Bird target detection method and device, computer equipment and storage medium

Abstract

The invention discloses a bird target detection method and device, computer equipment and a storage medium, and belongs to the technical field of target detection. The method comprises the following steps: preprocessing a collected bird image data set, and dividing the preprocessed bird image data set into a training set, a verification set and a test set; an SPPF module in a backbone network of the YOLOv8 model is replaced by a SimSPPF module, and an iRMB module is added in front of the SimSPPF module; training the YOLOv8 model based on the training set, and evaluating the YOLOv8 model in the training process based on the verification set to obtain a trained YOLOv8 model; and inputting the test set into the trained YOLOv8 model for bird target detection to obtain a detection result. By improving the YOLOv8 model, richer feature representation can be learned, so that accurate recognition of bird types, positions and confidence is improved.

Description

Technical Field

[0001] The present invention relates to the field of target detection technology, and more specifically, to a bird target detection method, device, computer equipment and storage medium. Background Art

[0002] With the increasing development of modern agricultural production, agricultural technology plays an increasingly important role in improving production efficiency, reducing costs, and ensuring the quality of agricultural products. Among them, the rapid development of computer vision and deep learning technology has brought new breakthroughs in the agricultural field. In the field of aquaculture, bird interference is a common problem, which may cause aquatic product losses and affect the stable development of the aquaculture industry. Therefore, it is of great significance to use advanced deep learning target detection technology to solve this problem.

[0003] Deep learning target detection technology refers to the use of deep neural networks to analyze images or videos, automatically identify the targets of interest and mark their locations and bounding boxes. Compared with traditional image processing methods, deep learning technology has higher accuracy and robustness, can handle complex and changing scenes, and achieve real-time target detection and recognition. The introduction of this technology can not only improve production efficiency, but also reduce human resource input, thereby reducing breeding costs and improving the overall benefits of agricultural production.

[0004] In open-air aquaculture, deep learning target detection technology can be used to monitor birds in the water in real time and detect their behaviors, such as eating or staying. Once birds are found entering the breeding area, relevant personnel can take immediate measures, such as making sounds or using driving tools such as drones to drive the birds away, thereby effectively preventing them from preying on aquatic products. This can not only protect aquatic resources, but also maintain the normal operation of the breeding industry and ensure the stability and sustainable development of production.

[0005] After searching, the Chinese patent application, application number 202410658497.X, published on August 9, 2024, discloses a bird target detection method based on deep learning. The method includes: building a bird detection model based on the YOLOv8 target detection network, obtaining a bird public data set and then training the bird detection model through the training set; introducing the BiFPN module to enhance the feature fusion capability of the Neck network module; at the same time, the Deformable Attention mechanism is introduced to enhance the feature information extraction capability of the Backbone network module; finally, the deep separable convolution is introduced to greatly reduce the number of model parameters to achieve lightweight. The model in this method is more lightweight, so it can be more easily deployed in mobile devices to achieve rapid detection of real-time scenes, but the detection accuracy of bird recognition targets in this method is low. Summary of the invention

[0006] 1. Technical problems to be solved

[0007] In view of the problems in the prior art that the detection speed of bird identification targets based on the YOLOv8 network structure is low and the detection accuracy is not high, the present invention provides a bird target detection method, device, computer equipment and storage medium, which can effectively improve the detection accuracy of bird identification targets without significantly reducing the detection speed.

[0008] 2. Technical solution

[0009] The purpose of the present invention is achieved through the following technical solutions.

[0010] A bird target detection method comprises the following steps:

[0011] Collect bird image datasets, preprocess the bird image datasets, and divide the preprocessed bird image datasets into training sets, validation sets, and test sets;

[0012] Improve the YOLOv8 model. The YOLOv8 model includes a backbone network, a neck network, and a head network. The SPPF module in the backbone network is replaced with a SimSPPF module, and an iRMB module is added before the SimSPPF module.

[0013] Train the YOLOv8 model based on the training set, evaluate the YOLOv8 model during the training process based on the validation set, and obtain the trained YOLOv8 model;

[0014] The test set is input into the trained YOLOv8 model to detect bird targets and obtain the detection results.

