Bird intelligent identification monitoring system and method based on image recognition and cloud computing

Abstract

The application discloses a bird intelligent recognition monitoring system and method based on image recognition and cloud computing, relates to the field of image recognition, and solves the problem that birds cannot be efficiently and accurately recognized due to fine features and similar parts, and comprises a body type extraction unit, a color extraction unit, a species prediction unit, an element analysis unit and an image recognition module; the body type extraction unit is used for extracting biological body type features of different parts of birds; the color extraction unit is used for extracting biological color features of different parts of birds according to actual images; the species prediction unit is used for predicting birds of known actual species according to actual area arrays and pixel interval sequences of different parts of birds; the element analysis unit is used for analyzing prediction accuracy of different target parts; and the image recognition module is used for recognizing birds in a to-be-recognized image according to recognition elements, and the application realizes efficient and accurate recognition of birds through image recognition.

Description

Technical Field

[0001] This invention belongs to the field of image recognition technology, specifically a bird intelligent identification and monitoring system and method based on image recognition and cloud computing. Background Technology

[0002] Birds are important indicator species of ecosystems. Their population size, species distribution, health status and behavioral patterns can directly reflect the regional ecological environment quality and biodiversity level. Conducting research on automated bird identification, monitoring and analysis has important practical significance and application value in fields such as wildlife protection, ecological environment assessment, biological population dynamics monitoring and popular science education. Existing technologies typically achieve bird identification and monitoring by constructing different recognition models. Although bird identification results can be obtained quickly, there is still room for improvement in the accuracy of the identification results. At the same time, because the same parts of different birds have certain color and size similarities, it is impossible to effectively extract subtle features in the image, thus making it impossible to efficiently identify and monitor birds in the image. Therefore, this invention proposes a bird intelligent identification and monitoring system and method based on image recognition and cloud computing. Summary of the Invention

[0003] The purpose of this invention is to propose an intelligent bird identification and monitoring system and method based on image recognition and cloud computing, so as to solve the problem mentioned in the background art of the inability to efficiently identify and monitor birds in images.

[0004] The objective of this invention can be achieved through the following technical solutions: The intelligent bird identification and monitoring system based on image recognition and cloud computing includes a data acquisition module, a feature extraction module, a feature matching module, and an image recognition module. The data acquisition module collects actual images of different bird species within the monitored area and sends these images to the feature extraction module. The feature extraction module includes a body shape extraction unit and a color extraction unit. The body shape extraction unit extracts biological body shape features of different parts of the bird based on the actual images, obtaining an array of actual areas for each part. The color extraction unit extracts biological color features of different parts of the bird based on the actual images, obtaining a sequence of pixel intervals for each part. The actual area arrays and pixel interval sequences of the different bird parts are then sent to the feature matching module. The feature matching module includes a species prediction unit and an element analysis unit. The species prediction unit is used to predict the known species of birds based on the actual area array and pixel interval sequence of different parts of the bird, and to obtain a primary predicted species and a secondary predicted species. The primary or secondary predicted species obtained based on the identification of different target parts is then sent to the element analysis unit. The element analysis unit is used to analyze the prediction accuracy of different target parts based on the primary or secondary predicted species, and to obtain the identification elements of different target parts. The identification elements of different target parts are then sent to the image recognition module. The data acquisition module is also used to acquire images of birds to be identified in the area to be monitored, and send the images of birds to be identified to the image recognition module; the image recognition module is used to perform image recognition on the birds in the images to be identified based on the recognition elements, and to identify the predicted species of birds in the images to be identified.

[0005] Furthermore, the extraction process of the body shape extraction unit includes the following steps: By inputting actual images of different birds into the target detection model, different parts of the birds in the actual images are detected; Select any actual image of a bird as the analysis object, obtain the actual shooting focal length and actual shooting distance of the corresponding actual image, count the number of pixels in any part of the actual image, and multiply the actual shooting distance by the number of pixels in the corresponding part and divide by the actual shooting focal length to obtain the actual area of ​​different parts in the actual image. Similarly, multiple actual images of the corresponding bird are selected as the analysis object, and the actual area of ​​different parts in all actual images is obtained. The different actual areas are associated with different parts to obtain an array of actual areas of different parts of the corresponding bird.

