Flying animal detection method and system

Abstract

The invention is suitable for the technical field of flying animal detection, and particularly relates to a flying animal detection method and system, and the method comprises the steps: receiving a detection request uploaded by a user, extracting a to-be-detected picture, backtracking a source device, and searching an edge node, deploying a pre-constructed global target detection model into the edge node, inputting the picture into the global target detection model, and outputting to obtain a plurality of flying animal instances and candidate boxes; calculating the brightness value of each pixel point by using a brightness formula, drawing a two-dimensional frequency base graph formed by a primary function, and segmenting the picture into a plurality of blocks; according to the method, by determining the intersection set and the union set, the suspected position can be verified, missing detection is avoided, false detection areas are effectively filtered, the flying animals can be marked in a man-machine cooperation mode by constructing the verification task, the professional threshold is lowered, the public participation degree is increased, and the marking efficiency and accuracy are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of flying animal detection, and in particular to a flying animal detection method and system. Background Art

[0002] In many aspects such as environmental impact assessment, pest and disease monitoring, and wildlife protection, monitoring equipment is often deployed to collect image data of flying animals in the area in order to monitor the species, number, and activity patterns of flying animals in the target area. The detection task of flying animals is generally completed by the target detection model.

[0003] However, traditional target detection models are prone to false detection and missed detection in complex backgrounds, and have limited ability to distinguish different types of flying animals, especially those with similar appearances or smaller sizes.

[0004] Therefore, “how to use the target detection model to perform multiple detections on the same image” is the technical problem that the present invention needs to solve. Summary of the Invention

[0005] The purpose of the present invention is to provide a flying animal detection method and system to solve the problem of "how to use the target detection model to perform multiple detections on the same image" raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions: A flying animal detection method, comprising: Receive detection requests uploaded by users, extract images to be detected, trace back to the source device, find edge nodes, deploy the pre-built global object detection model to the edge nodes, input the images into the global object detection model, and output several flying animal instances and candidate boxes; Using the brightness formula, the brightness value of each pixel is calculated, and a two-dimensional frequency basis map composed of basis functions is drawn. The image is divided into several blocks, and the blocks are quantized using the two-dimensional frequency basis map to obtain the transformation coefficients. Using the preset brightness matrix and chromaticity matrix, the high-frequency signal area is traversed and interpolated and amplified; Create a local object detection model and deploy it to a cloud server. Input the high-frequency signal area into the local object detection model, and output several flying animal instances and verification boxes. Compare the candidate boxes and verification boxes, calculate the intersection-over-union ratio, obtain the intersection and union, and insert the intersection into the flying animal instance. Integrate edge nodes and cloud servers to build a distributed processing architecture, upload the verification box to the cloud server, extract the fragment corresponding to the verification box from the image, and generate a verification task. When a detection request is received, issue the verification task, obtain the verification result, mark the flying animal instance and verification result in the image, and send it to the preset terminal.

[0007] Furthermore, the steps of tracing back the source device, finding the edge node, and deploying the pre-built global object detection model to the edge node include: Build a scheduling platform for edge nodes and collect evaluation indicators of each edge node, where the evaluation indicators include at least: latency and available bandwidth; Configure the weight value of each evaluation indicator, collect the real-time value of the evaluation indicator, calculate the evaluation score, upload it to the scheduling platform, and dynamically update it.

[0008] Furthermore, the method further comprises: Based on the evaluation scores, edge nodes are clustered into common nodes and super nodes, and the global object detection model is deployed to several super nodes; Identify the load of each supernode and integrate the load balancing mechanism into the scheduling platform.

[0009] Furthermore, the step of dividing the image into a plurality of blocks, quantizing the blocks using a two-dimensional frequency basis map, and obtaining transform coefficients includes: Set the basic range of each block and use the edge detection algorithm to traverse the edge pixels in the image; The number of edge pixels in each basic range is counted, and the basic range is adjusted based on the number, wherein the adjustment includes: expansion and contraction.

[0010] Furthermore, the steps of calculating the intersection-to-union ratio, obtaining the intersection and the union, and inserting the intersection into the flying animal instance include: Calculate the intersection-over-union (IoU) of each image, select a first set value and a second set value, define images with an IoU less than the first set value as negative samples, and define images with an IoU greater than the second set value as positive samples; Integrate positive samples and negative samples to generate a training set, and train the global object detection model and the local object detection model.

