the second bird data are monitoring data obtained by the rapid detection module.

[0060] Normally, this system uses the default monitoring module, which employs any existing image recognition algorithm for bird detection. The fast monitoring module, on the other hand, uses pixel comparison to detect birds. Compared to existing technologies, pixel comparison does not require a large amount of computation, so theoretically its recognition speed is higher than that of the default monitoring module. However, it is conceivable that to achieve a fast result, a sufficient amount of accurate data is needed for pixel comparison.

[0061] This embodiment, through the default monitoring module, can acquire accurate image data for pixel comparison. This monitoring data is temporarily stored in the cache module, which can also be used to monitor the number of birds in the current monitoring environment, thus ensuring the accuracy of the rapid monitoring module. On the one hand, the presence of a large number of birds in the environment is precisely where existing deep learning-based image recognition algorithms fail. On the other hand, a large number of birds means a large amount of monitoring data is acquired. In this case, the monitoring data in the cache module meets the conditions for the accurate operation of the rapid detection module—that is, the data is accurate and sufficient in quantity. Thus, this invention can flexibly adopt appropriate methods for bird monitoring and identification according to the actual situation, overcoming the shortcomings of existing technologies, balancing recognition accuracy and speed, and effectively improving the efficiency and accuracy of bird monitoring. Especially in scenarios with a large number of birds, it can quickly and accurately complete the monitoring task, providing strong support for the protection of birds in protected areas.

[0062] As can be seen, the preset trigger condition represents both the condition for determining the number of birds and the condition for determining whether the current monitoring data meets the requirements for accurate pixel comparison. Similarly, it can be understood that the specific definition of the preset trigger condition can be flexibly set according to the actual situation.

[0063] For example, in a preferred embodiment, the aforementioned preset triggering conditions specifically include:

[0064] Determine whether the number of monitoring data of the same bird species saved to the cache module within a preset time period has reached a preset threshold.

[0065] If so, the rapid monitoring module will be invoked for monitoring;

[0066] If not, the default monitoring module will be invoked for monitoring.

[0067] The advantage of this method lies in its ability to accurately determine the activity level and population size of a particular bird species within the current monitoring area based on the accumulated amount of monitoring data for the same bird species over a preset time period, thereby enabling intelligent switching of monitoring modes. Furthermore, this trigger condition setting based on the amount of monitoring data for the same bird species allows for flexible adjustment of the preset threshold based on actual monitoring needs, making it suitable for diverse monitoring scenarios across different protected areas, bird species, and monitoring periods, further enhancing the system's adaptability and practicality.

[0068] For example, in a preferred embodiment, each piece of monitoring data is temporarily stored in the cache module and then deleted periodically; the aforementioned preset triggering condition may specifically be:

[0069] Determine if the cache module has reached its capacity limit;

[0070] If so, the rapid monitoring module will be invoked for monitoring;

[0071] If not, the default monitoring module will be invoked for monitoring.

[0072] The advantage of this approach lies in its ability to intelligently select the appropriate monitoring module for bird monitoring based on the capacity of the cache module, thus achieving an effective balance between monitoring efficiency and accuracy. Specifically, when the monitoring data in the cache module has not reached its capacity limit, it indicates that the number of birds in the current monitoring environment is relatively small. In this case, calling the default monitoring module can fully leverage its high accuracy advantage based on machine learning algorithms, ensuring accurate identification and data recording of each bird, providing comprehensive and reliable data support for the protection of birds in the reserve. Conversely, when the cache module reaches its capacity limit, it means that the number of birds in the current monitoring area is large, resulting in a large amount of monitoring data. In this case, calling the fast monitoring module, utilizing its pixel-based rapid identification method, can complete the monitoring of a large number of birds in a short time, effectively avoiding monitoring delays or system crashes caused by excessive data volume, ensuring the continuity and timeliness of monitoring work. More importantly, compared to directly monitoring the number of birds, this trigger condition setting based on the cache module capacity is simple and easy to implement. It does not require complex algorithms or large amounts of computing resources to achieve intelligent switching of monitoring modes, reducing system operating costs and maintenance difficulty, and improving system stability and reliability.

