Control method of bird feeder interaction system, bird feeder and storage medium

Abstract

The invention discloses a control method of a bird feeder interaction system, a bird feeder and a storage medium, and relates to the technical field of computers.The method comprises the steps that after bird activities in a trough area are detected, relevant data are collected according to preset time nodes, and target image data and associated information are obtained through preprocessing; image features of the target image data are extracted and compared with a bird feature library for recognition, and a target recognition result is generated; according to the associated information and the target recognition result, bird images in the target image data are marked in a multi-dimensional mode, and mark information is generated; and after it is determined that the user pays attention to the birds through the mark information, interaction information corresponding to the user pays attention to the birds is issued to the terminal equipment. According to the method, the preprocessing data is collected after the activity of the birds in the trough is detected, the problem of poor data collection quality is solved through feature comparison and recognition, marking and issuing of concerned bird interaction information and synchronization of user marking feedback and model iteration, and the real-time performance of bird recognition and information pushing is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the computer technical field, and particularly relates to a control method of a bird feeder interactive system, a bird feeder and a storage medium. BACKGROUND

[0002] In an outdoor ecological feeding scene, bird image collection is a key link to meet the ecological observation demand. In the related technology, the bird feeder adopts a mode of indiscriminate image collection and feeding, which runs through the whole process of ecological observation data acquisition. However, such indiscriminate operation is prone to non-target biological interference and low proportion of effective bird images, thereby resulting in poor data collection quality.

[0003] The above content is only used to assist in understanding the technical solutions of the present application and does not represent the acknowledgement of the above content as prior art. SUMMARY

[0004] The main purpose of the present application is to provide a control method of a bird feeder interactive system, a bird feeder and a storage medium, aiming to solve the technical problem of poor data collection quality.

[0005] To achieve the above purpose, the present application provides a control method of a bird feeder interactive system, which comprises the following steps: After detecting bird activity in the feeding trough area, collecting relevant data according to a preset time node, and preprocessing to obtain target image data and associated information; Extracting the image features of the target image data, comparing and identifying with a bird feature library, and generating a target identification result; According to the associated information and the target identification result, marking the bird images in the target image data in multiple dimensions to generate marking information; After determining the user's attention bird through the marking information, the interactive information corresponding to the user's attention bird is sent to the terminal device.

[0006] In an embodiment, after receiving a bird feeder interactive initialization completion signal, a collection instruction is generated by detecting bird activity in the feeding trough area through an infrared detector, and a collection time node is determined according to a preset collection rule to obtain the collection instruction and the corresponding preset time node; Based on the collection instruction and the corresponding preset time node, the camera is controlled to shoot bird static images and short videos according to the preset time node, and the associated information is recorded synchronously, and the original image data and the associated information are output; The original image data is preprocessed to obtain optimized target image data, and the corresponding associated information is associated to obtain the target image data and the corresponding associated information.

[0007] In an embodiment, a lightweight bird identification model is loaded to extract features of a static bird image in the target image data, obtaining the image features; The image features are compared and analyzed with the bird feature library to output an initial result containing bird species and identification confidence; According to the pre-set confidence judgment rule, the validity of the initial result is judged, and the target identification result is generated in combination with the high-precision bird identification model.

[0008] In an embodiment, according to the pre-set confidence judgment rule, the relationship between the identification confidence in the initial result and the preset threshold is compared to obtain a comparison result; If the comparison result is that the identification confidence is greater than or equal to the preset threshold, the corresponding initial result is determined as the target identification result; If the comparison result is that the identification confidence is less than the preset threshold, the static bird image in the target image data and the initial result are packaged into secondary identification data to generate secondary identification data.

[0009] In an embodiment, the secondary identification data is uploaded to a cloud server to output a confirmation signal of successful reception of the cloud server; According to the confirmation signal of successful reception of the cloud server, the high-precision bird identification model is called to deeply analyze and identify the secondary identification data, and a precise identification result is output; The precise identification result is received, the identification source of the precise identification result is supplemented as cloud secondary identification, and the target identification result is integrated and generated.

[0010] In an embodiment, based on the target identification result and the associated information, a marked object is located, and a bird image to be marked and its corresponding associated information and target identification result are output; The bird image to be marked and its corresponding associated information and target identification result are converted into basic marking information and identification marking information, and intermediate marking data is output; After integrating and verifying the two types of marking information in the intermediate marking data, the corresponding relationship between the marking content and the bird image is determined, and the marking information is generated.

[0011] In an embodiment, the target bird identified in the marking information is compared with a list of birds of interest to determine whether the target bird is a bird of interest of a user, and a bird species comparison result is generated; If the comparison result is that the target bird is the bird of interest of the user, the relevant information of the target bird is integrated, and the interaction information corresponding to the target bird is output; The interaction information is sent to the terminal device through the communication module and displayed, and a confirmation signal of successful output of the interaction information is sent.

[0012] In an embodiment, manual annotation data of the user on the unidentified bird is collected, and feedback data is generated by integration; The feedback data is transmitted to a cloud server, and a high-precision bird identification model is retrained in combination with a new bird sample to generate an optimized high-precision bird identification model; The optimized high-precision bird identification model is processed to generate an update package adapted to the edge computing module of the bird feeder, so as to complete the update of the lightweight bird identification model.

[0013] In addition, to achieve the above-mentioned purpose, the application also provides a bird feeder, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the computer program is configured to implement the steps of the control method of the bird feeder interaction system as described above.

[0014] In addition, to achieve the above-mentioned purpose, the application also provides a storage medium, which is a computer-readable storage medium, and the storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the control method of the bird feeder interaction system as described above.

[0015] The application provides a control method of a bird feeder interaction system, which comprises collecting relevant data and preprocessing target image data and associated information according to a preset time node after detecting bird activity in a feeding trough area, extracting image features of the target image data and comparing and identifying bird features in a library to generate a target identification result, marking a bird image in the target image data in multiple dimensions in combination with the associated information and the target identification result to generate marking information, and issuing corresponding interaction information to a terminal device after determining that a user is interested in a bird, thereby solving the technical problems of a conventional bird feeder that cannot accurately identify visiting birds, cannot structurally manage bird visiting data, a user cannot obtain information about a bird of interest in a timely manner, and an identification model lacks an iterative optimization mechanism, improving the accuracy of bird identification and the coverage range of bird species, realizing standardized storage of bird visiting data, and enhancing user interaction experience.

[0016] In summary, the application collects and preprocesses data after detecting bird activity in a feeding trough, compares and identifies features, issues interaction information about a bird of interest, synchronizes user feedback and model iteration, solves the technical problem of poor data collection quality, improves the real-time performance of bird identification and information pushing, optimizes data processing standardization, and realizes dual improvement of process efficiency and user experience. BRIEF DESCRIPTION OF DRAWINGS

[0017] The accompanying drawings, which are incorporated into and form a part of the specification, illustrate preferred embodiments of the present application and, together with the description, serve to explain the principles of the application.

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings required by the embodiments or the prior art description will be briefly introduced. Obviously, for those skilled in the art, other drawings can also be obtained based on these drawings without creative labor.

[0019] Figure 1 Flowchart of a first embodiment of a control method of a bird feeder interactive system of the present application; Figure 2 Flowchart of a fourth embodiment of a control method of a bird feeder interactive system of the present application; Figure 3 Flowchart of a fifth embodiment of a control method of a bird feeder interactive system of the present application; Figure 4 Flowchart of an eighth embodiment of a control method of a bird feeder interactive system of the present application; Figure 5 Structure diagram of a bird feeder of the present application.

[0020] The object implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0021] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and do not limit the present application.

[0022] In the related art, the bird feeder adopts a mode of indiscriminate image collection and feeding of feed, and runs through the whole process of ecological observation data acquisition, but such indiscriminate operation is prone to non-target biological interference and low proportion of effective bird images, thereby resulting in poor data collection quality.

[0023] The present application provides a solution: first, after detecting bird activity in the feeding trough area, collecting relevant data according to a preset time node, preprocessing to obtain target image data and associated information, then extracting image features of the target image data, comparing and identifying with a bird feature library to generate a target identification result, then according to the associated information and the target identification result, multi-dimensionally marking bird images in the target image data to generate marking information, and finally determining a user's attention bird through the marking information, and issuing interactive information corresponding to the user's attention bird to a terminal device.

[0024] It should be noted that the execution subject of the embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a bird feeder, etc. The following takes the bird feeder as an example to describe the embodiment and each of the following embodiments.

[0025] In order to better understand the technical solutions of the present application, the following will be described in detail in combination with the drawings of the specification and specific embodiments.

[0026] The embodiment of the present application provides a control method of a bird feeder interaction system, referring to Figure 1 , Figure 1 The flowchart of the first embodiment of the control method of the bird feeder interaction system of the present application.

[0027] In the embodiment, the control method of the bird feeder interaction system comprises steps S10-S40: Step S10, after detecting the bird activity in the feeding trough area, collecting relevant data according to the preset time node, and preprocessing to obtain target image data and associated information.

[0028] In the embodiment, the feeding trough area refers to a specific area in the bird feeder for birds to forage. The preset time node refers to the trigger time set in advance. The preprocessing refers to the noise reduction, cropping, format conversion and other processing of the original data. The target image data refers to the image and video data that can be used for identification after preprocessing, and the associated information refers to the supporting information such as collection time, place, environmental data, etc.

[0029] As an optional implementation, after detecting the bird activity signal in the feeding trough area, immediately respond to the preset time node. Start the collection operation at a set interval at the activity trigger moment and during the activity, capture the original image and video data related to the birds, and record the associated information such as collection time, collection place, environmental temperature and humidity, etc. Then, perform noise reduction, cropping, format conversion and compression processing on the collected original data simultaneously, remove redundant interference information, integrate the processed image and video into target image data, and bind the associated information one by one with the target image data. This method has strong data collection timeliness and can completely capture the bird activity process, providing high-quality data support for subsequent rapid identification.

[0030] As another optional implementation, after detecting the bird activity in the feeding trough area, the activity start signal is recorded first. After the activity ends, the collection operation is started at the preset time node, and all original images and video data during the activity are obtained at one time. Then the original data is preprocessed in stages, and the invalid data with blur and no bird image is filtered out first. Then the effective data is sequentially subjected to noise reduction, format conversion and compression processing to obtain target image data. Finally, the related information such as collection time, place and environment data is uniformly arranged, and the corresponding relationship between the data collection sequence and the target image data is established. This method has low resource consumption, the data processing is more targeted, and the workload of subsequent processing is reduced.