[0006] Furthermore, the color extraction process of the color extraction unit includes the following steps: Obtain the pixel values ​​of all pixels within any part of the actual image, sum the pixel values ​​and take the average value to obtain the average pixel value of the corresponding part in the actual image; The standard deviation of pixel values ​​of all pixels in any part is calculated by the standard deviation formula. The first value of the interval is obtained by subtracting the standard deviation of pixel values ​​from the average value of pixels. The second value of the interval is obtained by summing the average value of pixels and the standard deviation of pixel values. The pixel reference intervals for different parts are constructed with the first value of the interval as the left endpoint and the second value of the interval as the right endpoint. Similarly, multiple actual images of the corresponding bird are selected as the analysis object, and the pixel reference intervals of different parts in all actual images are obtained. The different pixel reference intervals are then associated with different parts to obtain the pixel interval sequence of different parts of the corresponding bird.

[0007] Furthermore, the prediction process of the species prediction unit includes the following: Take pictures of any known bird species to obtain real-time images of the corresponding birds. Input the real-time images of the corresponding birds into the target detection model to detect different parts of the corresponding birds in the real-time images. Select any part of the real-time image as the analysis object and record it as the target part. Obtain the pixel values ​​of different pixels in the target part, and sum the pixel values ​​and take the average value to obtain the average pixel value of the target part in the real-time image. Obtain the pixel reference interval within all pixel interval sequences corresponding to the target area, and compare the average pixel value of the target area in the real-time image with the pixel reference interval.

[0008] Furthermore, the prediction process of the species prediction unit also includes the following: If the average pixel value of the target area in the real-time image falls within any pixel reference interval, then the bird species corresponding to the pixel reference interval is recorded as a predicted species. If the average pixel value of the target area in the real-time image does not belong to any pixel reference interval, the absolute value of the pixel value minus the midpoint value of all pixel reference intervals is obtained to get the pixel deviation value of the target area in the real-time image. The minimum pixel deviation value is obtained by iterating and comparing different pixel deviation values, and the bird species corresponding to the undetermined reference interval of the minimum pixel deviation value is recorded as a predicted species. Similarly, any part of the real-time image is selected as the target part for the same analysis, thereby identifying the corresponding prediction category based on different target parts.

[0009] Furthermore, the prediction process of the species prediction unit also includes the following: Obtain the real-time shooting focal length and real-time shooting object distance of the real-time image, count the number of pixels in the target area in the real-time image, and multiply the real-time shooting object distance by the number of pixels in the target area and divide by the real-time shooting focal length to obtain the real-time area of ​​the target area in the real-time image. Obtain the actual area from all the actual area arrays corresponding to the target part, and take the absolute value of the real-time area deviation of the target part by subtracting the different actual areas from the real-time area. Iterate and compare the different real-time area deviation values ​​to obtain the minimum real-time area deviation of the target part. The bird species corresponding to the minimum real-time area deviation is recorded as the secondary predicted species obtained based on the target location identification. Similarly, any part of the real-time image is selected as the target part for the same analysis, thereby identifying the corresponding secondary prediction category based on different target parts.

[0010] Furthermore, the analysis process of the elemental analysis unit includes the following: The predicted types and secondary predicted types obtained based on the same target location are compared with the actual types. If the predicted category obtained based on the same target location includes the actual category, but the predicted category does not include the actual category, then the accuracy of the first prediction for the corresponding target location is incremented by one. If the predicted category obtained based on the same target location does not contain the actual category, but the predicted category contains the actual category, then the accuracy of the second prediction for the corresponding target location is incremented by one. If both the first-prediction category and the second-prediction category contain the actual category, then the accuracy of the first-prediction and the accuracy of the second-prediction for the corresponding target part are incremented by one. If neither the first prediction category nor the second prediction category contains the actual category, no action will be taken. Similarly, the same known bird species is photographed again to obtain multiple real-time images of the corresponding bird. The target parts in different real-time images are analyzed repeatedly to obtain the accuracy of the first prediction and the accuracy of the second prediction for different target parts.