[0073] Furthermore, the steps performed by the aforementioned default monitoring module 120 are as follows: Based on a preset machine learning algorithm, the monitored image is identified to obtain a first bird image and corresponding first bird data. The preset machine learning algorithm can be a classic and widely used machine learning algorithm such as Convolutional Neural Network (CNN), Support Vector Machine (SVM), and Random Forest. It is understood that the above algorithms are all existing technologies that can be understood by those skilled in the art. The specific preset machine learning algorithm used and how bird monitoring is implemented are things that those skilled in the art can conceive of, therefore, this document will not elaborate further.

[0074] Furthermore, in combination Figure 2 As shown, in a preferred embodiment, under the premise that the monitoring data in the cache module meets the preset triggering conditions, the steps performed by the above-mentioned fast monitoring module 130 are as follows: Identifying the monitoring image based on pixel comparison with the first bird image to obtain the second bird image and the corresponding second bird data, specifically including:

[0075] S201. Normalize multiple first bird images to obtain multiple sample images of the same size;

[0076] S202. Based on the size distribution of multiple first bird images, the size of the monitoring images is adjusted multiple times to obtain target images of multiple size specifications;

[0077] S203. Set a detection window in the target image with the same size as the sample image, move the detection window and compare the content in the detection window with the sample image pixel by pixel to obtain the second bird image and the corresponding second bird data.

[0078] In the above process, because the first bird image is cropped from the monitoring image and only shows a partial image of the bird, its size may be inconsistent, so normalization processing is required. In this embodiment, normalization refers to adjusting the first bird images of different sizes to a uniform size to ensure that the sample images are of consistent size, which facilitates subsequent pixel comparison operations. For example, all first bird images can be adjusted to 64x64 pixel sample images. This ensures that the number and arrangement of pixels in each sample image are the same during pixel comparison, thereby improving the accuracy and efficiency of the comparison.

[0079] The target image refers to the monitoring image to be analyzed. During actual monitoring, due to factors such as shooting angle and distance, the size of birds in the same monitoring image may vary. Therefore, to ensure accurate identification of birds of different sizes, the monitoring image needs to be adjusted multiple times to generate target images of multiple sizes. For example, if the size of the first bird image is mainly concentrated in a small range, such as 32x32 pixels to 64x64 pixels, then the monitoring image can be adjusted to multiple sizes close to this range, such as 32x32, 48x48, 64x64, etc., to cover bird images of different sizes, thereby ensuring the completeness of the identification.

[0080] A detection window is a region in the target image that is the same size as the sample image. During pixel comparison, the detection window is moved across the target image, scanning the target image region by region. Within each detection window, the image content within the window is compared pixel by pixel with the sample image. By comparing pixel similarity, it is determined whether the region contains a bird image similar to the sample image. If the similarity reaches a preset threshold, the region is considered to contain a bird image, thus obtaining a second bird image and its corresponding data. For example, a pixel difference algorithm can be used to calculate the pixel difference between the image within the detection window and the sample image; if the difference is below a certain threshold, the two are considered similar. This pixel comparison method does not require complex feature extraction and machine learning calculations, therefore the recognition speed is relatively fast, making it suitable for rapid identification and counting in scenarios with a large number of birds.

[0081] Furthermore, the detection window in this embodiment is similar to the convolutional technology in the prior art. However, unlike convolution, which is only responsible for feature extraction and cannot directly complete the recognition work, requiring further processing of the extracted features (such as fully connected layers in neural networks), the detection window in this embodiment can directly obtain the results without subsequent data analysis. This means that by moving the window and using a preset mapping model, birds can be directly identified without the need for complex feature extraction and classification operations like those in convolutional neural networks, making the entire system more efficient.

[0082] Specifically, in a preferred embodiment, step S201, which involves normalizing multiple first bird images to obtain multiple sample images of the same size, specifically includes:

[0083] The first bird image is resized while preserving its features to obtain a standardized image.

[0084] The standardized image is transformed by pixel value transformation to obtain the sample image.