[0031] Step S20, extracting the image features of the target image data, comparing and identifying with the bird feature library to generate a target identification result.

[0032] In this embodiment, the image features refer to the key identification information of the bird shape, feather texture and contour extracted from the image. The bird feature library refers to a collection of stored corresponding feature information of known birds. The target identification result refers to the final identification conclusion including the bird species, identification confidence and identification source.

[0033] As an optional implementation, the core features of the static images in the target image data are extracted, including the contour, key feather texture and shape features, and are compared with the local stored common bird feature library one by one to calculate the feature coincidence degree and convert it into the identification confidence. If the coincidence degree reaches the preset standard, the bird species is directly determined, and if it does not reach the preset standard, it is marked as a to-be-confirmed category. The target identification result is generated by combining the bird species and the corresponding confidence. This method has fast comparison speed, does not require additional transmission resources, and is suitable for scenes with high identification speed requirements.

[0034] As another optional implementation, the full-quantity features of the static images and dynamic video key frames in the target image data are extracted, covering the detail texture and behavior posture related features. First, the local feature library is preliminarily compared to filter out high-similarity candidate categories. Then the full-quantity features and candidate category information are uploaded to the cloud complete bird feature library for multi-dimensional cross comparison and deep analysis. The comprehensive confidence is calculated based on the comparison results of the local and cloud, the identification source is determined, and the target identification result including the bird species, confidence and identification source is generated. This method has comprehensive identification species coverage and high accuracy, and can identify rare birds.

[0035] Step S30, according to the related information and the target identification result, multi-dimensionally marking the bird images in the target image data to generate marking information.

[0036] In this embodiment, multi-dimensional marking refers to marking from multiple dimensions such as basic information and identification information. Bird images in target image data refer to core image data that can be used for marking after preprocessing. Marking information refers to a structured information set formed after integrating multi-dimensional marking content.

[0037] As an optional implementation, after receiving the association information and the target identification result, a multi-dimensional marking process is immediately started for the bird images in the target image data. The collection time, collection location, environmental data and other association information are directly marked to the image corresponding field. Then the bird species, identification confidence, identification source and other identification result information are entered, and a unique corresponding identifier of the information and the image is established synchronously. After marking, complete structured marking information is immediately integrated. This method has fast marking response speed, high information and image binding accuracy, and avoids information misplacement or delay.

[0038] As another optional implementation, a certain amount of association information, target identification result and corresponding bird images are first collected, classified and summarized according to the identification source or bird species. Then the association information and identification result in each category are extracted in batches, and the bird images of the same category are uniformly marked with basic information. Each image is supplemented with exclusive identification information. Finally, the association between multi-dimensional information and images is established through classification numbering, and hierarchical structured marking information is integrated. This method has dispersed resource consumption, high batch processing efficiency, and is suitable for large data centralized processing scenarios, and improves the standardization of overall data management.

[0039] In step S40, after determining the user's attention bird through the marking information, the interaction information corresponding to the user's attention bird is pushed to the terminal device.

[0040] In this embodiment, the user's attention bird refers to a specific bird species that the user has previously set to focus on. The interaction information refers to an association information set containing clear images after preprocessing, structured marking information and reminder identifier. The terminal device refers to a terminal carrier used by the user to receive and display information.

[0041] As an optional implementation, the bird species information in the marking information is analyzed, and the bird species information is compared with the preset user attention bird list in real time. If the bird species is confirmed to be in the user's attention category, the corresponding target image data, complete marking information and preset reminder identifier are immediately extracted, and standardized interaction information is quickly integrated. The information is directly pushed to the terminal device through the communication link, ensuring the immediacy and integrity of information transmission. This method has fast response speed and can reach the user in the first time, meeting the user's demand for instant knowledge of the attention bird.

[0042] As another optional implementation, all marking information within a set period is collected first, and is classified and summarized according to bird species. Then, the classified marking information is compared with the user's list of birds of interest in batches, and all marking information and corresponding target image data that meet the conditions of interest are screened out, and are integrated to form summary interactive information containing multiple pieces of information about birds of interest, and are sent to the terminal device through the communication link after being indexed. This method reduces the transmission frequency and resource consumption by batch processing, adapts to the user's demand for batch information summary, and improves the information management efficiency.

[0043] Exemplarily, in the scenario of interaction between the bird feeder and the terminal, when the infrared detector detects bird activity in the feeding trough area, the acquisition is started at a preset time node (0.5 seconds after the activity is triggered), 1 1920x1080 pixel static image and 10 seconds of 1080P short video are acquired, and the associated information (acquisition time 2024-06-15 09:42:30, location N 30.21, E 120.35, environment temperature 26 degrees Celsius, humidity 58%) is recorded synchronously. After preprocessing such as Gaussian filter denoising, cropping to 800x600 pixels, and format conversion to JPEG format, the target image data is obtained. The HOG feature extraction algorithm is used to extract the features of the static image, and the extracted features are compared with the locally mounted ResNet18 lightweight bird feature library to generate a target recognition result (the bird species is white-headed bulbul, the recognition confidence is 94%, and the recognition source is local). Based on the associated information and the target recognition result, the white-headed bulbul image in the target image data is marked in multiple dimensions. The basic information includes the acquisition time, location, and environmental data, the recognition information includes the white-headed bulbul, 94% confidence, and local recognition, and the structured marking information is generated. After comparing with the user's preset list of birds of interest (containing white-headed bulbul), the interactive information (800x600 clear static image, 10 seconds of short video, complete marking information, and APP pop-up reminder “Your interested white-headed bulbul has arrived, click to view details”) is integrated, and is sent to the user's mobile terminal through the communication module. The terminal APP receives and displays the information in real time within 1.5 seconds.

[0044] Since the preprocessed data is collected after detecting bird activity in the feeding trough, the interactive information about birds of interest is recognized, marked, and sent after feature comparison, the user's feedback and model iteration are synchronized, the problem of poor data acquisition quality is solved, the real-time performance of bird recognition and information pushing is improved, the data processing standardization is optimized, and the process efficiency and user experience are improved.

[0045] Based on any of the above embodiments, in the second embodiment of the present application, the step S10 includes steps A11-A13: Step A11, after receiving the bird feeder interaction initialization completion signal, the bird activity in the feeding trough area is detected by the infrared detector to generate a collection instruction, and the collection time node is determined according to the preset collection rule to obtain the collection instruction and the corresponding preset time node.

[0046] In this embodiment, the bird feeder interaction initialization completion signal refers to a ready signal sent by the bird feeder after completing the preparation work such as starting and parameter configuration. The infrared detector refers to a detection component for sensing bird activity in the feeding trough area. The collection instruction refers to an instruction signal for triggering data collection operation. The preset collection rule refers to a rule set in advance for determining the collection opportunity.

[0047] As an optional implementation, after receiving the bird feeder interaction initialization completion signal, the feeding trough area is continuously monitored by the infrared detector, and as soon as bird activity is sensed, a collection instruction is immediately generated. At the same time, according to the preset collection rule, the key collection time nodes are dynamically adjusted in combination with the real-time conditions such as the duration of bird activity and the action amplitude. The important stages such as the initial activity and the foraging peak are preferentially covered, and finally the complete information containing the trigger signal and the dynamically adapted collection time nodes is obtained. This method is highly consistent with the actual activity of the bird in the collection opportunity, and the key action is captured comprehensively, avoiding missing important activity segments.

[0048] As another optional implementation, after receiving the bird feeder interaction initialization completion signal, the infrared detector enters a fixed period monitoring mode, and when bird activity is detected in the feeding trough area, a collection instruction is generated, and then the fixed time interval and the preset conditions of the fixed collection number of times in the preset collection rule are determined to determine the fixed and unchanged collection time nodes. The final result is obtained by integrating the basic collection instruction and the fixed collection time node. This method has simple rule execution, fast instruction generation speed, and low resource consumption.

[0049] Step A12, based on the collection instruction and the corresponding preset time node, the camera is controlled to shoot the bird static image and short video according to the preset time node, and the associated information is recorded synchronously, and the original image data and the associated information are output.

[0050] In this embodiment, the bird static image refers to a single-frame captured bird picture. The short video refers to a continuous shot bird dynamic image. The associated information refers to the supporting information recorded synchronously during shooting. The original image data refers to the unprocessed static image and short video obtained by shooting.

[0051] As an optional implementation, based on the collection instruction and the corresponding preset time node, the shooting operation is started in sequence according to the preset time node, and the bird static image and the short video are captured at the same time when the preset time node triggers each time. The collection time, the collection location, and the associated information of the environment related data corresponding to the preset time node are recorded in real time during the shooting process. After the shooting and information recording of each time node are completed, the static image, the short video, and the associated information of the node are temporarily bound, and after all the preset time nodes are executed, all the bound contents are integrated to output the complete original image data and the corresponding associated information. The image data and the associated information of this method have strong synchronization and accurate and unbiased corresponding relationship.

[0052] As another optional implementation, based on the collection instruction and the corresponding preset time node, the shooting of all bird static images is completed in sequence according to the node order. Then, the short videos are recorded in sequence according to the same time node, and the associated information key items of each node are recorded and temporarily stored during the shooting. After the shooting of all the static images and the short videos is completed, the associated information key items are supplemented, the corresponding relationship between the shooting node order and the corresponding static image and the short video is established one by one, and after the complete data set is integrated, the associated information is arranged in a standard and orderly manner, which is suitable for the multi-node dense collection scene and reduces the resource conflict risk in the shooting process.

[0053] Step A13, preprocessing the original image data to obtain optimized target image data, and associating the corresponding associated information to obtain the target image data and the corresponding associated information.

[0054] As an optional implementation, the original image data and the corresponding associated information are received, and the preprocessing is started in sections according to the shooting order of the image data. After the noise reduction, the cropping, and the format adjustment of each section of the original image are completed, the associated information corresponding to the section of the image is immediately called to establish the binding relationship between the two through the unique collection identifier. The process is continuously repeated until all the original images are preprocessed and associated, and finally a complete data set of each section of the target image data and the corresponding associated information is output. The preprocessing and the association of this method are synchronously promoted, the data corresponding relationship is unbiased, and there is no need for subsequent verification.

