Sea turtle instance segmentation method and system based on weight low-rank decomposition and feature enhancement

Abstract

The invention belongs to the technical field of image processing, and particularly relates to a sea turtle instance segmentation method and system based on weight low-rank decomposition and feature enhancement, and the method comprises the steps: obtaining an underwater sea turtle image; constructing an image encoder based on a weight low-rank decomposition and updating mechanism, and capturing global features of the obtained sea turtle image according to the image encoder and a multi-head self-attention mechanism; performing feature selective enhancement on the obtained global features of the sea turtle image to obtain a multi-scale reconstructed feature map; and generating a candidate region on the obtained multi-scale reconstructed feature map, aligning the multi-scale reconstructed feature map on the generated candidate region, and detecting and segmenting the aligned multi-scale reconstructed feature map through a multi-task branch network to complete the segmentation of the sea turtle instance.

Description

Technical Field

[0001] The present invention belongs to the field of image processing technology, and in particular relates to a turtle instance segmentation method and system based on weighted low-rank decomposition and feature enhancement. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Chelonium is an endangered species. Sea turtle instance segmentation can be used to investigate and sample sea turtle diversity, furthering conservation research. However, sea turtles live in a complex underwater environment. The water medium and suspended particles significantly absorb and reflect light, resulting in low contrast and blurred edges in underwater sea turtle images, which seriously affects the feature representation of sea turtle images.

[0004] Currently, traditional network models are unable to selectively enhance weak turtle features when performing underwater turtle target segmentation, which will greatly reduce the accuracy of turtle instance segmentation. At the same time, directly updating large weight matrices during model training has low training efficiency and weak model adaptability, making it impossible to quickly and accurately complete turtle instance segmentation in complex underwater environments. Summary of the Invention

[0005] To solve the above problems, the present invention proposes a turtle instance segmentation method and system based on weighted low-rank decomposition and feature enhancement. By designing a weighted low-rank decomposition and update mechanism, the adaptability and training efficiency of the network are improved, and an innovative turtle feature selective enhancement module is designed to effectively enhance the network's perception of fuzzy image features and improve the segmentation accuracy of underwater turtles.

[0006] According to some embodiments, a first solution of the present invention provides a turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement, which adopts the following technical solutions: A turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement, including: Acquire underwater turtle images; An image encoder is constructed based on a weighted low-rank decomposition and update mechanism. The global features of the acquired turtle images are captured using the image encoder and multi-head self-attention mechanism. The obtained global features of the turtle image are selectively enhanced to obtain a multi-scale reconstructed feature map; Candidate regions are generated on the obtained multi-scale reconstructed feature map, and the multi-scale reconstructed feature map is aligned on the generated candidate regions. The aligned multi-scale reconstructed feature map is detected and segmented through a multi-task branch network to complete the turtle instance segmentation.

[0007] As a further technical limitation, the constructed image encoder is equipped with a weighted low-rank decomposition update module; in the weighted low-rank decomposition update module, the weights are low-rank decomposed to obtain the decomposed low-rank matrix through the weighted low-rank decomposition and update mechanism, and the update is completed.

[0008] As a further technical limitation, in feature selective enhancement, a turtle image feature map is obtained based on the global features and selectivity coefficients of the turtle image, the obtained turtle image feature map is iterated layer by layer to obtain a multi-level feature map, and the obtained multi-level feature maps are fused to obtain a fused feature map; convolution and pooling are used to reconstruct the obtained fused feature map at different scales to obtain a multi-scale reconstructed feature map.

[0009] As a further technical limitation, the turtle instance segmentation includes a candidate region generation stage, an alignment stage, and a multi-task prediction stage; wherein, in the candidate region generation stage, the region generation network introduces an anchor box of preset size on the feature point, and based on the confidence of the anchor box prediction target and the regression offset of the bounding box, non-maximum suppression is used to screen all candidate boxes to obtain candidate regions, thereby completing the candidate region generation stage.

[0010] Furthermore, in the alignment stage, bilinear interpolation is used to obtain feature values ​​at the sampling positions, and the maximum pooling operation is combined to extract regional features of fixed size; the aligned features are sent to the multi-task branch network to complete the detection and segmentation tasks of the scale-reconstructed feature map respectively.

[0011] As a further technical limitation, the loss function of the turtle instance segmentation includes a region proposal network loss function, a detection head classification loss function, a detection head bounding box regression loss function and a segmentation mask loss function.

[0012] According to some embodiments, a second solution of the present invention provides a turtle instance segmentation system based on weighted low-rank decomposition and feature enhancement, which adopts the following technical solutions: A turtle instance segmentation system based on weighted low-rank decomposition and feature enhancement, including: An image acquisition module, fully configured to acquire underwater sea turtle images; The feature capture module is fully configured to build an image encoder based on a weighted low-rank decomposition and update mechanism, and to capture the global features of the acquired turtle images using the image encoder and the multi-head self-attention mechanism; A feature enhancement module is configured to selectively enhance the obtained global features of the turtle image to obtain a multi-scale reconstructed feature map; The instance segmentation module is configured to generate candidate regions on the obtained multi-scale reconstructed feature map, align the multi-scale reconstructed feature map on the generated candidate regions, and perform detection and segmentation of the aligned multi-scale reconstructed feature map through a multi-task branch network to complete turtle instance segmentation.

[0013] According to some embodiments, a third solution of the present invention provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium stores a program thereon, which, when executed by a processor, implements the steps of the turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement as described in the first embodiment of the present invention.