[0011] Furthermore, when a detection request is received, the steps of issuing a verification task and obtaining a verification result include: Setting up several distribution channels for the verification task, and calculating a confidence score for each verification box based on the verification results; The verification results with the confidence score greater than the threshold are marked in the image.

[0012] Furthermore, the system includes: The receiving module is used to receive detection requests uploaded by users, extract the images to be detected, trace back to the source device, find the edge node, deploy the pre-built global object detection model to the edge node, input the image into the global object detection model, and output several flying animal instances and candidate boxes; The amplification module is used to calculate the brightness value of each pixel using the brightness formula, draw a two-dimensional frequency basis map composed of basis functions, divide the image into several blocks, quantize the blocks using the two-dimensional frequency basis map, obtain the transformation coefficients, and use the preset brightness matrix and chrominance matrix to traverse the high-frequency signal area and perform interpolation amplification; The insertion module is used to create a local target detection model and deploy it to the cloud server. The high-frequency signal area is input into the local target detection model, and the output is several flying animal instances and verification boxes. The candidate boxes and verification boxes are compared, and the intersection-union ratio is calculated to obtain the intersection and union. The intersection is then inserted into the flying animal instance. The sending module is used to integrate edge nodes and cloud servers to build a distributed processing architecture. It uploads the verification box to the cloud server, extracts the fragment corresponding to the verification box from the image, and generates a verification task. When a detection request is received, it issues the verification task, obtains the verification result, marks the flying animal instance and the verification result in the image, and sends it to the preset terminal.

[0013] Furthermore, the receiving module includes: A collection unit, configured to construct a scheduling platform for edge nodes and collect evaluation indicators of each edge node, wherein the evaluation indicators include at least latency and available bandwidth; The update unit is used to configure the weight value of each evaluation indicator, collect the real-time value of the evaluation indicator, calculate the evaluation score, upload it to the scheduling platform, and dynamically update it.

[0014] Furthermore, the amplification module includes: The traversal unit is used to set the basic range of each block and use the edge detection algorithm to traverse the edge pixels in the image; The counting unit is used to count the number of edge pixels in each basic range, and adjust the basic range based on the number, wherein the adjustment includes: expansion and contraction.

[0015] Furthermore, the insertion module includes: A calculation unit is configured to calculate an IoU ratio of each image, select a first set value and a second set value, define an image with an IoU ratio less than the first set value as a negative sample, and define an image with an IoU ratio greater than the second set value as a positive sample; The training unit is used to integrate positive samples and negative samples to generate a training set and train the global target detection model and the local target detection model.

[0016] Compared with the prior art, the present invention has the following beneficial effects: By deploying the global target detection model to the edge node, the data transmission pressure can be reduced, flying animals can be quickly screened out, and suspected locations can be marked. By determining the high-frequency signal area, the edge and texture features of flying animals can be highlighted, the distinction between flying animals and the background can be enhanced, and the detection accuracy of flying animals can be improved. Through interpolation and amplification, the recognizability of small targets is enhanced, the ability to express edge details is greatly improved, and the detection accuracy is further improved. By constructing a local target detection model, the image can be secondary detected, which improves the recognition efficiency of small flying animals. By determining the intersection and union, the suspected location can be verified to avoid missed detection and effectively filter out the false detection area. By constructing the verification task, the flying animals can be labeled in a human-machine collaborative way, which increases public participation and greatly improves the labeling efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A flowchart of a flying animal detection method provided in an embodiment of the present invention; Figure 2 A flowchart of the first sub-process of the flying animal detection method provided in an embodiment of the present invention; Figure 3 A block diagram of a second sub-process of the flying animal detection method provided in an embodiment of the present invention; Figure 4 A block diagram of a third sub-process of the flying animal detection method provided in an embodiment of the present invention; Figure 5 A fourth sub-flow chart of the flying animal detection method provided in an embodiment of the present invention; Figure 6 A block diagram of a flying animal detection system according to an embodiment of the present invention; Figure 7 A block diagram of the receiving module in the flying animal detection system provided in an embodiment of the present invention; Figure 8 A block diagram of the components of the amplification module in the flying animal detection system provided in an embodiment of the present invention; Figure 9 A block diagram of the components of the insertion module in the flying animal detection system provided by an embodiment of the present invention; Figure 10 This is a block diagram of the composition of the sending module in the flying animal detection system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0019] In Example 1, Figure 1 The implementation process of the flying animal detection method provided by the embodiment of the present invention is shown and described in detail below: S100: Receive the detection request uploaded by the user, extract the image to be detected, trace back to the source device, find the edge node, deploy the pre-built global target detection model to the edge node, and input the image into the global target detection model to output several flying animal instances and candidate boxes.