[0055] As another optional implementation, all the original image data is collected for batch preprocessing, and the noise reduction, the cropping, and the format adjustment operation are uniformly performed on all the images to form a complete target image data set. Then, all the associated information is sorted according to the collection time sequence, and the sorted associated information is matched and associated with the target image data set one by one through the sequence identifier generated during the collection to establish a global corresponding relationship, and then a complete data set is integrated and output. This method has high batch processing efficiency, simple process, and concentrated resource occupation, and is suitable for processing scenes of a large amount of original image data.

[0056] Illustratively, in the scenario of bird feeder interacting with the terminal, after receiving the bird feeder interaction initialization completion signal, the HC-SR501 infrared detector monitors the feeding trough area within 3 seconds, generates a collection instruction 0.2 seconds after detecting bird foraging activity, determines 0.5 seconds, 1 second, and 2 seconds as three preset time nodes according to the preset collection rules, and obtains the collection instruction and corresponding node information. Based on the collection instruction and time nodes, the OV5640 camera sequentially captures 1920x1080 pixel static images at each node, synchronously records a 10-second short video with a frame rate of 30 frames, and simultaneously records related information such as collection time (2024-08-20 10:15:23), collection location (31.5 degrees north latitude and 120.8 degrees east longitude), and environment temperature 28 degrees Celsius and humidity 62%. The output is 3 pieces of original image data composed of static images and 3 pieces of short video and complete related information. The original image data is preprocessed by Gaussian noise reduction, cropping to 800x600 pixels, and format conversion to JPEG to obtain optimized target image data. The target image data and corresponding related information are bound one by one through the collection timestamp, and finally 3 sets of precisely associated target image data and corresponding related information are obtained.

[0057] Due to the rapid trigger collection by the infrared detector, the multi-node precise shooting by the camera, and the timestamp binding of the related information, the problems of high collection delay and mispositioning of related information in traditional bird feeders are solved, and the collection response speed, target image data clarity, and information association accuracy are improved.

[0058] Based on any of the above embodiments, in Embodiment Three of the present application, the step S20 includes steps B11-B13: Step B11, load the lightweight bird recognition model, extract the features of the static bird images in the target image data, and obtain the image features.

[0059] In this embodiment, the lightweight bird recognition model refers to a bird recognition related model with compact size, low resource consumption, and adaptation to fast feature extraction.

[0060] As an optional implementation, after loading the lightweight bird recognition model completely and completing the initialization, all static bird images in the target image data are batched and retrieved, and the global core features of each image are extracted synchronously, covering key information such as overall contour, feather texture distribution, and limb morphology. The feature dimensions are kept uniform during the extraction process, and the extracted feature information is preliminarily regularized to remove redundant content, generating a standardized global image feature set. This method has high batch processing efficiency, and the uniform feature dimensions facilitate subsequent comparison operations, making it suitable for fast recognition scenarios of common birds.

[0061] As another optional implementation, after loading the lightweight bird recognition model, parameter calibration is performed first, and then key image frames with clear bird posture and no occlusion are screened from the static bird images of the target image data. For the screened image frames, global contour features and local detail features, including feather texture details, beak morphology, and eye features, are extracted in steps. Then, the two types of features are integrated to supplement feature correlation information and improve recognition, and finally a multi-dimensional image feature set containing global and local information is obtained. This method has more comprehensive feature information, taking into account both global and local features.

[0062] Step B12, comparing and analyzing the image features with the bird feature library, outputting an initial result containing bird species and recognition confidence.

[0063] In this embodiment, the initial result refers to the preliminary identification conclusion containing the suspected bird species and the reliability of the conclusion generated after comparison.

[0064] As an optional implementation, the core dimensions of the extracted image features are first screened, and the key feature items with the highest recognition correlation are retained. Then, a quick rough screening comparison is performed between the key features and the core features of all birds in the bird feature library, and the candidate categories with high similarity are screened out. Then, the feature details of the candidate categories are matched one by one, the overall similarity is calculated and converted into recognition confidence, the bird species is determined in combination with the candidate category with the highest similarity, and finally the initial result containing the bird species and the corresponding recognition confidence is obtained. This method has fast comparison speed, can quickly narrow down the candidate range, and has low resource consumption.

[0065] As another optional implementation, the extracted complete image features are compared with the features of each bird species in the bird feature library in full dimension one by one, the global feature similarity and the local detail feature similarity are calculated respectively, and the two types of similarity are fused according to the preset weight to obtain the comprehensive similarity. The comprehensive similarity is converted into recognition confidence, the bird species with the highest comprehensive similarity is selected as the recognition result, and the matching situation of each dimension similarity is recorded to generate the initial result containing the bird species, the recognition confidence, and the key matching dimension description. This method has comprehensive comparison dimensions, can accurately capture feature differences, and has high recognition accuracy.

[0066] Step B13, judging the validity of the initial result according to the preset confidence judgment rule, and generating the target recognition result in combination with the high-precision bird recognition model.

[0067] In this embodiment, the preset confidence judgment rule refers to the judgment standard set in advance for evaluating the reliability of the initial result. Validity refers to whether the reliability of the initial result meets the preset requirements. The high-precision bird recognition model refers to an identification model with higher recognition accuracy and covering more bird species.

[0068] As an optional implementation, after receiving the initial result containing bird species and recognition confidence level, the recognition confidence level in the initial result is directly compared with a preset threshold according to a pre-set confidence judgment rule to determine the validity level of the initial result. If the recognition confidence level reaches or exceeds the preset threshold, the initial result is deemed valid, and the local preliminary recognition source annotation is directly added to it to form a complete target recognition result. If the recognition confidence level does not reach the preset threshold, the initial result is deemed invalid, and a static bird image from the corresponding target image data is immediately retrieved. This static bird image is input into a high-precision bird recognition model, and the model performs deep feature extraction and multi-dimensional comparison analysis to generate a secondary recognition result containing accurate bird species and high-confidence recognition confidence. Subsequently, the comparison record of the initial result and the core information of the secondary recognition result are integrated, and the recognition source annotation of the high-precision verification in the cloud is added to generate the target recognition result. This method specifically handles initial results with different levels of validity, and valid results can be directly reused, significantly saving computing resources and time, and providing a clear basis for subsequent data traceability.

[0069] As an alternative implementation, after receiving the initial result, the system first comprehensively analyzes the recognition confidence distribution and feature matching details of the initial result according to pre-set confidence judgment rules to determine its validity. During the judgment process, static bird images from the corresponding target image data are simultaneously retrieved and input into the high-precision bird recognition model along with the initial result. The model first reproduces the comparison logic of the initial result. Then, independent deep feature mining and full-category comparison are performed to generate a high-precision verification result. If the initial result is valid, the bird species and confidence levels of the two are cross-validated, and the conclusion with higher confidence is taken as the core. The feature matching advantages of the two are then combined to generate the target recognition result. If the initial result is invalid, the high-precision verification result is used as the main body, supplementing the feature comparison differences and invalidation reasons of the initial result to generate the target recognition result. At the same time, the complete process data of the two recognitions are recorded. This method, through dual recognition and cross-validation, minimizes the risk of misjudgment from single recognition and achieves extremely high recognition accuracy.

[0070] Exemplarily, in the scenario of the bird feeder interacting with the terminal, the MobileNetV3 lightweight bird recognition model is loaded and initialized, a 800x600 pixel static bird image is called from the target image data, the 2048-dimensional core features of the image are extracted, including contour, feather texture, limb morphology and other key information, and the standardized image features are obtained. The image features are quickly compared and analyzed with the feature library containing 500 common bird species stored locally, the feature coincidence degree is calculated and converted into recognition confidence, and the initial result containing the bird species as sparrow and recognition confidence of 82% is output. According to the pre-set confidence judgment rule (pre-set threshold 95%), the 82% confidence of the initial result is compared with the threshold, it is determined that the initial result is invalid, then the corresponding static bird image and the feature comparison record of the initial result are called, the EfficientNetB4 high-precision bird recognition model is input, the deep feature mining and multi-dimensional comparison of all categories are performed through the model, the high-precision review result of the bird species as sparrow and the recognition confidence of 96% is generated, the initial result invalid reason and the review result are integrated, the label of the recognition source "cloud high-precision review" is supplemented, and finally the complete target recognition result is generated.

[0071] Further, 100,000 bird image samples (covering 800 target bird species, including sparrows, acorn woodpeckers, red-billed blue magpies, white wagtails, etc. commonly seen in bird feeders) are collected, uniformly adjusted to 800x600 pixel JPEG format, and a standardized training data set without blur and occlusion is constructed. Each image in the data set is accurately labeled with 9 key feature point information, including bird species, head contour coordinates, body boundary, left and right wing endpoints, left and right leg length, tail shape, and left and right claw position, with an accuracy of 99.5%. The data set is randomly allocated in a ratio of 8:1:1 to obtain 80,000 training sets, 10,000 validation sets, and 10,000 test sets. The EfficientNetB4 feature extraction network is used to train the bird recognition model using the training set, the network weight parameters are adjusted in real time through the validation set, and the test set verifies the model performance. Finally, the bird recognition model is trained, the feature extraction network and the trained weight parameters constitute the bird feature library, and the initial recognition accuracy reaches 92%. In actual use, the bird feeder accumulates 1200 incorrect recognition images (such as misjudging a white wagtail as a yellow wagtail and a red-throated song thrush as a dark green painted eye bird), re-labels the species and feature points of these images, supplements them to the training set and trains the model again, updates the weight parameters of the feature library, and completes the iterative optimization of the feature library.

[0072] Due to the rapid feature extraction by the lightweight model, the preliminary comparison of the local feature library, and the deep review of the high-precision model, the problem of insufficient confidence caused by relying on a single model in traditional recognition and the misjudgment of similar bird species is solved, the accuracy of bird recognition is improved, and the authenticity of obtaining the information of the bird of interest is ensured.

[0073] Based on any of the above embodiments, in Embodiment Four of the control method of the bird feeder interactive system of the present application, refer to Figure 2 , Figure 2 is a flowchart of the fourth embodiment of the control method of the bird feeder interactive system of the present application. The step B13 includes steps C11-C13: Step C11, according to the pre-set confidence judgment rule, compare the recognition confidence in the initial result with the relationship of the pre-set threshold value, and obtain the comparison result.