[0015] Further, the backbone network includes a Conv layer, a Conv layer, a C2f module, a Conv layer, a C2f module, a Conv layer, a C2f module, a Conv layer, an iRMB module and a SimSPPF module connected in sequence;

[0016] Inputting a bird image into the backbone network, the bird image is sequentially subjected to a Conv layer, a Conv layer, a C2f module, a Conv layer, and a C2f module for feature extraction to obtain a first feature map;

[0017] The first feature map is sequentially subjected to feature extraction by the Conv layer and the C2f module to obtain a second feature map;

[0018] The second feature map is subjected to feature extraction by a Conv layer to obtain a shallow feature map.

[0019] Further, the iRMB module includes a Conv layer, a DW-Conv layer and a feature modulation mechanism;

[0020] The feature modulation mechanism obtains deep feature information in the shallow feature map by weighting and adjusting the number of channels of the shallow feature map;

[0021] The Conv layer and the DW-Conv layer extract deep feature information from the shallow feature map to obtain a deep feature map.

[0022] Further, the SimSPPF module includes a first Upsample layer, a first Concat layer, a first C2f module, a second Upsample layer, a second Concat layer, a second C2f module, a first Conv layer, a third Concat layer, a third C2f module, a second Conv layer, a fourth Concat layer and a fourth C2f module.

[0023] Furthermore, the deep feature map is input into the first Upsample layer, the first Upsample layer enlarges the resolution of the deep feature map so that the resolution of the deep feature map is the same as that of the second feature map; the first Concat layer splices the second feature map and the deep feature map with the enlarged resolution; the first C2f module extracts features and reduces the dimension of the spliced ​​feature map to obtain a third feature map; the second Upsample layer enlarges the size of the third feature map; the second Concat layer splices the first feature map and the enlarged third feature map to obtain a semantic feature map; the second C2f module extracts features and reduces the dimension of the semantic feature map to obtain a fourth feature map.

[0024] Further, the first Conv layer performs a convolution operation on the fourth feature map; the third Concat layer concatenates the third feature map with the fourth feature map after convolution to obtain an upsampled feature map; the third C2f module performs feature extraction and dimensionality reduction on the upsampled feature map to obtain a fifth feature map; the second Conv layer performs a convolution operation on the fifth feature map; the fourth Concat layer concatenates the deep feature map with the fifth feature map after convolution to obtain a global feature map; the fourth C2f module performs feature extraction and dimensionality reduction on the global feature map to obtain a sixth feature map.

[0025] Furthermore, the sixth feature map is feature fused through the neck network to obtain a multi-scale feature map; the head network performs bird target detection on the multi-scale feature map to obtain a detection result.

[0026] A bird target detection device, comprising:

[0027] A data processing module collects bird image data sets, preprocesses the bird image data sets, and divides the preprocessed bird image data sets into a training set, a validation set, and a test set;

[0028] Model improvement module, improves the YOLOv8 model, the YOLOv8 model includes the backbone network, the neck network and the head network; the SPPF module in the backbone network is replaced with the SimSPPF module, and the iRMB module is added before the SimSPPF module;

[0029] Model training module: trains the YOLOv8 model based on the training set, evaluates the YOLOv8 model during training based on the validation set, and obtains the trained YOLOv8 model;

[0030] The model testing module inputs the test set into the trained YOLOv8 model to detect bird targets and obtain the detection results.

[0031] A computer device comprises a memory and a processor, wherein the memory stores a computer program executable on the processor, and the processor implements any of the above-mentioned methods when executing the computer program.

[0032] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, any of the above methods is executed.

[0033] 3. Beneficial effects

[0034] Compared with the prior art, the advantages of the present invention are:

[0035] (1) A bird target detection method, apparatus, computer device and storage medium of the present invention replace the SPPF module of the backbone network in the YOLOv8 model with a SimSPPF module to capture different scale features and global long-term dependencies, thereby obtaining a comprehensive and discriminative feature representation to enhance the discrimination and generalization capabilities of the YOLOv8 model.