[0011] Furthermore, the analysis process of the elemental analysis unit also includes the following: The number of real-time images captured is counted, and the accuracy of the first prediction is divided by the number of images captured to obtain the accuracy of the first prediction for different target parts. The accuracy of the second prediction is divided by the number of images captured to obtain the accuracy of the second prediction for different target parts. Compare the accuracy of the first prediction with the accuracy of the second prediction for the same target location; If the accuracy of a single prediction for the same target part is greater than or equal to the accuracy of a double prediction, then the average pixel value of the corresponding target part will be used as the recognition element. If the accuracy of a single prediction for the same target part is less than the accuracy of a double prediction, then the real-time area of ​​the corresponding target part will be used as the identification element. Similarly, the accuracy rates of the first and second predictions for different target parts are compared to obtain the identification elements for different target parts.

[0012] Furthermore, the recognition process of the image recognition module includes the following: The image to be identified is input into the target detection model, which detects different parts of the bird in the image. Obtain the recognition elements corresponding to different parts. When the recognition element is the pixel average value, calculate the pixel average value of the corresponding part and analyze it to obtain a prediction category. When the identified element is a real-time area, the real-time area of ​​the corresponding part is calculated and analyzed to obtain the secondary prediction type; Similarly, based on the analysis of different parts, the primary and secondary predicted species of birds in the image to be identified are obtained; If there are duplicate predicted types in the first and second predictions, the corresponding predicted type is recorded as the undetermined type of bird in the image to be identified. At the same time, the number of times different undetermined types appear in the first and second predictions is counted and recorded as the prediction number. The different prediction numbers are compared to obtain the maximum prediction number, and the undetermined type corresponding to the maximum prediction number is recorded as the predicted type of bird in the image to be identified. If there are no duplicate predicted species in the primary and secondary predictions, then the primary and secondary predictions are collectively recorded as the predicted species of birds in the image to be identified.

[0013] Secondly, a bird intelligent identification and monitoring method based on image recognition and cloud computing, the method includes the following steps: Step S1: Extract the biological body shape features of different parts of the bird based on the actual image of the bird, and obtain the actual area array of different parts of the bird. Step S2: Based on the actual image of the bird, extract the biological color features of different parts of the bird to obtain the pixel interval sequence of different parts of the bird. Step S3: Based on the actual area array and pixel interval sequence of different parts of the bird, predict the known actual bird species and obtain the first predicted species and the second predicted species. Step S4: Analyze the prediction accuracy of different target parts based on the primary prediction type or the secondary prediction type, and obtain the identification elements of different target parts. Step S5: Based on the identification elements, perform image recognition on the birds in the image to be identified, and obtain the predicted species of birds in the image to be identified.

[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. The present invention first acquires actual images of different bird species in the bird population in the area to be monitored through the data acquisition module. The body shape extraction unit extracts the biological body shape features of different parts of the bird based on the actual images of the bird, and extracts the actual area array of different parts of the bird. Then, the color extraction unit extracts the biological color features of different parts of the bird based on the actual images, and extracts the pixel interval sequence of different parts of the bird, so as to realize the accurate extraction and identification of bird features. 2. This invention uses a species prediction unit to predict bird species based on the actual area array and pixel interval sequence of different parts of the bird, thereby predicting primary and secondary species. The primary or secondary predicted species obtained based on different target parts are sent to an element analysis unit. The element analysis unit then analyzes the prediction accuracy of different target parts based on the primary or secondary predicted species, thereby obtaining the identification elements of different target parts. The identification elements of different target parts are sent to an image recognition module. On the one hand, images of birds to be identified in the area to be monitored are also collected and sent to the image recognition module. Finally, the image recognition module performs image recognition on the birds in the images to be identified based on the identification elements, thereby identifying the predicted species of birds in the images to be identified. Through image recognition, efficient and accurate bird identification is achieved. Attached Figure Description