[0074] In this embodiment, the recognition confidence refers to a quantitative indicator that measures the reliability of the bird species determination in the initial result. The pre-set threshold value refers to a benchmark value for determining whether the recognition confidence meets the effective standard. The comparison result refers to the determination conclusion of high or equal after comparing the recognition confidence with the pre-set threshold value.

[0075] As an optional implementation, first, completely analyze the pre-set confidence judgment rule to determine the fixed value of the pre-set threshold value in the pre-set confidence judgment rule and the single dimension comparison logic. Then, accurately extract the specific value of the recognition confidence from the initial result, and perform format verification on the extracted confidence value to ensure that the confidence value meets the numerical specification required by the comparison rule. Subsequently, according to the comparison logic in the pre-set confidence judgment rule, directly compare the verified recognition confidence value with the pre-set threshold value in terms of numerical size, record the key data in the comparison process in real time, including the confidence value, the pre-set threshold value, the numerical difference, and finally generate a comparison result that clearly labels the recognition confidence as higher than, equal to, or lower than the pre-set threshold value, while also including a brief record of the comparison process. This method has a simple and intuitive comparison process, fewer operation steps, and can quickly obtain the comparison result, making it suitable for scenarios with high processing efficiency requirements.

[0076] As another optional implementation, first, deeply disassemble the pre-set confidence judgment rule to determine the multiple threshold value intervals and multi-dimensional comparison logic based on the matching integrity of bird characteristics in the rule. Then, extract the recognition confidence value and corresponding feature matching detail data from the initial result, including the global feature matching degree and the local feature matching degree, determine the currently applicable threshold value interval based on the feature matching detail data, and retrieve the pre-set threshold value range corresponding to the interval. Subsequently, perform multi-dimensional comparison of the recognition confidence value and the adapted threshold value range, not only judging whether the value is within the interval, but also analyzing whether the feature matching degree and the confidence value meet the rule requirements, and recording the comparison situation of each dimension and the threshold interval selection basis in detail. Finally, generate a comprehensive comparison result containing the threshold interval of the recognition confidence, the feature matching relevance determination, and the final high-low conclusion. This method dynamically adapts the threshold value based on feature matching details, and multi-dimensional comparison is more accurate, which can effectively avoid the misjudgment of edge values.

[0077] Step C12, if the comparison result is that the recognition confidence is greater than or equal to the preset threshold, the corresponding initial result is determined as the target recognition result.

[0078] As an optional implementation, after receiving the comparison result and the corresponding initial result, firstly, the association consistency of the two is verified, and it is confirmed that the bird species and the recognition confidence of the initial result belong to the same identification process as the values mentioned in the comparison result. Then, according to the judgment that the recognition confidence in the comparison result is greater than or equal to the preset threshold, the initial result is directly taken as the core content, the identification source label of the local effective identification is supplemented, the corresponding relationship between the label preset threshold and the actual recognition confidence is determined. Finally, the integrated information is processed in a standardized format, the field arrangement order is unified, the redundant temporary comparison records in the initial result are removed, the target recognition result with simple structure and clear core information is generated, and the time node and key verification information of the result determination are recorded simultaneously. This method is simple and efficient, does not require additional complex verification links, has low resource consumption, can quickly complete the determination of the target recognition result, and is suitable for scenes with high processing speed requirements and no need for complex tracing.

[0079] As another optional implementation, after receiving the comparison result, the generation logic of the comparison result is first verified in reverse, and it is confirmed that the comparison calculation process of the recognition confidence and the preset threshold is unbiased. Then, the feature comparison original data corresponding to the initial result is called, including the matching dimension and the coincidence degree distribution of the image features and the bird feature library, and the consistency of the bird species and the feature matching details in the initial result is checked to exclude the mismatching problem caused by data transmission or recording errors. Next, the initial result, the comparison result verification report, and the feature matching detail data are integrated, the detailed identification source description of the local effective identification and the feature matching that meet the standard are supplemented, the specific basis of the confidence that meets the standard is determined, and the corresponding target image data identifier is associated to establish multi-dimensional information association. Finally, all the information is arranged according to the structured standard to form a complete target recognition result containing the core identification conclusion, the verification process, the feature basis, and the associated identifier. This method has strong result certainty, avoids information errors to the greatest extent through multi-dimensional verification and detail supplement, and has high traceability and credibility.

[0080] Step C13, if the comparison result is that the recognition confidence is less than the preset threshold, the static bird image in the target image data and the initial result are packaged into secondary identification data to generate the secondary identification data.

[0081] In this embodiment, the secondary identification data refers to a data package that needs to be further accurately identified and integrates relevant images and preliminary conclusions.

[0082] As an optional implementation, after receiving the comparison result, the determination logic is reviewed to confirm that the recognition confidence calculation is correct and indeed less than the preset threshold. Then all static bird images in the target image data are retrieved, including pictures of different angles and different postures, without screening and keeping all. Then the complete content of the initial result is extracted, covering the details such as the suspected bird species, recognition confidence, full record of feature comparison, and possible reason analysis of low confidence. At the same time, the target image data identifier and the collection timestamp corresponding to the initial result are associated, and a uniform association identifier is added to all related data to establish a complete correspondence between the static bird image, the initial result, and the supporting information. Then, the data is fully integrated according to the standardized data structure to ensure that the classification of various information is clear and can be directly called. Finally, the integrated full-quantity static bird image, complete initial result and supporting associated information are packaged to generate comprehensive secondary identification data, and a data list is attached to explain the information categories and quantities contained. This method is comprehensive and complete, and can provide rich reference for secondary identification, helping accurate determination, and is suitable for scenes with very high identification accuracy requirements and the need for comprehensive reference of previous data.

[0083] Exemplarily, in the scene of the bird feeder interacting with the terminal, according to the preset confidence judgment rule (preset threshold 95%), the recognition confidence (initial result: bird species is red-beaked acacia bird, recognition confidence is 82%, after extracting features by MobileNetV3 lightweight model and comparing with local feature library containing 600 bird species) in the initial result is extracted. The numerical comparison between the recognition confidence of 82% and the preset threshold of 85% gives the comparison result of “recognition confidence less than preset threshold”. Since the comparison result does not meet “greater than or equal to preset threshold”, 3 static bird images of 800x600 pixels and JPEG format are retrieved from the target image data, including pictures of red-beaked acacia birds in different postures. The bird species, 82% confidence, and feature comparison differences (such as insufficient local feather texture matching degree) in the initial result are extracted completely, a uniform data number (2024082501) is added to establish an association, and the data is integrated and packaged in a structured format to generate secondary identification data containing static bird images and complete initial results.

[0084] Since the effectiveness of the initial result is determined by the preset threshold classification, the target recognition result is directly determined if it meets the standard, and the full-quantity associated data is packaged if it does not meet the standard, the problems of directly discarding data or misjudgment when the confidence is insufficient in traditional identification and the lack of support for previous data in secondary identification are solved, the utilization rate of the recognition result and the accuracy of the secondary identification are improved, and the reliability of the terminal user to obtain bird information is ensured.

[0085] Based on any of the above embodiments, in the fifth embodiment of the present application, refer to Figure 3 , Figure 3The flowchart of the control method of the bird feeder interaction system fifth embodiment. The step C13 described in the application includes steps D11~D13: Step D11, upload the secondary identification data to the cloud server, output the confirmation signal of successful reception of the cloud server.

[0086] In this embodiment, the cloud server refers to a remote server with data storage and deep processing capabilities. The confirmation signal of successful reception refers to the effective identification signal fed back by the cloud server after receiving the data completely.

[0087] As an optional implementation, the secondary identification data is subjected to integrity check to verify whether the static bird image, initial result and associated identifier are complete without missing, and the data is subjected to encryption processing to ensure transmission safety. Then a dedicated communication link with the cloud server is established, and the complete secondary identification data is transmitted in real time through the link. The transmission progress and status are continuously fed back during the transmission process. The cloud server automatically performs integrity and encryption check after receiving the data, and generates a reception success confirmation signal containing data number and reception time after the check is passed. The signal is returned along the original communication link, and after receiving and analyzing the confirmation signal, the whole process is completed and the full information of transmission and confirmation is recorded. This method has high data transmission safety and comprehensive check link, and can timely find data missing or transmission error.

[0088] As another optional implementation, the secondary identification data is subjected to block processing according to the preset specification, and a unique block identifier and an overall data association code are added to each block of data. Then enter the communication link monitoring state, and when it is monitored that the communication link bandwidth is sufficient, start the parallel upload of multiple blocks of data, without waiting for the completion of the transmission of a single block of data to initiate the transmission of the next block. After receiving each block of data, the cloud server sorts and splices the data through the block identifier and the association code. After splicing, the overall data integrity is checked. If the check is passed, a simple format of reception success confirmation signal is generated and fed back. After receiving the confirmation signal, the block temporary identifier is automatically cleaned up, and only the overall data transmission record is retained. This method has high efficiency of block parallel transmission, can fully utilize the bandwidth resources, has low congestion probability during multiple data transmission, and short transmission time.

[0089] Step D12, according to the confirmation signal of successful reception of the cloud server, calling the high-precision bird identification model to analyze and identify the secondary identification data, outputting the accurate identification result.

[0090] In this embodiment, the accurate identification result refers to the final identification conclusion with high credibility and detailed identification information generated after deep analysis.

[0091] As an optional implementation, the received cloud server receiving success confirmation signal is first verified for validity, and it is checked whether the data number in the signal is consistent with the associated identifier of the secondary identification data to be identified. Then the secondary identification data to be identified is disassembled, the static bird image and the initial result are separated, and the key information such as feature difference point and bird species type in the initial result is extracted as identification reference. Then the static bird image is input into the high-precision bird identification model, the deep feature mining process is started, the basic features such as global contour and feather texture are extracted, the local features of rare birds such as beak shape, eye details and feather texture distribution are analyzed, and the multi-dimensional cross comparison is performed combined with the reference information of the initial result and the built-in full-quantity bird feature library of the model. The uniqueness and accuracy of feature matching are repeatedly checked, and finally the accurate identification result containing accurate bird species, high-credibility identification confidence and core feature matching description is generated, and the original data identifier is associated for tracing. The identification process of the method is targeted, can fully utilize the reference information of the initial result to avoid invalid comparison, and deeply mine local detail features to improve identification accuracy.