[0036] (2) A bird target detection method, device, computer equipment and storage medium of the present invention, by introducing the iRMB reverse residual attention module, enhances the attention and accuracy of the YOLOv8 model when learning bird features, can more effectively focus on bird features, and improve the accuracy of bird recognition. The gradient vanishing problem in the deep network of the YOLOv8 model is avoided by the improved residual connection, so that the feature information can flow effectively between network layers. A feature modulation mechanism is also added to the iRMB module, which helps the YOLOv8 model focus on important features and suppress irrelevant features by adaptively weighting and adjusting the channels of the feature map, thereby enhancing the representation ability and robustness of the YOLOv8 model. Finally, the convolutional layer contained in the iRMB module further extracts and refines features to enhance the spatial structure modeling capability of the YOLOv8 model.

[0037] (3) The bird target detection method, device, computer equipment and storage medium of the present invention improve the YOLOv8 model, and can achieve better results for bird target detection by using only a lightweight detector. The use of a detector with a simpler structure in the bird target detection process can greatly reduce the resources required for calculation, improve the detection speed of the YOLOv8 model and reduce the memory usage required in the detection process, avoid unnecessary waste, and effectively utilize limited computing resources, thereby realizing the practical application of the YOLOv8 model in bird target detection, and having strong practicality and wide applicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flow chart of a bird target detection method according to an embodiment of the present invention;

[0039] Figure 2 A schematic diagram of the structure of the improved YOLOv8 model according to an embodiment of the present invention;

[0040] Figure 3 This is a schematic diagram of the structure of an iRMB module according to an embodiment of the present invention;

[0041] Figure 4 It is a structural diagram of the SimSPPF module of an embodiment of the present invention;

[0042] Figure 5 This is a performance indicator diagram of the improved YOLOv8 model according to an embodiment of the present invention;

[0043] Figure 6 This is a diagram of bird target detection results according to an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0045] Example

[0046] like Figure 1 As shown, a bird target detection method provided by this embodiment includes the following steps: collecting a bird image data set, preprocessing the bird image data set, and dividing the preprocessed bird image data set into a training set, a verification set, and a test set; improving the YOLOv8 model, the YOLOv8 model includes a backbone network, a neck network, and a head network; wherein the SPPF module in the backbone network is replaced with a SimSPPF module, and an iRMB module is added before the SimSPPF module; the YOLOv8 model is trained based on the training set, and the YOLOv8 model in the training process is evaluated based on the verification set to obtain a trained YOLOv8 model; the test set is input into the trained YOLOv8 model to perform bird target detection and obtain a detection result.

[0047] Specifically in this embodiment, a dataset of bird images commonly seen in fish ponds is collected, the bird image dataset is preprocessed, and converted into a training format for the YOLO algorithm. In this embodiment, the preprocessing includes labeling and classifying the bird image dataset using the existing labeling software LabelImg, that is, creating a rectangular frame, labeling and classifying the targets in the bird image using the rectangular frame. Then, the preprocessed bird image dataset is divided into a training set, a validation set, and a test set.

[0048] Further, the YOLOv8 model is improved, and the YOLOv8 model includes a backbone network, a neck network, and a head network. In this embodiment, the SPPF module in the backbone network is replaced with a SimSPPF module, and an iRMB (Improved Residual Modulation Block) module is added before the SimSPPF module. In this embodiment, the backbone network is used for feature extraction, the neck network is used for feature fusion, and the head network is used for bird target detection output.

[0049] like Figure 2 As shown in the figure, in the backbone network of the YOLOv8 model, the original SPPF module is replaced by the SimSPPF module, which serves as a new spatial pyramid pooling layer. At the same time, the iRMB module is introduced, so that the YOLOv8 model can focus on bird features more effectively and improve the accuracy of bird recognition.

[0050] In this embodiment, the backbone network is composed of a series of convolutional layers and custom modules. Specifically, the backbone network includes a Conv layer, a Conv layer, a C2f module, a Conv layer, a C2f module, a Conv layer, a C2f module, a Conv layer, an iRMB module, and a SimSPPF module connected in sequence.

[0051] It should be noted that, in this embodiment, the Conv layers have different numbers of output channels and strides, respectively, to gradually reduce the size of the feature map. A series of repeated custom modules, including the C2f module and the iRMB module, are used to further extract the features of the bird images and retain the residual connection. Finally, the SimSPPF module is used to capture features of different scales in the bird images. By improving the backbone network of the YOLOv8 model, multi-level and multi-scale features can be effectively extracted to support subsequent bird target detection tasks.