[0020] Receive detection requests uploaded by users and determine the images that need to be detected, where the images can also be video snapshots. For example, the automatic monitoring cameras in a nature reserve collect image data on a daily basis and upload them to the cloud platform. One day, the administrator initiates a detection request through the management system to find flying animals that appeared in the reserve on that day or in the recent period. If the data collected by the monitoring camera is video data, several snapshots are captured from it and used to detect flying animals.

[0021] Determine the source device of the image, and find the edge node that the source device can access based on the deployment location of the source device and the network topology architecture, where the edge node can be an intelligent network or an edge server, etc.; build a global target detection model, where the global target detection model is a target detection model in the existing technology for detecting flying animals in the image, and the target detection model is built based on a convolutional neural network and is iteratively trained with flying animal features; deploy the global target detection model to the edge node, and input the image into the global target detection model, identify the flying animals in the image by performing feature extraction and multi-layer deep convolution on the image, and output it as an instance, which refers to a specific target or individual in the image, such as a person, tree, and vehicle, etc.; the global target detection model also generates multiple candidate boxes, which delineate the location area in the image that may contain flying animals.

[0022] S200: Calculate the brightness value of each pixel using the brightness formula, draw a two-dimensional frequency basis map composed of basis functions, divide the image into several blocks, quantize the blocks using the two-dimensional frequency basis map to obtain transformation coefficients, and use the preset brightness matrix and chromaticity matrix to traverse the high-frequency signal area and perform interpolation and amplification.

[0023] The brightness formula (Y=0.299R+0.587G+0.114B, where Y represents the brightness value of a pixel and R, G, and B represent the intensity values ​​of the red, green, and blue color channels of the pixel, respectively) is used to calculate the brightness value of each pixel in the image and convert the image into a grayscale image. A two-dimensional frequency basis function from the prior art is used to decompose the grayscale image to obtain a corresponding two-dimensional frequency basis map, which shows the intensity distribution of the image at different spatial frequencies. The image is divided into several fixed-size blocks (e.g., 8×8 pixel blocks). Each block is discretely transformed using the two-dimensional frequency basis map to obtain a corresponding transformation coefficient matrix. The transformation coefficients of each block are quantized using the brightness matrix and chromaticity matrix from the prior art. Regions where the transformation coefficients are greater than a threshold or non-zero are defined as high-frequency signal regions. These regions often contain a large amount of detailed information and edge features, which are key to identifying smaller, more hidden flying animals. Interpolation amplification technology is used to locally amplify the high-frequency signal regions.

[0024] S300: Create a local target detection model and deploy it to a cloud server. Input the high-frequency signal area into the local target detection model, and output several flying animal instances and verification boxes. Compare the candidate boxes and verification boxes, calculate the intersection-union ratio, obtain the intersection and union, and insert the intersection into the flying animal instance.

[0025] A detection model dedicated to small-scale target recognition, namely the local target detection model, is constructed. Both the global target detection model and the local target detection model are constructed by the target detection model. The difference between the two lies in the training method and training data. The local target detection model is trained using the high-frequency signal area after interpolation and amplification, so that the local target detection model has stronger edge feature perception ability and small target recognition accuracy; the local target detection model is deployed to a cloud server, and the high-frequency signal area is input into the local target detection model, and several flying animal instances and verification boxes are output. The verification box is similar to the candidate box and refers to the location area that may contain flying animals.

[0026] Compare the candidate box and the verification box, define the overlapping area of ​​the two as the intersection, define the total area after the merger as the union, and calculate the intersection-in-union ratio, where the intersection-in-union ratio refers to the number of pixels in the intersection of the two divided by the number of pixels in the union; insert the instance in the intersection area into the flying animal instance.

[0027] S400: Integrate edge nodes and cloud servers to build a distributed processing architecture, upload the verification box to the cloud server, extract the fragment corresponding to the verification box from the image, generate a verification task, and when a detection request is received, issue the verification task, obtain the verification result, mark the flying animal instance and verification result in the image, and send it to the preset terminal.