[0092] As another optional implementation, all secondary identification data corresponding to the cloud receiving success confirmation signal is first summarized, grouped and classified according to the bird species type in the initial result, and an independent identification task identifier is assigned to each group of data. Then the high-precision bird identification model is called in batches, and the deep analysis is performed on multiple groups of data simultaneously by using parallel processing mechanism. During the processing, the static bird image in each group of data is first subjected to batch feature extraction, and the global and local core features are uniformly mined. Then, according to the initial result difference point of each group of data, the corresponding exclusive feature comparison module in the model is called for targeted verification, and the feature commonness and difference of the same type of birds are cross-group verified. After completing the feature comparison of all groups, the identification data of each group is integrated to generate the accurate identification result containing the accurate bird species, comprehensive identification confidence and group feature consistency analysis of each group of data, and the multi-dimensional association is established according to the task identifier and the original data identifier. The method adopts a grouping parallel processing mode, can fully utilize the model operation resources, greatly improves the identification efficiency of batch data, and reduces the misidentification probability of the same type of birds through cross-group verification.

[0093] Step D13, receiving the accurate identification result, supplementing the identification source of the accurate identification result as cloud secondary identification, and integrating to generate the target identification result.

[0094] In this embodiment, the identification source identifier identifies the key information of the identification process.

[0095] As an optional implementation, the accurate recognition result returned by the cloud is received, the core information such as bird species, recognition confidence, feature matching description, etc. in the result is checked for integrity, and after confirming that there is no missing or logical contradiction, the identification source: cloud secondary identification is supplemented in the specified field of the result. At the same time, the initial result in the secondary identification data corresponding to the accurate recognition result is called, and the initial result is integrated as auxiliary explanation information, and the association between the core identification conclusion and the auxiliary explanation is established. Then, the information is regularized according to the fixed structure of the identification source, the accurate species, the recognition confidence and the auxiliary explanation, and the redundant and repetitive contents are removed. Finally, the integrated information is logically checked to ensure that each part of the content is coherent and consistent, and the target recognition result is generated. The target recognition result of this method is comprehensive, has core conclusion and auxiliary background, and has strong traceability, and is suitable for scenes with high requirements for identification process transparency.

[0096] As another optional implementation, a plurality of accurate recognition results returned by the cloud are received in batches, grouped and sorted according to the data association identifier, and the identification source: cloud secondary identification is directly supplemented to each group of results. Subsequently, the core information in each group of results is extracted, only the bird species and the recognition confidence are retained, and the auxiliary information such as the feature matching description and the initial result is discarded, the core information and the source annotation are integrated according to the preset simple data format, the field arrangement order and the expression standard are unified, the merging processing of multiple groups of information is quickly completed, and a standardized target recognition result set containing core identification information and clear source annotation is generated. This method has high processing efficiency and can quickly complete batch data integration with low resource consumption, and is suitable for large-scale data processing scenarios.

[0097] Exemplarily, in the scenario of the bird feeder interacting with the terminal, the three 800x600 pixel JPEG format static bird images (containing different angle unobstructed pictures) and the initial result (the bird species is speculated to be the dark green painted eye bird, the identification confidence is 83%, and the feature comparison difference point is the throat feather texture matching) will be uploaded to the cloud server through the communication link after encryption processing. After the cloud server completes the data integrity check, the receiving success confirmation signal containing the data number 20240912003 and the receiving time 2024-09-12 14:38:15 is output. According to the confirmation signal, the EfficientNetB5 high-precision bird identification model is called, the image and the initial result are separated from the secondary identification data, and the initial result is used as a reference to focus on mining local detail features such as eye feather texture and beak length ratio in the image. Cross comparison is performed with the 800 bird full feature library built in the cloud in multiple dimensions, and the accurate identification result (the bird species is dark green painted eye bird, the identification confidence is 97%, and the core feature matching description is that the throat feather texture meets the subspecies characteristics) is output. After receiving the accurate identification result, the result specified field is supplemented with the label "identification source: cloud secondary identification", the accurate identification result and the label information are integrated, and the complete target identification result is generated according to the structure of "identification source, bird species, identification confidence, and feature matching description".

[0098] Due to the encryption transmission to ensure the integrity of the secondary identification data, the deep mining of detail features, and the supplement of cloud secondary identification source labels, the problems of easy loss of traditional secondary identification data transmission, low accuracy of low confidence scene identification, and no traceable result source are solved, the accuracy of secondary identification is improved, and the information reliability of the bird feeder and the terminal interaction is enhanced.

[0099] Based on any of the above embodiments, in the sixth embodiment of the present application, the step S30 includes steps E11-E13: Step E11, based on the target identification result and the associated information, positioning the marking object, outputting the bird image to be marked and its corresponding associated information and target identification result.

[0100] In this embodiment, positioning the marking object refers to the operation of accurately locking the bird main body area that needs to be marked from the bird image. The bird image to be marked refers to the core bird image data that needs to be marked in multiple dimensions.

[0101] As an optional implementation, the target recognition result is accurately matched with the associated information through a unique collection identifier, so that each target recognition result can correspond to unique associated information and the bird image to which it belongs. Subsequently, for each matched data set, the contour boundary of the bird subject is identified from the corresponding bird image, the background clutter interference is excluded, the core area of the marked object is accurately positioned, and the key feature area related to the target recognition result in the image is extracted as an auxiliary labeling basis. Then, the bird image to be marked, the complete associated information, and the target recognition result are structured and integrated according to the preset field order to determine the one-to-one correspondence of the three, and finally the integrated data with accurate positioning, complete information, and correct binding is output. This method has high positioning accuracy, no data correlation deviation, and can improve the subsequent marking pertinence.

[0102] As another optional implementation, all target recognition results are classified and summarized according to the bird species or recognition source in the target recognition result, and a unified classification identifier is assigned to each type of data. Then, the corresponding bird images and associated information of each category are batch retrieved, the bird subject in each image is quickly identified using a contour screening mechanism, and only the images with complete and non-severely obstructed bird subjects are retained as the images to be marked. Subsequently, the to-be-labeled bird images of the same category are sorted by collection time, and the associated information and target recognition result of the corresponding category are batch bound according to the classification identifier, the field format is uniformly regularized, and the batch integrated data set divided by category is directly output. This method has high batch processing efficiency, simple process, can quickly complete the positioning and integration of a large amount of data, has low resource consumption, and meets the needs of quickly advancing the overall process.

[0103] Step E12, converting the to-be-labeled bird image and its corresponding associated information and target recognition result into basic labeling information and recognition labeling information, and outputting intermediate labeling data.

[0104] In this embodiment, the basic labeling information refers to the basic labeling content converted from the associated information. The recognition labeling information refers to the recognition type labeling content converted based on the target recognition result. The intermediate labeling data refers to the transitional labeling data formed after integrating the basic and recognition labeling information, and the output refers to providing the intermediate labeling data to the subsequent operation.

[0105] As an optional implementation, the bird images to be marked, the corresponding associated information and the target recognition result are first called by group, and after confirming the binding of the three through the unique identification, the associated information is split according to the categories such as collection time, collection place and environmental data, and is converted into structured basic marking information. At the same time, the bird species, recognition confidence, recognition source and core feature matching description in the target recognition result are extracted, and are split and converted into standardized recognition marking information according to the recognition dimension. Then the deep association of the basic marking information and the recognition marking information is established, the field mapping relationship of the two types of marking information corresponding to the same bird image is determined, and the data integrity check identifier is supplemented. Finally, the two types of marking information and the corresponding bird image association identifier are integrated to form intermediate marking data containing complete fields and clear associations and are output. The marking information splitting of this method is detailed, the fields are complete, the association logic is rigorous, and it can provide comprehensive support for subsequent multi-dimensional marking.

[0106] As another optional implementation, all bird images to be marked, associated information and target recognition results are batch classified and summarized according to the recognition source or bird species in the target recognition result, and a unified classification code is assigned to each type of data. Then the same type of data is batch processed, the core basic fields in the associated information are extracted and converted into simplified basic marking information, and the core recognition fields in the target recognition result are extracted and converted into simplified recognition marking information. Then the same type of simplified basic marking information and recognition marking information are batch bound through classification coding, data batch check identifiers are added, and intermediate marking data sets divided by category and simplified fields are directly integrated and output. This method has high batch processing efficiency, simple process and simplified fields, which reduces resource consumption, can quickly promote subsequent links and meet the demand for rapid processing of massive data.

[0107] Step E13, after integrating and verifying the two types of marking information in the intermediate marking data and determining the correspondence between the marking content and the bird image, the marking information is generated.

[0108] In this embodiment, the two types of marking information refer to the basic marking information converted based on the associated information and the recognition marking information converted from the target recognition result in the intermediate marking data. The marking content refers to the specific content of the basic information and the recognition information to be marked on the bird image. The correspondence of the bird image refers to the unique binding relationship between the marking content and the bird image to which it belongs.

[0109] As an optional implementation, the basic mark information and the identification mark information in the intermediate mark data are called in groups. First, the logical consistency of the two types of information is verified. Whether the collection time and environmental data in the basic mark information and the bird species and identification source in the identification mark information are reasonably related is checked. The abnormal items with data contradictions or missing fields are excluded and the problem causes are recorded. Then, through the unique association identifier, the ownership relationship between each group of mark content and the corresponding bird image is confirmed one by one. It is determined that the image, the basic mark information and the identification mark information in a single group correspond to each other without misplacement. Subsequently, according to the hierarchical integration of the basic information, the identification information and the image association, the content that passes the verification is supplemented with verification identification and data traceability code, the field arrangement order is standardized, and redundant temporary records are removed. Finally, the mark information is generated. This method has comprehensive verification dimension, accurate data correspondence, no logical contradiction and misplacement problem in mark information, and high reliability.

[0110] As another optional implementation, all data is batch grouped according to the classification identifier in the intermediate mark data. A unified verification code is assigned to each group. Batch format verification is performed on the two types of mark information in each group. Whether the core fields of the basic mark information and the identification mark information are complete and the format is uniform is checked. Then, through the classification identifier and the collection timestamp, the corresponding relationship between the mark content and the bird image in the same group is batch established. The random sampling method is used to extract part of the data to verify the binding accuracy. Subsequently, the two types of mark information that pass the verification in each group are batch integrated. The field format and integration structure are unified. The abnormal data groups found by sampling are removed. The batch verification identification, sampling proportion and qualified condition description are supplemented. The mark information set that is classified and has complete core information is generated. This method has very high batch processing efficiency. It can quickly complete the integration of massive data. The resource consumption is low. The processing throughput is greatly improved.