[0052] It is worth noting that, in this embodiment, Figure 3 As shown, the iRMB module is an enhancement module for deep neural networks, which aims to improve the efficiency of feature learning and information transfer. The gradient vanishing problem in the deep network is avoided by introducing an improved residual connection, so that the feature information can flow effectively between network layers. In this embodiment, the iRMB module includes a Conv layer and a DW-Conv layer. The Conv layer and DW-Conv layer contained in the iRMB module are used to further extract and refine features and enhance the spatial structure modeling capability of the YOLOv8 model. In particular, a feature modulation mechanism is also introduced in the iRMB module, which helps the backbone network focus on important features and suppress irrelevant features by adaptively weighting and adjusting the channels of the feature map, thereby enhancing the representation ability and robustness of the YOLOv8 model.

[0053] It is worth noting that, in this embodiment, Figure 4 As shown, the SimSPPF module combines convolution operations, pooling and feature fusion to extract multi-scale features. By multiple pooling and splicing feature maps of different scales together, the SimSPPF module provides richer spatial information, so that the YOLOv8 model can better detect targets at different scales. Specifically, the SimSPPF module includes a first Upsample layer, a first Concat layer, a first C2f module, a second Upsample layer, a second Concat layer, a second C2f module, a first Conv layer, a third Concat layer, a third C2f module, a second Conv layer, a fourth Concat layer and a fourth C2f module. Thus, in this embodiment, after the SimSPPF module processes the feature information in the feature map in layers, the extracted feature information is input into the neck network for feature fusion. In this embodiment, the difference between the improved SimSPPF module and the existing SPPF module is that the global long-term dependencies can be captured, thereby obtaining a comprehensive and discriminative feature representation.

[0054] Therefore, in this embodiment, the bird image is input into the backbone network of the YOLOv8 model, and the bird image is sequentially subjected to the Conv layer, the Conv layer, the C2f module, the Conv layer, and the C2f module for feature extraction to obtain a first feature map. The first feature map is sequentially subjected to the Conv layer and the C2f module for feature extraction to obtain a second feature map, and at this time, the number of channels of the second feature map is 512 dimensions. The second feature map is then subjected to feature extraction by the Conv layer to obtain a shallow feature map.

[0055] Specifically, the first Conv layer has 64 output channels, a kernel size of 3, and a stride of 2. This Conv layer downsamples the input bird image to 1 / 2 of the original. Then, the backbone network downsamples through the second Conv layer, which has 128 output channels, a kernel size of 3, and a stride of 2. This Conv layer downsamples the feature map to 1 / 4 of the original. Then, the backbone network passes through the first C2f module, which has 128 output channels, repeats the feature extraction operation 3 times in the first C2f module, and gradually fuses and extracts more features by applying the C2f module multiple times, thereby improving the expressiveness of the features. The feature map is convolved through the first C2f module to extract higher-level features. Next, the backbone network downsamples through the third Conv layer, which has 256 output channels, a kernel size of 3, and a stride of 2. This Conv layer downsamples the feature map to 1 / 8 of the original. Then, the backbone network passes through the second C2f module, which has 256 output channels and repeats the feature extraction operation three times in the second C2f module. The second C2f module performs a convolution operation on the feature map to extract higher-level features and obtain the first feature map.

[0056] The backbone network is downsampled through the fourth Conv layer, which has 512 output channels, a convolution kernel size of 3, and a stride of 2. This Conv layer downsamples the feature map to 1 / 16 of the original. Then, the backbone network passes through the third C2f module, which has 512 output channels and repeats the feature extraction operation 6 times in the third C2f module. The third C2f module convolves the feature map to extract higher-level features and obtain the second feature map.

[0057] The backbone network is downsampled through the fifth Conv layer, which has 1024 output channels, a convolution kernel size of 3, and a stride of 2. This Conv layer downsamples the feature map to 1 / 32 of the original, thereby obtaining a shallow feature map.

[0058] The feature modulation mechanism in the iRMB module of the backbone network obtains the deep feature information in the shallow feature map by weighting and adjusting the number of channels of the shallow feature map. Furthermore, the Conv layer and the DW-Conv layer extract the deep feature information in the shallow feature map to obtain the deep feature map. In this embodiment, the iRMB module has 1024 output channels.