[0015] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0016] Figure 1 This is an overall system block diagram of the present invention; Figure 2 This is a flowchart of the feature extraction module in this invention; Figure 3 This is a flowchart of the feature matching module in this invention; Figure 4 This is a flowchart of the method of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0018] Example 1, please refer to Figures 1-3 As shown, the technical solution provided by the present invention is: a bird intelligent identification and monitoring system based on image recognition and cloud computing. The system first extracts biological features from different parts of actual images of bird flocks that have previously inhabited the monitoring area. Then, it analyzes the importance of biological features corresponding to different parts using real-time images of known bird species to obtain identification elements corresponding to different parts. Next, it uses the identification elements to perform image recognition on images of birds to be identified in the monitoring area to identify the predicted species of birds in the images to be identified. The system includes a data acquisition module, a feature extraction module, a feature matching module, and an image recognition module. In this embodiment, the data acquisition module is used to acquire actual images of different bird species among the bird flocks inhabiting the area to be monitored, and send the actual images of different birds to the feature extraction module; Specifically, the actual images include, but are not limited to, images of birds in different states such as flying, standing, and gliding. The actual images also include images of birds under different lighting conditions. The actual images are stored in a database. When the data acquisition module sends different actual images to the feature extraction module, the feature extraction module performs cloud computing on the features of different actual images to extract the biological characteristics of birds. In the specific implementation process, the feature extraction module includes a body shape extraction unit and a color extraction unit; In this embodiment, the body shape extraction unit is used to extract the biological body shape features of different parts of the bird based on the actual image of the bird, and to extract the actual area array of different parts of the bird. The extraction process includes the following steps: By inputting actual images of different birds into the target detection model, different parts of the birds in the actual images are detected; In this embodiment, the target detection model is the existing Mask R-CNN model. The target detection model can depict the outline of different parts of the bird in the actual image and annotate the corresponding part names. The target detection models that appear later in this embodiment are the same as the target detection model here, and will not be described again later. Select any actual image of a bird as the analysis object, obtain the actual shooting focal length and actual shooting distance of the corresponding actual image, count the number of pixels in any part of the actual image, and multiply the actual shooting distance by the number of pixels in the corresponding part and divide by the actual shooting focal length to obtain the actual area of ​​different parts in the actual image. Similarly, multiple actual images of the corresponding bird are selected as the analysis object, and the actual area of ​​different parts in all actual images is obtained. The different actual areas are associated with different parts to obtain an array of actual areas of different parts of the corresponding bird. Specifically, a bird has multiple body parts, and each body part has a corresponding actual area array. Since the size of different individuals of the same bird species also varies, the actual area array contains multiple elements, each of which represents the actual area of ​​the corresponding body part of the bird. For example, for the beak of bird A, there are multiple actual areas in the actual area array corresponding to the beak. In this embodiment, the actual shooting focal length and the actual shooting object distance can be obtained from the shooting parameters of the corresponding actual image, and the size of a single pixel is 1×1.

[0019] In this embodiment, the color extraction unit is used to extract the biological color features of different parts of the bird based on the actual image, and to extract the pixel interval sequence of different parts of the bird. The extraction process includes the following steps: Obtain the pixel values ​​of all pixels within any part of the actual image, sum the pixel values ​​and take the average value to obtain the average pixel value of the corresponding part in the actual image; The standard deviation of pixel values ​​of all pixels in any part is calculated by the standard deviation formula. The first value of the interval is obtained by subtracting the standard deviation of pixel values ​​from the average value of pixels. The second value of the interval is obtained by summing the average value of pixels and the standard deviation of pixel values. The pixel reference intervals for different parts are constructed with the first value of the interval as the left endpoint and the second value of the interval as the right endpoint. Similarly, multiple actual images of the corresponding bird are selected as the analysis object, and the pixel reference intervals of different parts in all actual images are analyzed. The different pixel reference intervals are then associated with different parts to obtain the pixel interval sequence of different parts of the corresponding bird. Specifically, since there are certain differences in color between different individuals of the same bird species, there are multiple elements in the pixel interval sequence, each element representing the pixel reference interval of the corresponding part of the bird. The feature extraction module sends the actual area array and pixel interval sequence of different parts of the bird to the feature matching module.