[0111] Exemplarily, in the scenario of the bird feeder interacting with the terminal, based on the target recognition result (bird species: northern redstart, recognition confidence: 97%, recognition source: cloud secondary recognition) and the associated information (collection time: November 11, 2024 08:45:30, location: north latitude 32.1 degrees east longitude 119.7 degrees, environmental temperature: 25 degrees Celsius humidity: 60%), the contour recognition algorithm is used to locate the marked object, lock the bird main body core area (excluding the feeding trough background interference), and output 3 bird images in 800x600 pixel JPEG format to be marked and their corresponding complete associated information, target recognition result; the collection time / location / environmental data in the associated information of the bird image to be marked are converted into basic marking information, and the northern redstart, 97% confidence, and cloud secondary recognition in the target recognition result are converted into recognition marking information, which are regularized according to the “basic marking, recognition marking, and image identification” field, and structured intermediate marking data is output. The two types of marking information in the integrated intermediate marking data are checked, the logical consistency of the collection time and the recognition process is checked, the adaptability of the environmental data and the habits of the northern redstart is checked, the unique correspondence between the marking content and the bird image is determined through the image identification, 1 abnormal item with environmental data entry error is excluded, complete marking information containing basic marking information, recognition marking information, and image associated identification is generated, and each piece of marking information is attached with a verification passed identification.

[0112] Due to the accurate positioning of the marked object by the contour recognition algorithm, the field splitting and conversion of the marking information, and the multi-dimensional integration and verification of the data, the problems of blurred positioning of the bird main body, mixed basic information and recognition information, and misaligned data correspondence in the traditional marking are solved, and the accuracy and consistency of the marking information are improved.

[0113] Based on any of the above embodiments, in the seventh embodiment of the present application, the step S40 includes steps F11-F13: Step F11, comparing the target bird recognized in the marking information with the list of birds of interest, determining whether the target bird is a bird of interest of the user, and generating a bird species comparison result.

[0114] In this embodiment, the list of birds of interest refers to a set of bird species that the user needs to pay special attention to in advance. The bird species comparison result refers to the determination conclusion of whether the target bird belongs to the bird of interest of the user after matching.

[0115] As an optional implementation, the core information of the target bird is first extracted from the marking information, including the correct name, subspecies name and common alias of the bird. Then the user's attention bird list is called, and full-dimensional comparison is performed in turn according to the order of bird species in the list. The correct name is matched first, and if the correct name is not consistent, the alias is checked, and if the subspecies classification is involved, the subspecies attribution is further checked. Each step of comparison records the matching nodes and difference points. If there is a complete match in any dimension, it is determined that the user's attention bird, and if there is no matching item in all dimensions, it is determined that the non-attention bird. Finally, the detailed bird species comparison result containing the determination result, matching dimension and comparison process record is generated. This method has comprehensive comparison dimensions, can accurately identify the matching relationship of the correct name, alias and subspecies, avoids misjudgment caused by name difference, and ensures the pertinence of subsequent interactive information delivery.

[0116] As another optional implementation, the core correct name of the target bird is first extracted from the marking information, and the user's attention bird list is preprocessed to establish an index directory according to the bird family classification. Then the core correct name of the target bird is batch fuzzy matched with the index directory to directly search whether there is a consistent core correct name under the family classification. If a matching item is found, it is quickly determined that the user's attention bird, and if not, it is determined that the non-attention bird. Finally, the simple bird species comparison result containing only the core determination result and the matching family identification is generated. This method uses family index and core correct name fuzzy matching, has extremely fast processing speed, can adapt to scenes with many entries in the attention bird list, has low resource consumption and high throughput.

[0117] Step F12, if the comparison result is that the target bird is the user's attention bird, integrating the related information of the target bird, outputting the interactive information corresponding to the target bird.

[0118] In this embodiment, the related information refers to the data set of marking information, associated information, target identification result and the like related to the target bird. The interactive information refers to the standardized bird-related information integrated for the terminal to display to the user, and the output refers to the operation of delivering the interactive information to the terminal.

[0119] As an optional implementation, after confirming that the comparison result is the target bird species that the user is interested in, all relevant data of the target bird species are comprehensively retrieved, including the basic markers and identification markers in the marker information, the core feature matching description in the target identification result, and the associated images of the bird species to be marked. These data are subjected to multi-dimensional screening and integration, and redundant information is removed. The data are sorted according to the logical levels of the bird basic information, identification verification information, collection scene information, and image resources, and the data integrity verification identifier is supplemented. At the same time, the field expression and format are standardized, and finally the all-around interactive information containing text description, data parameters, and image links is generated. The interactive information obtained by this method is detailed and comprehensive in dimension, and can fully meet the user's demand for in-depth understanding of bird-related information, and is suitable for scenarios with high demand for information detail.

[0120] As another optional implementation, after confirming that the comparison result is the target bird species that the user is interested in, the core relevant information of the target bird species is quickly extracted, and only the bird species, identification confidence, collection time, core collection location, and a clear bird image core data are retained. The data are integrated according to the preset minimalist interactive format, the field order and expression standard are unified, the accuracy of the core data is verified, and after ensuring that there is no obvious error, the lightweight interactive information containing core text information and image resources is directly generated. The data extraction and integration process of this method is simple and efficient, the processing speed is fast, the interactive information volume is small, the storage and transmission resources are less occupied, and the interactive delay caused by information redundancy is avoided.

[0121] Step F13, the interactive information is transmitted to the terminal device through the communication module and displayed, and a confirmation signal of successful transmission of the interactive information is output.

[0122] In this embodiment, the communication module refers to a software and hardware communication component for transmitting interactive information. Display refers to the process of presenting interactive information in a visual form by a terminal device. The confirmation signal of successful transmission of the interactive information refers to the effective identifier fed back after the terminal device successfully receives and completes the display.

[0123] As an optional implementation, the integrity of the interactive information is checked to confirm that the core content of the text description, data parameters, and image resources is not missing or damaged. Then, the information is encrypted to ensure transmission security, the communication module is started to establish a dedicated real-time communication link with the terminal device, the complete interactive information is transmitted through the link, and the transmission progress is fed back in real time during the transmission process. After receiving the information, the terminal device automatically decrypts and checks, and after the check is passed, the visual presentation is completed according to the preset display template, and the confirmation signal containing the terminal identifier, reception time, and display state is generated and returned along the original link. After receiving the confirmation signal, the transmission duration, encryption check result, and terminal feedback details are recorded, and the whole process is completed. This method is safe and reliable in transmission, the display format is uniform, the confirmation signal contains multi-dimensional information, and the traceability is strong.

[0124] As another optional implementation, the plurality of interaction information is first grouped and classified according to terminal device identifiers, a unified group code is added to each group of information, each group of information is compressed to reduce the transmission volume, the communication module is started to enter the multi-link parallel transmission mode, and the compressed interaction information of the corresponding group is issued to different terminal devices. After receiving, the terminal device automatically decompresses, quickly completes the display according to the default display rule of the device, generates a simple confirmation signal containing only the group code and a successful issuance identifier after successful display, and returns in batches. After receiving the confirmation signal, the terminal device is matched according to the group code, and only whether the issuance is successful is recorded, without tracking detailed transmission and display details. This method has high batch parallel transmission efficiency, saves bandwidth through compression processing, can quickly complete multi-terminal and multi-information issuance, has low resource consumption, and is suitable for large-scale terminal concurrent reception scenarios.

[0125] Exemplarily, in the scene of interaction between the bird feeder and the terminal, the correct name, subspecies name and common alias of the target bird are extracted from the marking information, the user's attention bird list is called, including the common subspecies of the Eurasian Jay, the Red-billed Blue Magpie and the Magpie, and the correct name, subspecies and alias are compared in all dimensions. It is confirmed that the common subspecies of the Eurasian Jay completely matches the list entries, and the bird species comparison result that the target bird is the user's attention bird is generated. The related information of the target bird is integrated: the basic mark in the marking information (collection time 2024-10-5 09:20:18, location north latitude 31.5 degrees east longitude 120.3 degrees, environment temperature 23 degrees Celsius humidity 55%), identification mark (identification confidence 96%, identification source cloud secondary identification), core feature description in the target identification result (the head crown spread form conforms to the common subspecies characteristics), and 1 1080P clear bird image. According to the structure of “core introduction, collection scene, identification verification and image display”, the interaction information containing text description, data parameters and image resources is output. Through the communication module, an encrypted link is established, and the interaction information is issued to the user's mobile phone APP terminal. After the APP receives, it is displayed according to the preset template (top image, middle core text and bottom data parameters). After successful display, the interaction information issuance success confirmation signal containing the terminal ID (SN20240518) and the receiving time 2024-10-5 09:20:21 is fed back, and the transmission process data is recorded after receiving.

[0126] Due to the full-dimensional subspecies-level comparison, structured integration of interaction information and encrypted communication link issuance, the problems of subspecies missed judgment caused by traditional comparison only matching the correct name, disorganized interaction information and information leakage in the issuance process are solved, and the bird comparison accuracy, interaction information readability and issuance security are improved.

[0127] Based on any of the above embodiments, in the eighth embodiment of the present application, refer to Figure 4 , Figure 4The flowchart of the eighth embodiment of the control method of the bird feeder interaction system of the present application is shown. After step S40, steps G11-G13 are further included. Step G11, collect user's manual annotation data for unidentified birds, and integrate to generate feedback data.

[0128] In this embodiment, unidentified birds refer to bird-related data for which the system has failed to determine the species. Manual annotation data refers to supplementary information such as species and characteristics manually input by the user for unidentified birds. Feedback data refers to standardized data formed by integrating user annotation data and original data of unidentified birds, which is used to optimize recognition capability.