[0028] Integrate edge nodes and cloud servers to build a distributed processing architecture with collaborative computing capabilities. In this processing architecture, edge nodes are mainly used to preliminarily detect flying animals in images, while cloud servers are used to undertake tasks such as brightness calculation, high-frequency area extraction, interpolation and amplification, candidate frame and verification frame comparison, and perform fine-grained detection on images. Furthermore, after completing the preliminary processing, the edge nodes will upload the detected suspected target areas and their corresponding verification frames to the cloud servers. The cloud servers have stronger computing resources and model libraries, which facilitate the deployment of more accurate local target detection models. Receive the verification frames from the edge nodes and perform deep recognition and intersection-comparison analysis on them. , target fusion and instance confirmation and other processing operations, and upload the processed verification box to the cloud server to identify the flying animal instance again; in the picture, the fragment corresponding to the verification box is cut out, and a verification task is generated, where the verification task refers to: integrating the fragments into the same thumbnail and writing the instructions: "Please find the flying animals in the following pictures" (similar to the picture verification code); after receiving the detection request, the verification task is sent to the user, and the user selects the flying animal from it. The user's selection result is also the verification result. All flying animal instances and verification results are marked in the picture and sent to the preset terminal, where the preset terminal is the user's device terminal.

[0029] In Example 2, Figure 2 The implementation process of the flying animal detection method provided by an embodiment of the present invention is shown. The following details the steps of tracing back to the source device, finding the edge node, and deploying the pre-built global target detection model to the edge node. S101: Build a scheduling platform for edge nodes and collect evaluation indicators of each edge node, where the evaluation indicators include at least delay and available bandwidth.

[0030] Build a scheduling platform, which is mainly used to uniformly manage and schedule all edge nodes. The scheduling platform continuously collects and dynamically updates the evaluation indicators of each edge node, including: communication delay, available bandwidth, node load, and CPU utilization of edge nodes.

[0031] S102: Configure the weight value of each evaluation indicator, collect the real-time value of the evaluation indicator, calculate the evaluation score, upload it to the scheduling platform, and dynamically update it.

[0032] According to the importance of each evaluation indicator, a corresponding weight value is configured. The weight value can be set according to the needs of the actual application scenario or formulated by the management personnel of the scheduling platform. The real-time value of the evaluation indicator of each edge node is collected, the real-time value of the evaluation indicator is multiplied by the weight value, and all the calculation results are superimposed to obtain the evaluation score. The evaluation score is marked in the scheduling platform and dynamically updated; the higher the evaluation score, the better the performance of the corresponding edge node. When deploying the global target detection model, edge nodes with higher evaluation scores are given priority.

[0033] In Example 3, different from Example 1, in this embodiment of the present invention, the method further includes: Based on the evaluation scores, edge nodes are clustered into common nodes and super nodes, and the global object detection model is deployed to several super nodes; Identify the load of each supernode and integrate the load balancing mechanism into the scheduling platform.

[0034] Based on the evaluation scores, edge nodes are divided into ordinary nodes and super nodes. Super nodes are edge nodes with higher evaluation scores mentioned above, and global target detection models are deployed in super nodes. Based on the load, multiple super nodes can also be selected, and a global target detection model is deployed in each super node. CPU usage, memory occupancy, task queue length and number of concurrent connections are used as load indicators to judge the load status of edge nodes. When the load of an edge node exceeds the threshold, new tasks will no longer be assigned first. A load balancing mechanism is integrated into the scheduling platform. The load balancing mechanism refers to assigning new tasks to edge nodes with smaller loads first. It should be noted that the task here refers to identifying flying animal instances using the global target detection model in the edge node.

[0035] In Example 4, Figure 3 The implementation process of the flying animal detection method provided by an embodiment of the present invention is shown. The following details the steps of dividing the image into a plurality of blocks, quantizing the blocks using a two-dimensional frequency basis map, and obtaining transform coefficients. S201: Set the basic range of each block and use the edge detection algorithm to traverse the edge pixels in the image.

[0036] Set the basic range of each block, where the basic range can be 8×8, and use the edge detection algorithm to determine the edge pixels in each image.

[0037] S202: Count the number of edge pixels in each basic range, and adjust the basic range based on the number, where the adjustment includes: expansion and contraction.