[0020] In this embodiment, the feature matching module includes a category prediction unit and an element analysis unit; In the specific implementation process, the species prediction unit is used to predict the species of birds based on the actual area array and pixel interval sequence of different parts of the bird, and to obtain the primary predicted species and the secondary predicted species. The prediction process includes the following: Take pictures of any known bird species to obtain real-time images of the corresponding birds. Input the real-time images of the corresponding birds into the target detection model to detect different parts of the corresponding birds in the real-time images. Select any part of the real-time image as the analysis object and record it as the target part. Obtain the pixel values ​​of different pixels in the target part, and sum the pixel values ​​and take the average value to obtain the average pixel value of the target part in the real-time image. Obtain the pixel reference interval within all pixel interval sequences corresponding to the target area, and compare the average pixel value of the target area in the real-time image with the pixel reference interval; If the average pixel value of the target area in the real-time image falls within any pixel reference interval, then the bird species corresponding to the pixel reference interval is recorded as a predicted species. If the average pixel value of the target area in the real-time image does not belong to any pixel reference interval, the absolute value of the pixel value minus the midpoint value of all pixel reference intervals is obtained to get the pixel deviation value of the target area in the real-time image. The minimum pixel deviation value is obtained by iterating and comparing different pixel deviation values, and the bird species corresponding to the undetermined reference interval of the minimum pixel deviation value is recorded as a predicted species. Similarly, any part of the real-time image is selected as the target part for the same analysis, thereby identifying the corresponding prediction category based on different target parts; It should be noted that a single prediction can contain multiple categories; Obtain the real-time shooting focal length and real-time shooting object distance of the real-time image, count the number of pixels in the target area in the real-time image, and multiply the real-time shooting object distance by the number of pixels in the target area and divide by the real-time shooting focal length to obtain the real-time area of ​​the target area in the real-time image. Obtain the actual area from all the actual area arrays corresponding to the target part, and take the absolute value of the real-time area deviation of the target part by subtracting the different actual areas from the real-time area. Iterate and compare the different real-time area deviation values ​​to obtain the minimum real-time area deviation of the target part. The bird species corresponding to the minimum real-time area deviation is recorded as the secondary predicted species obtained based on the target location identification. Similarly, any part of the real-time image is selected as the target part for the same analysis, thereby identifying the corresponding secondary prediction category based on different target parts; It should be noted that there can be multiple types of secondary predictions; The type prediction unit sends the primary or secondary prediction type obtained based on the identification of different target parts to the element analysis unit.

[0021] In this embodiment, the element analysis unit is used to analyze the prediction accuracy of different target parts based on the primary prediction type or the secondary prediction type, and to obtain the identification elements of different target parts. The analysis process includes the following: The predicted types and secondary predicted types obtained based on the same target location are compared with the actual types. If the predicted category obtained based on the same target location includes the actual category, but the predicted category does not include the actual category, then the accuracy of the first prediction for the corresponding target location is incremented by one. If the predicted category obtained based on the same target location does not contain the actual category, but the predicted category contains the actual category, then the accuracy of the second prediction for the corresponding target location is incremented by one. If both the primary and secondary predictions of the same target location contain the actual target type, then the accuracy of the primary and secondary predictions for the corresponding target location will be incremented by one. If neither the first prediction category nor the second prediction category contains the actual category, no action will be taken. It should be noted that if the predicted species obtained based on the same target location does not include the actual species, and the predicted species does not include the actual species, the corresponding bird can be considered to be an unknown bird. This embodiment will not conduct a specific analysis. Similarly, the same known bird species is photographed again to obtain multiple real-time images of the corresponding bird. The target parts in different real-time images are analyzed repeatedly to obtain the accuracy of the first prediction and the accuracy of the second prediction for different target parts. The number of real-time images captured is counted, and the accuracy of the first prediction is divided by the number of images captured to obtain the accuracy of the first prediction for different target parts. The accuracy of the second prediction is divided by the number of images captured to obtain the accuracy of the second prediction for different target parts. Compare the accuracy of the first prediction with the accuracy of the second prediction for the same target location; If the accuracy of a single prediction for the same target part is greater than or equal to the accuracy of a double prediction, then the average pixel value of the corresponding target part will be used as the recognition element. If the accuracy of a single prediction for the same target part is less than the accuracy of a double prediction, then the real-time area of ​​the corresponding target part will be used as the identification element. Similarly, the accuracy rates of the first and second predictions for different target parts are compared to obtain the identification elements for different target parts. The feature matching module sends the recognition elements of different target parts to the image recognition module.