[0085] In the above process, feature-preserving size adjustment refers to using a method that preserves key image features when resizing the image. This ensures that the adjusted image retains the bird's distinctive features, such as outlines and textures, while maintaining a uniform size, facilitating subsequent identification and comparison. Pixel value transformation refers to adjusting the pixel values ​​of the standardized image to meet the requirements of subsequent processing. This includes normalizing pixel values ​​to a specific range or converting the color image to grayscale, reducing data volume and improving processing efficiency. Through feature-preserving size adjustment and pixel value transformation, sample images with uniform size, preserved features, and suitable pixel values ​​can be obtained, facilitating subsequent pixel comparison operations and improving the accuracy and efficiency of the comparison.

[0086] Specifically, in a preferred embodiment, the above step of: performing feature-preserving size adjustment on the first bird image to obtain a standardized image, specifically includes:

[0087] Adjust the size based on the target to obtain the first pixel and its neighboring pixels;

[0088] The pixel values ​​of the first pixel and its neighboring pixels are weighted and averaged to obtain the standardized pixels, and multiple standardized pixels constitute the standardized image.

[0089] In the above process, the target adjustment size is the size of the sample image to be adjusted. The first pixel is the pixel to be adjusted. The specific selection range of neighboring pixels can be flexibly set based on the target adjustment size, depending on the specific situation. For example, when adjusting to a smaller size, a larger range of neighboring pixels can be selected for weighted averaging to better preserve features. The advantage of this adjustment method is that the sample image still has high recognizability even after the size is reduced, facilitating accurate comparison with the content in the detection window.

[0090] Further, the above steps involve: performing pixel value transformation on the standardized image to obtain a sample image, including:

[0091] Get the preset mapped pixel range;

[0092] In the standardized image, the pixel values ​​located in the preset mapping pixel range are transformed to 1, and the remaining pixel values ​​are transformed to 0 to obtain the sample image.

[0093] In the above process, the preset pixel interval refers to a range of pixel values ​​pre-defined based on the characteristics of birds and monitoring requirements. For example, if birds in a certain habitat area are black, then the range of black or near-black colors can be set as the preset mapped pixel interval. The advantage of this is that it effectively distinguishes the bird portion of the image from the background or other non-bird parts, improving the contrast and feature salience of the sample image, thereby enhancing the accuracy and efficiency of subsequent pixel comparison. By transforming pixel values ​​within the preset mapped pixel interval to 1 and other pixel values ​​to 0, a binarized sample image can be generated, making the bird's outline and main features clearer, facilitating rapid identification and comparison.

[0094] Furthermore, in combination Figure 3 As shown, in a preferred embodiment, step S203, which involves setting a detection window in the target image with the same size as the sample image, moving the detection window, and comparing the content in the detection window with the sample image pixel by pixel to obtain the second bird image and the corresponding second bird data, specifically includes:

[0095] S301. Obtain a preset mapping model, wherein the preset mapping model is used to map different images to unique feature values;

[0096] S302. Input the sample images into the preset mapping model to obtain the sample label value corresponding to each sample image;

[0097] S303. Set a detection window in the target image with the same size as the sample image, move the detection window and input the contents of the detection window into the preset mapping model each time it is moved to obtain the set of output feature values.

[0098] S304. Match the sample label value with the feature value in the output feature value set, and obtain the position information of the detection window corresponding to the successfully matched feature value.

[0099] S305. Extract a portion of the monitored image based on the location information and use it as the initial bird image corresponding to the sample label value;

[0100] S306. Based on the correspondence between the initial bird images, sample label values, sample images, first bird images, and first bird data, summarize all the initial bird images to obtain the second bird images and the corresponding second bird data.

[0101] In the above process, the sole purpose of the pre-defined mapping model output is to ensure that each sample image can be accurately identified and distinguished. Specifically, the pre-defined mapping model maps different images to unique feature values, giving each sample image a unique identifier, namely a sample label value. This allows for accurate determination of whether the content within the detection window matches the sample image during feature value matching. The sample label value is the feature value corresponding to the sample image.

[0102] Understandably, if a mechanical pixel comparison is performed using conditional judgments, the pixel comparison process is slow because this method involves repeatedly making numerous conditional judgments, resulting in a large computational load and consuming significant computing resources (mainly due to the time delay caused by the computer reading data). In this embodiment, however, the comparison is performed by calculating feature values ​​using a preset mapping model. This transforms complex pixel comparisons into simple numerical comparisons, greatly reducing conditional judgments and thus improving comparison speed. Especially when processing large amounts of data, this embodiment can complete simultaneous comparisons of large batches of data without having to compare them one by one according to the movement order of the detection window, making its speed advantage even more pronounced.