[0038] The number of edge pixels in each basic range is calculated. When the number is greater than the threshold, the basic range is reduced, that is, the basic range is adjusted to 4×4. Conversely, if the number is less than or equal to the threshold, the basic range is expanded to 16×16.

[0039] In this embodiment, by expanding the basic range, the complexity of data processing can be reduced and the processing efficiency can be improved. By narrowing the basic range, it is easier to capture local features and improve the richness of details.

[0040] In Example 5, Figure 4 The implementation process of the flying animal detection method provided by an embodiment of the present invention is shown. The steps of calculating the intersection-union ratio, obtaining the intersection and union, and inserting the intersection into the flying animal instance are described in detail below. S301: Calculate the intersection-over-union (IoU) of each image, select a first set value and a second set value, define images with an IoU less than the first set value as negative samples, and define images with an IoU greater than the second set value as positive samples.

[0041] In each picture, the candidate box and the verification box are determined, and the intersection-over-union (IoU) is calculated, where each picture corresponds to an IoU; based on the IoU, two set values ​​are selected, namely the first set value and the second set value. The picture with an IoU less than the first set value is defined as a negative sample, and the picture with an IoU greater than the second set value is defined as a positive sample.

[0042] For example, assuming the first setting value is 0.4 and the second setting value is 0.5, if the IoU of a certain image is less than 0.4, it means that the global target detection model or the local target detection model has a low recognition accuracy for this image and there are many false detections; if the IoU is greater than 0.5, it means that the model (the target detection model and the local target detection model are collectively referred to as the model) is more accurate in recognition. When the IoU of the image is between 0.4-0.5, the corresponding image is not used for training. In other words, using such images for training cannot improve the recognition ability of the model.

[0043] S302: Integrate positive samples and negative samples to generate a training set, and train the global object detection model and the local object detection model.

[0044] Use positive samples and negative samples to generate a training set and conduct training. In the actual model training process, positive samples need to be used as targets so that the model can learn "what is the target", and negative samples are also needed to train the model to identify "what is not the target".

[0045] In Example 6, Figure 5The implementation process of the flying animal detection method provided by an embodiment of the present invention is shown. The following details the steps of issuing a verification task and obtaining a verification result when a search request is received. S401: Setting a number of distribution channels for the verification task, and calculating a confidence score for each verification box based on the verification result.

[0046] Set up several distribution channels for verification tasks, each distribution channel represents a specific usage scenario or platform (such as web page, mobile terminal, third-party interface, etc.), and distribute the verification task to the user through the distribution channel; that is, use the fragment corresponding to the verification box to generate a thumbnail, and send it to the user in the form of a verification code; obtain the user's verification result, collect the user's behavior data during the verification process, and determine the confidence score of each verification box. For example, when the verification box is selected, add 1 point value to the confidence score of the corresponding verification box.

[0047] S402: Mark the verification results with confidence scores greater than a threshold in the image.

[0048] After multiple users have processed the verification code, the confidence score of each verification box is calculated, and the verification results with a confidence score greater than the threshold are marked in the image.

[0049] For example, an automatic surveillance camera (No. A) in a nature reserve collects image data at regular intervals every day and uploads it to the cloud platform. One day, administrator A initiates a detection request through the management system to find flying animals that have appeared in the reserve within the past 5 hours. At this time, A needs to verify by means of a verification code. The verification code is a thumbnail of the verification frame of the picture captured by other surveillance cameras or devices. It should be noted that the verification code is not composed of the verification frame of the picture captured by A. If the verification frame in the picture captured by A is not verified in time by means of the verification code when A initiates the detection request (the reason may be that manual verification resources are available, network congestion, etc.), then the user will be sent a picture with only flying animal instances marked (without the verification frame and verification result).