[0129] As an optional implementation, the complete original data of unidentified birds is retrieved, including the image to be labeled, associated collection time / location / environment data, and previous recognition failure records. A unique traceable identifier is assigned to each set of data, and then a structured annotation template is provided to the user, which includes fields such as bird name, subspecies, core feature description, annotation basis notes, etc., guiding the user to fill in the complete information according to the template. After collecting the user's manual annotation data, the field completeness is first verified, and invalid data with missing key information is removed. Then the reasonableness of the annotation content is checked, and the annotation data that passes the verification is accurately bound with the original data of the corresponding unidentified bird according to the traceable identifier, supplemented with metadata such as annotation submission time and user identifier, and structured according to the original data, annotation data, and metadata, to generate detailed feedback data containing multi-dimensional information. This method has comprehensive feedback data information and high credibility, and can provide accurate support for recognition model optimization and feature library supplementation, suitable for scenarios with strict requirements for feedback data quality.

[0130] As another optional implementation, only the core original data of unidentified birds is retrieved, and a minimalist annotation interface is provided to the user, with only one required field of bird name. The user does not need to fill in additional information and can complete the annotation quickly by submitting. After collecting all user-submitted annotation data, obviously incorrect annotations are filtered out. Then, the core original data of unidentified birds is batch-associated according to the data identifier, and the annotation submission timestamp is batch-supplemented. The data is integrated according to the simple structure of data identifier, bird name, and annotation time to generate a set of simplified feedback data. This method has a very low user operation threshold and high annotation efficiency, can quickly collect a large amount of data, and has short processing time and low resource consumption for batch integration.

[0131] Step G12, transmit the feedback data to the cloud server, combine with the new bird samples, retrain the high-precision bird recognition model, and generate an optimized high-precision bird recognition model.

[0132] In the present embodiment, the new bird sample refers to the bird feature image and associated information data that are not included in the original model training and are supplemented. The optimized high-precision bird recognition model refers to the final model with improved recognition ability after training with new data.

[0133] As an optional implementation, the feedback data is subjected to secondary verification of integrity and rationality, and abnormal items with labeling contradictions and data missing are excluded. Then, the effective feedback data is subjected to encryption processing and transmitted to the cloud server through a dedicated communication link. After receiving, the server classifies and regularizes the new bird samples according to bird families and morphological characteristics, merges the same data to form a special training data set, and then fuses the data set with the original model's basic training data in proportion. For the bird categories that the original model recognizes weakly, the training weight of the feature extraction module is strengthened, and targeted multi-round iterative training is carried out. The recognition accuracy change is monitored in real time during the training process, and the optimized high-precision bird recognition model that adapts to the weak category recognition is generated after the accuracy stabilizes and meets the standard. This method has strong training pertinence and can accurately compensate for the recognition shortcomings of the original model. After optimization, the human recognition accuracy of specific categories is significantly improved, and the comprehensiveness and accuracy of model recognition are improved.

[0134] As another optional implementation, the feedback data is batch compressed in the original format and quickly uploaded to the cloud server through a parallel transmission link. After receiving, the server directly mixes the feedback data with the new bird samples to form a large-capacity comprehensive training data set. Without distinguishing between strong and weak categories of the original model, the incremental training mode is adopted, the comprehensive data set is input into the original high-precision bird recognition model as a whole, the training framework and basic parameters of the original model are followed, and only the feature matching library and parameter threshold of the model are updated through multiple rounds of data iteration. During the training process, only the overall recognition coverage and average accuracy are monitored. After the overall indicators meet the standard, the optimized high-precision bird recognition model is generated. The data processing and training process of this method is simple and efficient, batch operation greatly shortens the training period, resource consumption is low, and model iteration can be quickly completed.

[0135] Step G13, the optimized high-precision bird recognition model is subjected to lightweight processing to generate an update package that adapts to the edge computing module of the bird feeder, so as to complete the update of the lightweight bird recognition model.

[0136] In the present embodiment, lightweight processing refers to a processing process that simplifies the model structure and reduces resource occupation under the premise of retaining core recognition functions. The edge computing module of the bird feeder refers to a hardware component built-in the bird feeder with local data processing capability. The update package refers to a standardized installation package containing the model file after lightweight processing and an update script. The update of the lightweight bird recognition model refers to the operation of replacing the original local model with the optimized lightweight model.

[0137] As an optional implementation, the network structure and parameter distribution of the optimized high-precision bird recognition model are analyzed, the core recognition layer and the redundant auxiliary layer are identified, the key structure directly related to bird feature extraction and species matching is retained, and the repeated calculation module and redundant parameters are removed. Then, the remaining core parameters are quantized and compressed, the data format is unified to reduce storage occupation, and the hardware architecture and operation capacity of the bird feeder edge computing module are adapted, the model running dependency library and interface protocol are adjusted to ensure compatibility. Subsequently, the processed lightweight model is integrated with the automatic update script and version verification file, packaged according to the storage path specification of the bird feeder edge computing module, and an exclusive update package containing the model core file, update process instructions, and verification mechanism is generated. The standardized format ensures that the update package can be correctly identified and replaced by the edge computing module, completing the update of the original lightweight bird recognition model. The model generated by this method has strong adaptability, accurately matches the resource occupation of the edge computing module, and has minimal loss of core recognition ability.

[0138] As another optional implementation, the optimized high-precision bird recognition model is directly compressed in bulk, a general lightweight algorithm is used to reduce the model file size and parameter total, and the format is converted according to the industry-standard edge device model specification to ensure basic compatibility. Subsequently, the compressed model and the general update script are packaged, the script contains basic instructions such as automatic decompression, path matching, and version coverage, and a simple structure and small size general update package is generated. This update package can adapt to the interface protocol of most mainstream bird feeder edge computing modules, and the edge computing module automatically executes the update process after receiving the update package, replacing the original lightweight model. This method has a simple and efficient processing flow, fast batch generation speed, and wide adaptation range, without the need for large individual adaptation costs, improving the overall recognition system update efficiency.

[0139] Exemplarily, in the scenario of bird feeder interacting with terminal, the un-identified bird data uploaded by the bird feeder (containing 2 clear 1080P static images, associated information: collected at 30.8 degrees north latitude and 120.1 degrees east longitude on October 12, 2024, at 10:15, and the environmental temperature is 22 degrees Celsius) is collected. Through the terminal APP, a structured labeling template is provided to the user, and the user manually inputs the labeling data (the bird species is gray-headed bunting, the core feature is gray crown on the head, and the labeling basis is the color distribution of the feathers). The un-identified bird original data and the manually labeled data are integrated to generate feedback data containing data identifier 20241012007 and labeling integrity verification passed. The feedback data is encrypted and transmitted to the cloud server through the communication link, merged with the new bird sample (50 characteristic images of gray-headed bunting in different poses and different lightings and subspecies information), and forms a special training data set. Based on the original EfficientNetB5 high-precision bird recognition model, the gray-headed bunting feature extraction weight is supplemented, and the model is retrained for 10 iterations to generate an optimized high-precision bird recognition model (the recognition accuracy of gray-headed bunting reaches 95%). The optimized model is lightened by pruning 30% redundant network layers, quantizing 32-bit floating-point parameters to 8-bit integer, and removing non-core auxiliary function modules. The model is adapted to the ARM Cortex-A7 architecture edge computing module of the bird feeder, generates a standardized update package with a size of 8MB, and is pushed to the bird feeder through OTA to replace the original MobileNetV3 lightweight bird recognition model, and the update is completed.

[0140] Further, in the scenario of bird feeder interacting with terminal, an outdoor bird feeder interaction method with the body user, applied to intelligent outdoor bird feeder (hereinafter referred to as "bird feeder"), cloud server and user terminal, the bird feeder includes food storage, food trough, image acquisition module, edge computing module, communication module, control module, the interaction method includes the following steps, initialization interaction and image acquisition configuration: the user scans the bird feeder unique identification code through the terminal APP, completes the device activation and encrypted connection, and at the same time, the APP "image setting" interface is configured to collect parameters: trigger mode: select "bird approaching trigger" or "timed collection"; image parameters: set resolution, shooting mode, night light opening threshold. After the bird feeder control module receives the configuration parameters, it is synchronized to the image acquisition module and edge computing module, and the interaction initialization is completed. Bird image acquisition and preprocessing: image acquisition trigger: the infrared detector detects that there are birds in the food trough area, immediately sends "collection instruction" to the control module, and starts the camera shooting; "timed collection", the control module triggers the camera shooting according to the preset time node, and records the collection time, environmental temperature and humidity. Image preprocessing: the image preprocessing unit optimizes the collected original image, including noise reduction, defogging and cropping; the preprocessed image is converted to a format supported by the edge computing module, and the image size is compressed to reduce data transmission bandwidth occupation. Bird image recognition and classification: edge computing module local preliminary identification: the edge computing module is built-in lightweight bird recognition model (covering common outdoor bird species, such as sparrow, pigeon, embroidered eye bird, white-headed pheasant, etc., the model weight is deployed in advance through cloud training); input the preprocessed bird image, the model extracts the image features, compares with the built-in bird feature library, and outputs the preliminary identification result, including bird species, recognition confidence; if the recognition confidence is greater than or equal to the preset threshold, it directly enters the marking link; if the confidence is less than the threshold, the image and the preliminary identification result are packaged and uploaded to the cloud server through the communication module for secondary accurate identification. Cloud server secondary identification: the cloud server carries high-precision bird recognition model (covering more rare bird species, and supporting real-time update), receives low-confidence image data uploaded by the bird feeder; adopts multi-feature fusion algorithm for deep identification, and outputs accurate identification result. Identification result marking and building: multi-dimensional marking: the bird feeder control module combines the identification result and associated data to mark each bird image in multiple dimensions, and the marking information includes: basic information: image acquisition time, acquisition location, environmental data; identification information: bird species, recognition confidence, identification source; user associated information: if the bird is the user's preset "attention bird", it is marked as "target attention bird"; at the same time, the user's exclusive "bird observation database" is established in the cloud to avoid data loss.Data delivery and user interaction: real-time delivery trigger: if the recognition result is "user interested in birds", the bird feeder immediately delivers the following data to the user terminal through the communication module: real-time image / short video: clear bird image or short video after preprocessing; structured label information: collection time, bird species, recognition confidence, environment data; reminder mark: APP pop-up window reminds "your interested white-headed babbler has arrived, click to view details"; if it is not an interested bird, no classified image data and label information will be delivered. User terminal data presentation and interaction: the user terminal APP displays the received data in real time, and the user can manually label the bird for the un-identified user, and the labeling result is fed back to the cloud server for optimizing the recognition model. Model and data update optimization: recognition model iteration update: the cloud server regularly counts the manual labeling data fed back by the user, combines the new bird samples, re-trains the cloud high-precision recognition model, performs lightweight processing on the trained model, generates an update package suitable for the edge computing module of the bird feeder, pushes it to the bird feeder through the communication module, and completes the local model update after user confirmation, continuously improving the recognition accuracy.