[0059] Furthermore, the backbone network passes through the SimSPPF module, which has 1024 output channels and uses 5 pooling layers of different sizes. The SimSPPF module can extract features at different scales, including global information features, local information features, and spatial structure information features.

[0060] Specifically, the deep feature map is input into the first Upsample layer, and the first Upsample layer amplifies the resolution of the deep feature map so that the resolution of the deep feature map is the same as that of the second feature map, in preparation for subsequent feature fusion. The first Concat layer splices the second feature map and the deep feature map with amplified resolution. Through the splicing method, the backbone network can obtain higher resolution features and enhance the semantic information of the YOLOv8 model. The spliced ​​feature map is 512 dimensions. The first C2f module extracts features and reduces the dimension of the spliced ​​feature map, thereby obtaining the third feature map, making the backbone network more lightweight and more computationally efficient.

[0061] The second Upsample layer enlarges the size of the third feature map without increasing the amount of calculation, preparing for subsequent feature fusion and providing more feature information for more precise target location. The second Concat layer concatenates the first feature map and the enlarged third feature map to obtain a semantic feature map, further enriching the semantic information and enhancing the YOLOv8 model's ability to represent the target. The second C2f module extracts features and reduces the dimension of the semantic feature map, thereby obtaining the fourth feature map to enhance the discrimination and generalization capabilities of the YOLOv8 model.

[0062] The first Conv layer performs convolution operation on the fourth feature map and outputs a 512-dimensional feature map with a convolution kernel size of 3 and a step size of 2, thereby further increasing the abstraction and representation capabilities of the feature map. The third Concat layer concatenates the third feature map with the fourth feature map after convolution to obtain an upsampled feature map, completes the fusion of feature maps of different scales, and improves the target detection accuracy of the YOLOv8 model. The third C2f module extracts features and reduces the dimension of the upsampled feature map to obtain the fifth feature map.

[0063] The second Conv layer performs convolution operation on the fifth feature map and outputs a 1024-dimensional feature map. The convolution kernel size is 3 and the step size is 2, which further enhances the feature expression ability of the YOLOv8 model. The fourth Concat layer concatenates the deep feature map with the fifth feature map after convolution to obtain a global feature map. The fourth C2f module extracts features and reduces the dimension of the global feature map to obtain the sixth feature map, making the YOLOv8 model more generalizable and adaptable.

[0064] It can be seen that the SimSPPF module has the characteristics of intra-layer adjustment and top-down global adjustment, which can obtain more comprehensive and discriminative feature representations.

[0065] The sixth feature map is subjected to feature fusion by the neck network to obtain a multi-scale feature map. In this embodiment, the head network performs bird target detection on the multi-scale feature map to obtain a detection result.

[0066] Specifically, the fourth feature map is input into the first detection head (Detect) of the head network for target detection to obtain primary detection results, including bird species, position and confidence. The fifth feature map is input into the second detection head of the head network for target detection to obtain intermediate detection results, including bird species, position and confidence. The multi-scale feature map is input into the third detection head of the head network for target detection to obtain the final detection result, including bird species, position and confidence. Therefore, in this embodiment, by comparing the primary detection results, the intermediate detection results and the final detection results, it can be seen that by replacing the SimSPPF module, it helps the backbone network learn richer feature representations, thereby significantly improving the accurate recognition of bird species, positions and confidences, making the YOLOv8 model more practical.

[0067] In this embodiment, the YOLOv8 model is trained based on the training set, and the YOLOv8 model in the training process is evaluated based on the verification set to obtain a trained YOLOv8 model. Specifically, the improved YOLOv8 model is trained, and the number of training times and hyperparameters are set according to the target and hardware performance to achieve accurate recognition of the target bird species, position and confidence. As the number of training times increases, the loss function of the training YOLOv8 model will gradually converge. When the loss function tends to be stable, the optimal YOLOv8 model is obtained and the weight parameters are generated. The improved YOLOv8 model loads the weight parameters, and the YOLOv8 model in the training process is evaluated based on the verification set to obtain a trained YOLOv8 model.