[0050] Figure 6 The structure block diagram of the flying animal detection system provided by an embodiment of the present invention is shown. The flying animal detection system 1 includes: The receiving module 11 is used to receive the search request uploaded by the user, extract the image to be detected, trace back the source device, find the edge node, deploy the pre-built global object detection model to the edge node, input the image into the global object detection model, and output a number of flying animal instances and candidate boxes; Amplification module 12 is used to calculate the brightness value of each pixel using a brightness formula, draw a two-dimensional frequency basis map composed of basis functions, divide the image into several blocks, quantize the blocks using the two-dimensional frequency basis map to obtain transformation coefficients, traverse the high-frequency signal area using a preset brightness matrix and chrominance matrix, and perform interpolation amplification; Insertion module 13 is used to create a local target detection model and deploy it to the cloud server. The high-frequency signal area is input into the local target detection model, and the output is a number of flying animal instances and verification boxes. The candidate boxes and verification boxes are compared, and the intersection-union ratio is calculated to obtain the intersection and union, and the intersection is inserted into the flying animal instance. The sending module 14 is used to integrate edge nodes and cloud servers to build a distributed processing architecture, upload the verification box to the cloud server, extract the fragment corresponding to the verification box from the image, generate a verification task, and when a search request is received, issue the verification task, obtain the verification result, mark the flying animal instance and the verification result in the image, and send it to the preset terminal.

[0051] Figure 7 The structure block diagram of the flying animal detection system provided by an embodiment of the present invention is shown. The receiving module 11 includes: The collection unit 111 is used to build a scheduling platform for edge nodes and collect evaluation indicators of each edge node, wherein the evaluation indicators include at least: delay and available bandwidth; The updating unit 112 is used to configure the weight value of each evaluation indicator, collect the real-time value of the evaluation indicator, calculate the evaluation score, upload it to the scheduling platform, and perform dynamic updates.

[0052] Figure 8 The structure block diagram of the flying animal detection system provided by an embodiment of the present invention is shown. The amplification module 12 includes: The traversal unit 121 is used to set the basic range of each block and use the edge detection algorithm to traverse the edge pixels in the image; The counting unit 122 is configured to count the number of edge pixels within each basic range, and adjust the basic range based on the number, wherein the adjustment includes: expansion and contraction.

[0053] Figure 9 The structure block diagram of the flying animal detection system provided by an embodiment of the present invention is shown. The insertion module 13 includes: A calculation unit 131 is configured to calculate an IoU ratio of each image, select a first set value and a second set value, define images with an IoU ratio less than the first set value as negative samples, and define images with an IoU ratio greater than the second set value as positive samples; The training unit 132 is used to integrate positive samples and negative samples to generate a training set and train the global object detection model and the local object detection model.

[0054] Figure 10 The structure block diagram of the flying animal detection system provided by an embodiment of the present invention is shown. The sending module 14 includes: A setting unit 141 is used to set a number of distribution channels for the verification task and calculate a confidence score for each verification box based on the verification result; The marking unit 142 is configured to mark the verification result with a confidence score greater than a threshold in the image.

[0055] The receiving module 11 is mainly used to complete step S100, the amplifying module 12 is mainly used to complete step S200, the inserting module 13 is mainly used to complete step S300, and the sending module 14 is mainly used to complete step S400; The acquisition unit 111 is mainly used to complete step S101, and the updating unit 112 is mainly used to complete step S102; The traversal unit 121 is mainly used to complete step S201, and the statistical unit 122 is mainly used to complete step S202; The calculation unit 131 is mainly used to complete step S301, and the training unit 132 is mainly used to complete step S302; The setting unit 141 is mainly used to complete step S401, and the marking unit 142 is mainly used to complete step S402.

[0056] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0057] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

[0058] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A flying animal detection method, characterized in that: The method comprises: Receive detection requests uploaded by users, extract images to be detected, trace back to the source device, find edge nodes, deploy the pre-built global object detection model to the edge nodes, input the images into the global object detection model, and output several flying animal instances and candidate boxes; Using the brightness formula, the brightness value of each pixel is calculated, and a two-dimensional frequency basis map composed of basis functions is drawn. The image is divided into several blocks, and the blocks are quantized using the two-dimensional frequency basis map to obtain the transformation coefficients. Using the preset brightness matrix and chromaticity matrix, the high-frequency signal area is traversed and interpolated and amplified; Create a local object detection model and deploy it to a cloud server. Input the high-frequency signal area into the local object detection model, and output several flying animal instances and verification boxes. Compare the candidate boxes and verification boxes, calculate the intersection-over-union ratio, obtain the intersection and union, and insert the intersection into the flying animal instance. Integrate edge nodes and cloud servers to build a distributed processing architecture, upload the verification box to the cloud server, extract the fragment corresponding to the verification box from the image, and generate a verification task. When a detection request is received, issue the verification task, obtain the verification result, mark the flying animal instance and verification result in the image, and send it to the preset terminal.