[0022] In this embodiment, the data acquisition module is also used to acquire images of birds to be identified within the area to be monitored, and send the images of birds to be identified to the image recognition module; In practical implementation, the image recognition module is used to perform image recognition on birds in the image to be recognized based on the recognition elements, and to obtain the predicted species of birds in the image to be recognized. The recognition process includes the following: The image to be identified is input into the target detection model, which detects different parts of the bird in the image. Obtain the recognition elements corresponding to different parts. When the recognition element is the pixel average value, calculate the pixel average value of the corresponding part and analyze it to obtain a prediction category. When the identified element is a real-time area, the real-time area of ​​the corresponding part is calculated and analyzed to obtain the secondary prediction type; It should be explained that the analysis process for obtaining the first predicted species here is the same as the analysis process for obtaining the first predicted species in the species prediction unit, and the analysis process for obtaining the second predicted species here is the same as the analysis process for obtaining the first predicted species in the species prediction unit, which will not be repeated here. Similarly, based on the analysis of different parts, the primary and secondary predicted species of birds in the image to be identified are obtained; If there are duplicate predicted types in the first and second predictions, the corresponding predicted type is recorded as the undetermined type of bird in the image to be identified. At the same time, the number of times different undetermined types appear in the first and second predictions is counted and recorded as the prediction number. The different prediction numbers are compared to obtain the maximum prediction number, and the undetermined type corresponding to the maximum prediction number is recorded as the predicted type of bird in the image to be identified. If there are no overlapping predicted categories in the primary and secondary prediction categories, then the primary and secondary prediction categories are collectively recorded as the predicted bird category in the image to be identified. In this embodiment, when there is a duplicate predicted species between the first and second predicted species, a unique species of bird in the image to be identified can be obtained. When there is no duplicate predicted species between the first and second predicted species, a species similar to the bird in the image to be identified can be identified for reference.

[0023] Example 2, please refer to Figure 4 As shown, based on another concept of the same invention, a bird intelligent identification and monitoring method based on image recognition and cloud computing is proposed, including the following steps: Step S1: Extract the biological body shape features of different parts of the bird based on the actual image of the bird, and obtain the actual area array of different parts of the bird. Step S2: Based on the actual image of the bird, extract the biological color features of different parts of the bird to obtain the pixel interval sequence of different parts of the bird. Step S3: Based on the actual area array and pixel interval sequence of different parts of the bird, predict the known actual bird species and obtain the first predicted species and the second predicted species. Step S4: Analyze the prediction accuracy of different target parts based on the primary prediction type or the secondary prediction type, and obtain the identification elements of different target parts. Step S5: Based on the identification elements, perform image recognition on the birds in the image to be identified, and obtain the predicted species of birds in the image to be identified.

[0024] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A bird intelligent identification and monitoring system based on image recognition and cloud computing, characterized in that, include: The data acquisition module is used to collect actual images of different bird species among the bird populations inhabiting the area to be monitored; The feature extraction module includes a body shape extraction unit and a color extraction unit; The body shape extraction unit is used to extract the biological body shape features of different parts of the bird based on the actual image of the bird, and to extract the actual area array of different parts of the bird; The color extraction unit is used to extract the biological color features of different parts of the bird based on the actual image, and to extract the pixel interval sequence of different parts of the bird. The feature matching module includes a category prediction unit and an element analysis unit; The species prediction unit is used to predict the species of birds based on the actual area array and pixel interval sequence of different parts of the bird, and to obtain the first predicted species and the second predicted species. The element analysis unit is used to analyze the prediction accuracy of different target parts based on the primary prediction type or the secondary prediction type, and to obtain the identification elements of different target parts. The data acquisition module is also used to acquire images of birds to be identified within the area to be monitored; The image recognition module is used to perform image recognition on birds in the image to be recognized based on the recognition elements, and to obtain the predicted species of birds in the image to be recognized.