[0068] In this embodiment, the head network includes regression loss and classification loss. The regression loss function is expressed as:

[0069] L IoU=1-IoU

[0070] Among them, IoU represents the overlap between the predicted box and the real box, L IoU Represents regression loss. It should be noted that, in this embodiment, the IoU (Intersection over Union) loss is used to measure the overlap between the predicted box and the true box. IoU is a very important indicator in object detection, which is used to measure the ratio of the intersection and union of the predicted box and the true box.

[0071] The classification loss function is expressed as:

[0072]

[0073] Among them, i represents a natural number, C represents, y i Indicates that p i Indicates that L cls (p, y) represents the classification loss. In this embodiment, the classification loss receives the output features from the second layer MaxPool2d, the third layer MaxPool2d and the fourth layer MaxPool2d of the SimSPPF module, and finally outputs the detection result after processing.

[0074] like Figure 5 As shown in the figure, the YOLOv8 model aims to accurately identify bird species, positions, and confidence levels during training. By reasonably setting the number of training times and hyperparameters and combining hardware performance, the training efficiency and YOLOv8 model performance are balanced. As the number of training times increases, the YOLOv8 model will gradually learn more refined feature representations. Among them, the iRMB module can effectively capture key information in bird images and promote the expression ability of the YOLOv8 model. The loss function will gradually converge to stability. After reaching a stable state, the YOLOv8 model will reach the optimal state, and the generated weight parameters have high recognition accuracy and generalization ability, and can accurately identify bird species, positions, and confidence levels.

[0075] like Figure 6 As shown in the figure, the test set is input into the trained YOLOv8 model to detect bird targets and obtain the detection results. Specifically, the target bird species, location and confidence are annotated through the test set and the detection results are output. According to the detection results and the report generated by the training, it is judged whether the detection accuracy meets the expectations.

[0076] Therefore, the bird target detection method provided in this embodiment, by improving the YOLOv8 model, can achieve better results for bird target detection by using only a lightweight detector. Using a detector with a simpler structure in the bird target detection process can greatly reduce the resources required for calculation, improve the detection speed of the YOLOv8 model and reduce the video memory occupancy required in the detection process, avoid unnecessary waste, and effectively utilize limited computing resources, thereby realizing the practical application of the YOLOv8 model in bird target detection, which has strong practicality and wide applicability.

[0077] The present embodiment also provides a bird target detection device, including a data processing module, a model improvement module, a model training module and a model testing module. The data processing module is used to collect a bird image data set, preprocess the bird image data set, and divide the preprocessed bird image data set into a training set, a verification set and a test set. The model improvement module is used to improve the YOLOv8 model, and the YOLOv8 model includes a backbone network, a neck network and a head network; wherein the SPPF module in the backbone network is replaced with a SimSPPF module, and an iRMB module is added before the SimSPPF module. The model training module trains the YOLOv8 model based on the training set, and evaluates the YOLOv8 model in the training process based on the verification set to obtain a trained YOLOv8 model. The model testing module inputs the test set into the trained YOLOv8 model model to perform bird target detection and obtain a detection result. The bird target detection device provided in the present embodiment can implement any of the bird target detection methods, and the specific working process of a bird target detection device can refer to the corresponding process in the embodiment of the bird target detection method. The method and device provided in this embodiment can be implemented in other ways. For example, the device embodiment described above is only schematic; for example, the division of a module is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual connection or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, or it can be an electrical, mechanical or other form of connection.

[0078] This embodiment also provides a computer device. A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the bird target detection method when executing the computer program.

[0079] This embodiment also provides a computer-readable storage medium. A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, a bird target detection method described in this embodiment is executed. The computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, device or device; the program code contained on the computer-readable medium can be transmitted using any appropriate medium, including but not limited to wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0080] The above schematically describes the invention and its implementation methods, which is not restrictive. Without departing from the spirit or basic features of the invention, the invention can be implemented in other specific forms. What is shown in the accompanying drawings is only one of the implementation methods of the invention. The actual structure is not limited thereto, and any figure mark in the claims should not limit the claims involved. Therefore, if a person of ordinary skill in the art is inspired by it, without departing from the purpose of the invention, a structural method and an embodiment similar to the technical solution are designed without creativity, which should all belong to the protection scope of the present invention. In addition, the word "including" does not exclude other elements or steps, and the word "one" before the element does not exclude the inclusion of "multiple" elements. The multiple elements stated in the product claim can also be implemented by one element through software or hardware. The words first, second, etc. are used to indicate the name, and do not indicate any specific order.