2. The flying animal detection method according to claim 1, characterized in that: The steps of tracing back the source device, finding the edge node, and deploying the pre-built global object detection model to the edge node include: Build a scheduling platform for edge nodes and collect evaluation indicators of each edge node, where the evaluation indicators include at least: latency and available bandwidth; Configure the weight value of each evaluation indicator, collect the real-time value of the evaluation indicator, calculate the evaluation score, upload it to the scheduling platform, and dynamically update it.

3. The flying animal detection method according to claim 2, characterized in that: The method further comprises: Based on the evaluation scores, edge nodes are clustered into common nodes and super nodes, and the global object detection model is deployed to several super nodes; Identify the load of each supernode and integrate the load balancing mechanism into the scheduling platform.

4. The flying animal detection method according to claim 1, characterized in that: The step of dividing the image into a plurality of blocks, quantizing the blocks using a two-dimensional frequency basis map, and obtaining transform coefficients includes: Set the basic range of each block and use the edge detection algorithm to traverse the edge pixels in the image; The number of edge pixels in each basic range is counted, and the basic range is adjusted based on the number, wherein the adjustment includes: expansion and contraction.

5. The flying animal detection method according to claim 1, characterized in that: The steps of calculating the intersection-union ratio, obtaining the intersection and the union, and inserting the intersection into the flying animal instance include: Calculate the intersection-over-union (IoU) of each image, select a first set value and a second set value, define images with an IoU less than the first set value as negative samples, and define images with an IoU greater than the second set value as positive samples; Integrate positive samples and negative samples to generate a training set, and train the global object detection model and the local object detection model.

6. The flying animal detection method according to claim 1, characterized in that: When a detection request is received, the steps of issuing a verification task and obtaining a verification result include: Setting up several distribution channels for the verification task, and calculating a confidence score for each verification box based on the verification results; The verification results with the confidence score greater than the threshold are marked in the image.

7. A flying animal detection system, characterized in that: The system comprises: The receiving module is used to receive detection requests uploaded by users, extract the images to be detected, trace back to the source device, find the edge node, deploy the pre-built global object detection model to the edge node, input the image into the global object detection model, and output several flying animal instances and candidate boxes; The amplification module is used to calculate the brightness value of each pixel using the brightness formula, draw a two-dimensional frequency basis map composed of basis functions, divide the image into several blocks, quantize the blocks using the two-dimensional frequency basis map, obtain the transformation coefficients, and use the preset brightness matrix and chrominance matrix to traverse the high-frequency signal area and perform interpolation amplification; The insertion module is used to create a local target detection model and deploy it to the cloud server. The high-frequency signal area is input into the local target detection model, and the output is several flying animal instances and verification boxes. The candidate boxes and verification boxes are compared, and the intersection-union ratio is calculated to obtain the intersection and union. The intersection is then inserted into the flying animal instance. The sending module is used to integrate edge nodes and cloud servers to build a distributed processing architecture. It uploads the verification box to the cloud server, extracts the fragment corresponding to the verification box from the image, and generates a verification task. When a detection request is received, it issues the verification task, obtains the verification result, marks the flying animal instance and the verification result in the image, and sends it to the preset terminal.

8. The flying animal detection system according to claim 7, characterized in that: The receiving module includes: A collection unit, configured to construct a scheduling platform for edge nodes and collect evaluation indicators of each edge node, wherein the evaluation indicators include at least latency and available bandwidth; The update unit is used to configure the weight value of each evaluation indicator, collect the real-time value of the evaluation indicator, calculate the evaluation score, upload it to the scheduling platform, and dynamically update it.

9. The flying animal detection system according to claim 7, characterized in that: The amplification module includes: The traversal unit is used to set the basic range of each block and use the edge detection algorithm to traverse the edge pixels in the image; The counting unit is used to count the number of edge pixels in each basic range, and adjust the basic range based on the number, wherein the adjustment includes: expansion and contraction.

10. The flying animal detection system according to claim 7, characterized in that: The insertion module comprises: A calculation unit is configured to calculate an IoU ratio of each image, select a first set value and a second set value, define an image with an IoU ratio less than the first set value as a negative sample, and define an image with an IoU ratio greater than the second set value as a positive sample; The training unit is used to integrate positive samples and negative samples to generate a training set and train the global target detection model and the local target detection model.

Images