2. The intelligent bird identification and monitoring system based on image recognition and cloud computing according to claim 1, characterized in that, The extraction process of the body shape extraction unit includes the following steps: By inputting actual images of different birds into the target detection model, different parts of the birds in the actual images are detected; Select any actual image of a bird as the analysis object, obtain the actual shooting focal length and actual shooting distance of the corresponding actual image, count the number of pixels in any part of the actual image, and multiply the actual shooting distance by the number of pixels in the corresponding part and divide by the actual shooting focal length to obtain the actual area of ​​different parts in the actual image. Similarly, multiple actual images of the corresponding bird are selected as the analysis object, and the actual area of ​​different parts in all actual images is obtained. The different actual areas are associated with different parts to obtain an array of actual areas of different parts of the corresponding bird.

3. The intelligent bird identification and monitoring system based on image recognition and cloud computing according to claim 1, characterized in that, The color extraction process of the color extraction unit includes the following steps: Obtain the pixel values ​​of all pixels within any part of the actual image, sum the pixel values ​​and take the average value to obtain the average pixel value of the corresponding part in the actual image; The standard deviation of pixel values ​​of all pixels in any part is calculated by the standard deviation formula. The first value of the interval is obtained by subtracting the standard deviation of pixel values ​​from the average value of pixels. The second value of the interval is obtained by summing the average value of pixels and the standard deviation of pixel values. The pixel reference intervals for different parts are constructed with the first value of the interval as the left endpoint and the second value of the interval as the right endpoint. Similarly, multiple actual images of the corresponding bird are selected as the analysis object. The pixel reference intervals of different parts in all actual images are analyzed and obtained. The different pixel reference intervals are associated with different parts to obtain the pixel interval sequence of different parts of the corresponding bird.

4. The intelligent bird identification and monitoring system based on image recognition and cloud computing according to claim 1, characterized in that, The prediction process of the type prediction unit includes the following: Take pictures of any known bird species to obtain real-time images of the corresponding birds. Input the real-time images of the corresponding birds into the target detection model to detect different parts of the corresponding birds in the real-time images. Select any part of the real-time image as the analysis object and record it as the target part. Obtain the pixel values ​​of different pixels in the target part, and sum the pixel values ​​and take the average value to obtain the average pixel value of the target part in the real-time image. Obtain the pixel reference interval within all pixel interval sequences corresponding to the target area, and compare the average pixel value of the target area in the real-time image with the pixel reference interval.

5. The intelligent bird identification and monitoring system based on image recognition and cloud computing according to claim 4, characterized in that, The prediction process of the type prediction unit also includes the following: If the average pixel value of the target area in the real-time image falls within any pixel reference interval, then the bird species corresponding to the pixel reference interval is recorded as a predicted species. If the average pixel value of the target area in the real-time image does not belong to any pixel reference interval, the absolute value of the pixel value minus the midpoint value of all pixel reference intervals is obtained to get the pixel deviation value of the target area in the real-time image. The minimum pixel deviation value is obtained by iterating and comparing different pixel deviation values, and the bird species corresponding to the undetermined reference interval of the minimum pixel deviation value is recorded as a predicted species. Similarly, any part of the real-time image is selected as the target part for the same analysis, thereby identifying the corresponding prediction category based on different target parts.

6. The intelligent bird identification and monitoring system based on image recognition and cloud computing according to claim 5, characterized in that, The prediction process of the type prediction unit also includes the following: Obtain the real-time shooting focal length and real-time shooting object distance of the real-time image, count the number of pixels in the target area in the real-time image, and multiply the real-time shooting object distance by the number of pixels in the target area and divide by the real-time shooting focal length to obtain the real-time area of ​​the target area in the real-time image. Obtain the actual area from all the actual area arrays corresponding to the target part, and take the absolute value of the real-time area deviation of the target part by subtracting the different actual areas from the real-time area. Iterate and compare the different real-time area deviation values ​​to obtain the minimum real-time area deviation of the target part. The bird species corresponding to the minimum real-time area deviation is recorded as the secondary predicted species obtained based on the target location identification. Similarly, any part of the real-time image is selected as the target part for the same analysis, thereby identifying the corresponding secondary prediction category based on different target parts.