[0103] Specifically, in a preferred embodiment, the preset mapping model includes:

[0104] The characteristic value is calculated using the following formula:

[0105] V = ∑c × p;

[0106] Where V represents the feature value corresponding to the input image of a preset mapping model, p represents a pixel value in the input image, and c represents the preset coefficient corresponding to each pixel value, wherein the preset coefficients are different powers of two.

[0107] This model can work in conjunction with the pixel value transformation method described above, which maps pixels to 0 and 1, thereby ensuring the uniqueness of the output of the preset mapping model.

[0108] Furthermore, as mentioned earlier, there is a one-to-one correspondence between the initial bird image, sample label value, sample image, first bird image, and first bird data. Therefore, once the initial bird image is determined, the bird species in the image can be directly identified. However, it's understandable that during feature value matching, the feature values ​​in the feature dataset are obtained based on target images of different sizes. Therefore, when mapping the successfully matched feature values ​​back to the monitoring image, the corresponding detection window sizes will differ. This can lead to multiple initial bird images having varying sizes and potentially overlapping. Therefore, it's necessary to summarize the data again to obtain accurate second bird images and data. By summarizing all initial bird images, duplicate counting can be avoided, thereby improving the accuracy and reliability of the monitoring data.

[0109] Specifically, in a preferred embodiment, step S306, based on the correspondence between the initial bird images, sample label values, sample images, first bird images, and first bird data, summarizes all initial bird images to obtain second bird images and corresponding second bird data, specifically including:

[0110] Determine the coverage relationship among multiple initial bird images;

[0111] If an initial bird image covers more than a preset threshold of other initial bird images, then the covered initial bird images will be used as the second bird images.

[0112] If the number of other initial bird images covered by an initial bird image does not exceed a preset threshold, then the covered initial bird images are used as the second bird images.

[0113] Based on the correspondence between the second bird image, the initial bird image, the sample label value, the sample image, the first bird image, and the first bird data, the first bird data corresponding to the second bird image is taken as its corresponding second bird data.

[0114] By determining the coverage relationship of the initial bird images and identifying the second bird image based on whether the coverage exceeds a preset threshold, the overlapping coverage problem between the initial bird images can be effectively solved. Specifically:

[0115] When an initial bird image covers more than a preset threshold of other initial bird images, the covered initial bird images are used as the second bird images. This preserves a more complete and clearer bird image, avoiding information loss or double counting due to overlap. Conversely, when an initial bird image covers no more than a preset threshold of other initial bird images, the covered initial bird images may only contain misidentified features. In this case, the covered initial bird images are used as the second bird images. This ensures that the selected second bird images are highly representative and accurate, reducing errors caused by image overlap.

[0116] This invention provides a machine learning-based bird monitoring and protection system for protected areas, comprising a cache module, an acquisition module, a default monitoring module, a fast monitoring module, and a cached monitoring module. The acquisition module acquires monitoring images. The default monitoring module identifies the monitoring images based on a preset machine learning algorithm, obtaining a first bird image and corresponding first bird data, and stores the first bird image and first bird data as monitoring data in the cache module. The fast monitoring module identifies the monitoring images by comparing pixels with the first bird image, obtaining a second bird image and corresponding second bird data, and stores the second bird image and second bird data as monitoring data in the cache module. The cached monitoring module determines whether the monitoring data in the cache module meets preset trigger conditions. If yes, the fast monitoring module is invoked for monitoring; otherwise, the default monitoring module is invoked. The preset trigger conditions are used to determine the quantity of birds currently to be monitored based on the monitoring data. This invention cleverly utilizes a small improvement by using a caching module to temporarily store data and achieves preliminary analysis of monitoring data through preset trigger conditions, thereby determining the current number of monitored birds. Under normal circumstances, the default monitoring module is used to accurately monitor the birds through machine learning, while when the number of birds is large, the fast monitoring module is called to quickly monitor the birds through pixel comparison, thus solving the problem of difficulty in identifying a large number of birds simultaneously in a short period of time in the prior art.