Claims

1. A bird target detection method, comprising the following steps: Collect bird image datasets, preprocess the bird image datasets, and divide the preprocessed bird image datasets into training sets, validation sets, and test sets; Improve the YOLOv8 model. The YOLOv8 model includes a backbone network, a neck network, and a head network. The SPPF module in the backbone network is replaced with a SimSPPF module, and an iRMB module is added before the SimSPPF module. Train the YOLOv8 model based on the training set, evaluate the YOLOv8 model during training based on the validation set, and obtain the trained YOLOv8 model; The test set is input into the trained YOLOv8 model to detect bird targets and obtain the detection results.

2. A bird target detection method according to claim 1, characterized in that: The backbone network includes a Conv layer, a Conv layer, a C2f module, a Conv layer, a C2f module, a Conv layer, a C2f module, a Conv layer, an iRMB module and a SimSPPF module connected in sequence; Inputting a bird image into the backbone network, the bird image is sequentially subjected to a Conv layer, a Conv layer, a C2f module, a Conv layer, and a C2f module for feature extraction to obtain a first feature map; The first feature map is sequentially subjected to feature extraction by the Conv layer and the C2f module to obtain a second feature map; The second feature map is subjected to feature extraction by a Conv layer to obtain a shallow feature map.

3. A bird target detection method according to claim 2, characterized in that: The iRMB module includes a Conv layer, a DW-Conv layer and a feature modulation mechanism; The feature modulation mechanism obtains deep feature information in the shallow feature map by weighting and adjusting the number of channels of the shallow feature map; The Conv layer and the DW-Conv layer extract deep feature information from the shallow feature map to obtain a deep feature map.

4. A bird target detection method according to claim 3, characterized in that: The SimSPPF module includes a first Upsample layer, a first Concat layer, a first C2f module, a second Upsample layer, a second Concat layer, a second C2f module, a first Conv layer, a third Concat layer, a third C2f module, a second Conv layer, a fourth Concat layer and a fourth C2f module.

5. A bird target detection method according to claim 4, characterized in that: The deep feature map is input into the first Upsample layer, and the first Upsample layer enlarges the resolution of the deep feature map so that the resolution of the deep feature map is the same as that of the second feature map; the first Concat layer splices the second feature map and the deep feature map with the enlarged resolution; the first C2f module extracts features and reduces the dimension of the spliced ​​feature map to obtain a third feature map; the second Upsample layer enlarges the size of the third feature map; the second Concat layer splices the first feature map and the enlarged third feature map to obtain a semantic feature map; the second C2f module extracts features and reduces the dimension of the semantic feature map to obtain a fourth feature map.

6. A bird target detection method according to claim 5, characterized in that: The first Conv layer performs a convolution operation on the fourth feature map; the third Concat layer concatenates the third feature map with the fourth feature map after convolution to obtain an upsampled feature map; the third C2f module performs feature extraction and dimensionality reduction on the upsampled feature map to obtain a fifth feature map; the second Conv layer performs a convolution operation on the fifth feature map; the fourth Concat layer concatenates the deep feature map with the fifth feature map after convolution to obtain a global feature map; the fourth C2f module performs feature extraction and dimensionality reduction on the global feature map to obtain a sixth feature map.

7. A bird target detection method according to claim 6, characterized in that: The sixth feature map is subjected to feature fusion through the neck network to obtain a multi-scale feature map; the head network performs bird target detection on the multi-scale feature map to obtain a detection result.

8. A bird target detection device, characterized in that: include: A data processing module collects bird image data sets, preprocesses the bird image data sets, and divides the preprocessed bird image data sets into a training set, a validation set, and a test set; Model improvement module, improves the YOLOv8 model, the YOLOv8 model includes the backbone network, the neck network and the head network; the SPPF module in the backbone network is replaced with the SimSPPF module, and the iRMB module is added before the SimSPPF module; Model training module: trains the YOLOv8 model based on the training set, evaluates the YOLOv8 model during training based on the validation set, and obtains the trained YOLOv8 model; The model testing module inputs the test set into the trained YOLOv8 model to detect bird targets and obtain the detection results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is executed.

Images