7. The intelligent bird identification and monitoring system based on image recognition and cloud computing according to claim 1, characterized in that, The analysis process of the elemental analysis unit includes the following: The predicted types and secondary predicted types obtained based on the same target location are compared with the actual types. If the predicted category obtained based on the same target location includes the actual category, but the predicted category does not include the actual category, then the accuracy of the first prediction for the corresponding target location is incremented by one. If the predicted category obtained based on the same target location does not contain the actual category, but the predicted category contains the actual category, then the accuracy of the second prediction for the corresponding target location is incremented by one. If both the primary and secondary predictions of the same target location contain the actual target type, then the accuracy of the primary and secondary predictions for the corresponding target location will be incremented by one. If neither the first prediction category nor the second prediction category contains the actual category, no action will be taken. Similarly, the same known bird species is photographed again to obtain multiple real-time images of the corresponding bird. The target parts in different real-time images are analyzed repeatedly to obtain the accuracy of the first prediction and the accuracy of the second prediction for different target parts.

8. The intelligent bird identification and monitoring system based on image recognition and cloud computing according to claim 7, characterized in that, The analysis process of the elemental analysis unit also includes the following: The number of real-time images captured is counted, and the accuracy of the first prediction is divided by the number of images captured to obtain the accuracy of the first prediction for different target parts. The accuracy of the second prediction is divided by the number of images captured to obtain the accuracy of the second prediction for different target parts. Compare the accuracy of the first prediction with the accuracy of the second prediction for the same target location; If the accuracy of a single prediction for the same target part is greater than or equal to the accuracy of a double prediction, then the average pixel value of the corresponding target part will be used as the recognition element. If the accuracy of a single prediction for the same target part is less than the accuracy of a double prediction, then the real-time area of ​​the corresponding target part will be used as the identification element. Similarly, the accuracy rates of the first and second predictions for different target parts are compared to obtain the identification elements for different target parts.

9. The intelligent bird identification and monitoring system based on image recognition and cloud computing according to claim 1, characterized in that, The recognition process of the image recognition module includes the following: The image to be identified is input into the target detection model, which detects different parts of the bird in the image. Obtain the recognition elements corresponding to different parts. When the recognition element is the pixel average value, calculate the pixel average value of the corresponding part and analyze it to obtain a prediction category. When the identified element is a real-time area, the real-time area of ​​the corresponding part is calculated and analyzed to obtain the secondary prediction type; Similarly, based on the analysis of different parts, the primary and secondary predicted species of birds in the image to be identified are obtained; If there are duplicate predicted types in the first and second predictions, the corresponding predicted type is recorded as the undetermined type of bird in the image to be identified. At the same time, the number of times different undetermined types appear in the first and second predictions is counted and recorded as the prediction number. The different prediction numbers are compared to obtain the maximum prediction number, and the undetermined type corresponding to the maximum prediction number is recorded as the predicted type of bird in the image to be identified. If there are no duplicate predicted species in the primary and secondary predictions, then the primary and secondary predictions are collectively recorded as the predicted species of birds in the image to be identified.

10. A bird intelligent identification and monitoring method based on image recognition and cloud computing, characterized in that, The bird intelligent identification and monitoring system based on image recognition and cloud computing, as described in any one of claims 1-9, comprises the following steps: Step S1: Extract the biological body shape features of different parts of the bird based on the actual image of the bird, and obtain the actual area array of different parts of the bird. Step S2: Based on the actual image of the bird, extract the biological color features of different parts of the bird to obtain the pixel interval sequence of different parts of the bird. Step S3: Based on the actual area array and pixel interval sequence of different parts of the bird, predict the known actual bird species and obtain the first predicted species and the second predicted species. Step S4: Analyze the prediction accuracy of different target parts based on the primary prediction type or the secondary prediction type, and obtain the identification elements of different target parts. Step S5: Based on the identification elements, perform image recognition on the birds in the image to be identified, and obtain the predicted species of birds in the image to be identified.

Images