[0117] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0118] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A machine learning-based bird monitoring and protection system for protected areas, characterized in that, It includes a caching module, a data acquisition module, a default monitoring module, a fast monitoring module, and a cache monitoring module, among which: The acquisition module is used to acquire monitoring images; The default monitoring module is used to identify monitoring images based on a preset machine learning algorithm, obtain the first bird image and the corresponding first bird data, and save the first bird image and the first bird data as monitoring data in the cache module; The rapid monitoring module is used to normalize multiple first bird images to obtain multiple sample images of the same size; Based on the size distribution of multiple first bird images, the size of the monitoring images was adjusted multiple times to obtain target images of multiple size specifications; Obtain a preset mapping model, whereby the preset mapping model is used to map different images to unique feature values; The sample images are input into a preset mapping model to obtain the sample label value corresponding to each sample image; Set a detection window with the same size as the sample image in the target image, move the detection window and input the contents of the detection window into the preset mapping model each time it is moved to obtain the set of output feature values; The sample label value is matched with the feature value in the output feature value set, and the position information of the detection window corresponding to the successfully matched feature value is obtained. Based on the location information, a portion of the monitored image content is extracted and used as the initial bird image corresponding to the sample label value; Determine the coverage relationship among multiple initial bird images; If an initial bird image covers more than a preset threshold of other initial bird images, then the covered initial bird images will be used as the second bird images. If the number of other initial bird images covered by an initial bird image does not exceed a preset threshold, then the covered initial bird images are used as the second bird images. Based on the correspondence between the second bird image, the initial bird image, the sample label value, the sample image, the first bird image, and the first bird data, the first bird data corresponding to the second bird image is taken as its corresponding second bird data. The cached monitoring module is used to determine whether the monitoring data in the cached module meets the preset trigger conditions. If yes, the fast monitoring module is called to perform monitoring; otherwise, the default monitoring module is called to perform monitoring. The preset trigger conditions are used to determine the current number of birds to be monitored based on the monitoring data.

2. The machine learning-based bird monitoring and protection system for protected areas according to claim 1, characterized in that, Normalize multiple first bird images to obtain multiple sample images of the same size, including: The first bird image is resized while preserving its features to obtain a standardized image. The standardized image is transformed by pixel value transformation to obtain the sample image.

3. The machine learning-based bird monitoring and protection system for protected areas according to claim 2, characterized in that, The first bird image is resized with feature preservation to obtain a standardized image, including: Adjust the size based on the target to obtain the first pixel and its neighboring pixels; The pixel values ​​of the first pixel and its neighboring pixels are weighted and averaged to obtain the standardized pixels, and multiple standardized pixels constitute the standardized image.

4. The machine learning-based bird monitoring and protection system for protected areas according to claim 2, characterized in that, The standardized image is transformed by pixel values ​​to obtain a sample image, including: Get the preset mapped pixel range; In the standardized image, the pixel values ​​located in the preset mapping pixel range are transformed to 1, and the remaining pixel values ​​are transformed to 0 to obtain the sample image.

5. The machine learning-based bird monitoring and protection system for protected areas according to claim 4 is characterized in that, The preset mapping models include: The characteristic value is calculated using the following formula: ； in, This represents the feature value corresponding to the input image of a predefined mapping model. This represents a pixel value in the input image. This indicates that each pixel value corresponds to a preset coefficient, where the preset coefficients are different powers of two.

6. The machine learning-based bird monitoring and protection system for protected areas according to claim 1, characterized in that, Each piece of monitoring data is temporarily stored in the cache module and then deleted periodically; Determine whether the monitoring data in the cache module meets the preset trigger conditions, including: Determine if the cache module has reached its capacity limit; If so, the rapid monitoring module will be invoked for monitoring; If not, the default monitoring module will be invoked for monitoring.

7. The machine learning-based bird monitoring and protection system for protected areas according to claim 1, characterized in that, Determine whether the monitoring data in the cache module meets the preset trigger conditions, including: Determine whether the number of monitoring data of the same bird species saved to the cache module within a preset time period has reached a preset threshold. If so, the rapid monitoring module will be invoked for monitoring; If not, the default monitoring module will be invoked for monitoring.

Images