[0141] Due to the supplement of un-identified bird data by user manual labeling, the iteration training model combined with new samples in the cloud, and the lightweight adaptation of the edge module, the problems of traditional model unable to identify rare / unrecorded birds, update package too large to adapt to the hardware of the bird feeder, and lack of user feedback support for model iteration are solved, improving the bird recognition coverage, model adaptability and iteration efficiency, and ensuring the local recognition efficiency and hardware adaptability.

[0142] The application provides a bird feeder, which comprises at least one processor and a memory in communication connection with the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the control method of the bird feeder interaction system in Embodiment I.

[0143] Reference will be made to Figure 5 which shows a structural schematic diagram of a bird feeder suitable for implementing the embodiments of the application. The bird feeder in the embodiments of the application can include but is not limited to mobile terminals such as mobile phones, notebook computers, automatic feeding devices, personal digital assistants (PDA), tablet computers (PAD), portable media players (PMP), intelligent recognition algorithm systems, and fixed terminals such as multi-modal recognition devices and desktop computers. Figure 5 The illustrated bird feeder is only an example and should not impose any limitation on the functions and use range of the embodiments of the application.

[0144] As shown in Figure 5 the bird feeder can include a processing device 1001 (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. Various programs and data required for the operation of the bird feeder are also stored in the random access memory 1004. The processing device 1001, the read-only memory 1002, and the random access memory 1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; the storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the bird feeder to communicate with other devices wirelessly or by wire to exchange data. Although the bird feeder with various systems is shown in the figure, it should be understood that all the systems shown are not required to be implemented or possessed. More or fewer systems can be alternatively implemented or possessed.

[0145] In particular, the processes described above with reference to the flowcharts can be implemented as a computer software program according to embodiments of the present disclosure. For example, embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer readable medium, the computer program containing program code for executing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network by a communication device, or installed from the storage device 1003, or installed from the read-only memory 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the methods of embodiments of the present disclosure are performed.

[0146] The bird feeder provided by the present application adopts the control method of the bird feeder interactive system in the above-mentioned embodiments, and can solve the technical problem of poor data acquisition quality. Compared with the prior art, the bird feeder provided by the present application has the same beneficial effects as the control method of the bird feeder interactive system provided by the above-mentioned embodiments, and other technical features in the bird feeder are the same as those disclosed in the previous embodiment method, which will not be repeated here.

[0147] It should be understood that various aspects of the disclosure can be implemented in hardware, software, firmware, or combinations thereof, to achieve the above described functionality. In the description above, specific terminology has been used to describe particular features, structures, materials or characteristics. Such terminology is used in the descriptive sense and not for the purposes of limitation.

[0148] The above description is only specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0149] The present application provides a computer readable storage medium having stored thereon computer readable program instructions (i.e., a computer program) for performing the control method of the bird feeder interaction system in the above-described embodiments.

[0150] The computer readable storage medium provided by the present application may, for example, be a U disk, but is not limited to an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, system, or device, or any combination thereof. More specific examples of the computer readable storage medium can include, but are not limited to, an electrical connection having one or more conductive wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read only memory (ROM), an erasable programmable read only memory (EPROM or flash memory), an optical fiber, a portable compact disk read only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present embodiment, the computer readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer readable storage medium can be transmitted in any suitable medium, including but not limited to electrical wire, optical cable, radio frequency (RF), and the like, or any suitable combination thereof.

[0151] The above computer readable storage medium can be contained in the bird feeder; or can exist separately without being assembled into the bird feeder.

[0152] The aforementioned computer-readable storage medium carries one or more programs. When the bird feeder executes these programs, it causes the bird feeder to: detect bird activity in the feeding trough area, collect relevant data according to preset time nodes, preprocess the data to obtain target image data and associated information; extract image features from the target image data, compare and identify them with a bird feature database, and generate target recognition results; based on the associated information and the target recognition results, label the bird images in the target image data in multiple dimensions to generate labeling information; and after determining the bird that the user is interested in through the labeling information, send the corresponding interactive information for the bird to the terminal device.

[0153] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0154] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0155] The modules described in the embodiments of the present application can be implemented in the form of software or in the form of hardware. In some cases, the name of the module does not constitute a limitation on the module itself.

[0156] The readable storage medium provided by the present application is a computer readable storage medium, which stores computer readable program instructions (i.e. computer programs) for executing the control method of the bird feeder interaction system, and can solve the technical problem of poor data acquisition quality. Compared with the prior art, the computer readable storage medium provided by the present application has the same beneficial effects as the control method of the bird feeder interaction system provided by the above-mentioned embodiments, and will not be described here.

[0157] The above only describes some embodiments of the present application, and does not limit the patent scope of the present application. Any equivalent structural transformation made by using the content of the present application specification and drawings, or direct / indirect application in other related technical fields is included in the patent protection scope of the present application.

Claims

1. A control method for a bird feeder interactive system, characterized in that, The method includes: After detecting bird activity in the feeding area, relevant data are collected according to preset time nodes, and preprocessed to obtain target image data and related information; Extract image features from the target image data, compare and identify them with a bird feature database, and generate target recognition results; Based on the association information and the target recognition result, bird images in the target image data are labeled in multiple dimensions to generate labeling information; After determining the bird that the user is interested in based on the tagging information, the corresponding interactive information for the bird is sent to the terminal device.

2. The control method of the bird feeder interactive system as described in claim 1, characterized in that, After detecting bird activity in the feeding trough area, the steps of collecting relevant data according to preset time nodes and preprocessing to obtain target image data and related information include: After receiving the bird feeder interaction initialization completion signal, the infrared detector detects bird activity in the feeding trough area, generates a collection command, and determines the collection time node according to the preset collection rules, thus obtaining the collection command and the corresponding preset time node; Based on the acquisition command and its corresponding preset time node, the camera is controlled to capture static images and short videos of birds according to the preset time node, and the associated information is collected simultaneously, and the original image data and the associated information are output. The original image data is preprocessed to obtain optimized target image data, which is then associated with the corresponding association information to obtain the target image data and its corresponding association information.

3. The control method of the bird feeder interactive system as described in claim 1, characterized in that, The step of extracting image features from the target image data, comparing them with a bird feature database, and generating a target recognition result includes: A lightweight bird recognition model is loaded to extract features from static bird images in the target image data, thereby obtaining the image features; The image features are compared and analyzed with the bird feature database to output initial results including bird species and recognition confidence. The validity of the initial result is determined according to the preset reliability judgment rules, and the target recognition result is generated by combining the high-precision bird recognition model.

4. The control method of the bird feeder interactive system as described in claim 3, characterized in that, The step of determining the validity of the initial result according to a preset reliability judgment rule and generating the target recognition result in combination with a high-precision bird recognition model includes: Based on the preset confidence judgment rule, the relationship between the recognition confidence in the initial result and the preset threshold is compared to obtain the comparison result; If the comparison result is that the recognition confidence is greater than or equal to the preset threshold, then the corresponding initial result is determined as the target recognition result; If the comparison result indicates that the recognition confidence is less than the preset threshold, then the static bird images in the target image data are packaged with the initial result into data to be recognized a second time, and the data to be recognized a second time is generated.

5. The control method of the bird feeder interactive system as described in claim 4, characterized in that, After the step of packaging the static bird images in the target image data with the initial result into data to be identified a second time, if the comparison result is that the recognition confidence is less than the preset threshold, and generating the data to be identified a second time, the control method of the bird feeder interaction system further includes: The data to be identified a second time is uploaded to the cloud server, and a confirmation signal indicating that the cloud server has successfully received the data is output. Based on the successful confirmation signal received by the cloud server, the high-precision bird recognition model is invoked to deeply analyze and identify the data to be identified a second time, and an accurate recognition result is output. The system receives the accurate recognition result, adds a note indicating that the source of the accurate recognition result is secondary recognition in the cloud, and integrates these to generate the target recognition result.

6. The control method of the bird feeder interactive system as described in claim 1, characterized in that, The step of generating labeling information by multi-dimensionally labeling bird images in the target image data based on the association information and the target recognition result includes: Based on the target recognition results and the associated information, the target object is located and labeled, and the image of the bird to be labeled, its corresponding associated information, and the target recognition results are output. The bird images to be labeled, along with their corresponding association information and target recognition results, are converted into basic labeling information and recognition labeling information, and intermediate labeling data is output. After integrating and verifying the two types of marker information in the intermediate marker data and determining the correspondence between the marker content and the bird image, the marker information is generated.

7. The control method of the bird feeder interactive system as described in claim 1, characterized in that, The step of determining the bird a user is interested in using the tagging information and then sending the corresponding interaction information for that bird to the terminal device includes: The target birds identified in the tagging information are compared with the list of birds of interest to determine whether the target birds are birds of interest to the user, and a bird species comparison result is generated. If the comparison result indicates that the target bird is the bird that the user is interested in, then the relevant information of the target bird is integrated and the corresponding interaction information of the target bird is output. The interactive information is sent to the terminal device via the communication module and displayed, and a confirmation signal indicating successful transmission of the interactive information is output.

8. The control method of the bird feeder interactive system as described in claim 1, characterized in that, After determining the bird the user is interested in using the tagging information and sending the corresponding interactive information to the terminal device, the control method of the bird feeder interaction system further includes: Collect user-generated annotations of unidentified birds and integrate them to generate feedback data; The feedback data is transmitted to the cloud server, and combined with the new bird samples, the high-precision bird recognition model is retrained to generate an optimized high-precision bird recognition model. The optimized high-precision bird recognition model is processed in a lightweight manner, and an update package adapted to the edge computing module of the bird feeder is generated to complete the update of the lightweight bird recognition model.

9. A bird feeder, characterized in that, The bird feeder includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the control method of the bird feeder interactive system as described in any one of claims 1 to 8.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the control method of the bird feeder interactive system as described in any one of claims 1 to 8.

Images