[0014] According to some embodiments, a fourth solution of the present invention provides an electronic device, which adopts the following technical solution: An electronic device comprises a memory, a processor, and a program stored in the memory and running on the processor. When the processor executes the program, the steps of the turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement as described in the first embodiment of the present invention are implemented.

[0015] According to some embodiments, a fifth solution of the present invention provides a computer program product, which adopts the following technical solution: A computer program product includes software code, wherein the program in the software code executes the steps in the turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement as described in the first embodiment of the present invention.

[0016] Compared with the prior art, the present invention has the following beneficial effects: This paper improves the adaptability and training efficiency of the network by designing a low-rank decomposition and update mechanism for weights, and innovatively designs a turtle feature selective enhancement module to effectively enhance the network's perception of fuzzy image features and improve the segmentation accuracy of underwater turtles. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings constituting a part of the specification of this embodiment are used to provide a further understanding of this embodiment. The schematic embodiments and descriptions of this embodiment are used to explain this embodiment and do not constitute an improper limitation on this embodiment.

[0018] Figure 1 Flowchart of the turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement in Example 1 of the present invention; Figure 2 This is a network architecture diagram of the turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement in Example 1 of the present invention; Figure 3This is a structural diagram of an image encoder based on weighted low-rank decomposition and updating in Embodiment 1 of the present invention; Figure 4 Schematic diagram of the structure of weighted feature fusion in embodiment 1 of the present invention; Figure 5 Schematic diagram of the turtle instance head segmentation in the first embodiment of the present invention; Figure 6 Schematic diagram of the turtle instance segmentation result in the first embodiment of the present invention; Figure 7 This is a structural block diagram of the turtle instance segmentation system based on weighted low-rank decomposition and feature enhancement in Example 2 of the present invention. DETAILED DESCRIPTION

[0019] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0020] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0021] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0022] In the present invention, terms such as "upper", "lower", "left", "right", "front", "back", "vertical", "horizontal", "side", "bottom", etc. indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are relational words determined only for the convenience of describing the structural relationships of the various parts or elements of the present invention, and do not specifically refer to any part or element in the present invention, and should not be understood as limiting the present invention.

[0023] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.

[0024] Example 1 Embodiment 1 of the present invention introduces a turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement.

[0025] like Figure 1 A turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement is shown, including: Acquire underwater turtle images; An image encoder is constructed based on a weighted low-rank decomposition and update mechanism. The global features of the acquired turtle images are captured using the image encoder and multi-head self-attention mechanism. The obtained global features of the turtle image are selectively enhanced to obtain a multi-scale reconstructed feature map; Candidate regions are generated on the obtained multi-scale reconstructed feature map, and the multi-scale reconstructed feature map is aligned on the generated candidate regions. The aligned multi-scale reconstructed feature map is detected and segmented through a multi-task branch network to complete the turtle instance segmentation.

[0026] like Figure 2 As shown, the network of this embodiment includes: an image encoder based on weighted low-rank decomposition and update, which improves the model's adaptability and segmentation capabilities for underwater images by designing a weighted low-rank decomposition and update mechanism, while reducing the computing resources and time required for training; a turtle feature selection enhancement module, which consists of a key feature selection module and a multi-scale feature generator. The feature selection module selects key features output by the image encoder and performs multi-level weighted fusion; the multi-scale feature generator performs multi-scale segmentation and generation on the fused feature map; and a turtle instance segmentation head, which uses a built-in region generation network to generate candidate regions. After the region alignment module, the candidate regions are accurately aligned with the original image, achieving accurate instance segmentation.

[0027] As one or more implementations, the image encoder based on weighted low-rank decomposition update uses a multi-head self-attention mechanism to capture global context information in the image, thereby extracting high-level feature representations. Figure 3 As shown in Figure 1, the encoder mainly consists of five modules: image block encoding module, position encoding module, encoder module, feature fusion module, and weighted low-rank decomposition update module. The image block encoding module is used to divide the image into image blocks of size 16×16 and use a two-dimensional convolution operation with a convolution kernel size and a stride of 16 to map each image block to a 768-dimensional feature vector; that is, ; in, I represents the input image, represents the convolution operation, N is the total number of image blocks.

[0028] The position encoding module adds a learnable position vector to each image block to preserve the spatial position information, allowing the encoder to model the structural features of the image. The encoder module uses a multi-head self-attention mechanism and a feedforward neural network to model the complex relationship between global and local features. The multi-layer encoder can extract multi-level high-dimensional semantic features, providing rich and diverse feature representations for subsequent segmentation tasks; that is, ; ; ; in, Represent the query value, key value and original value matrices respectively, and the three matrices come from the input features. It is i The linear projection matrix of the attention heads, Output projection matrix used to fuse the outputs of all attention heads; is the feature dimension of each attention head, usually ,in h is the number of attention heads; is the attention weight, which is used to measure the similarity between the query value and the key value. The feature fusion module fuses and optimizes the features extracted by the encoder.

[0029] Since the encoder module adopts a multi-head self-attention mechanism, there is a problem of high computational complexity. In order to improve the segmentation performance of the network and reduce the computational complexity, this embodiment designs a weight low-rank decomposition and update mechanism to perform lightweight and efficient fine-tuning on the encoder; the core idea of ​​the weight low-rank decomposition and update mechanism is to introduce a low-rank parameterization method to perform low-rank decomposition on some weights of the pre-trained model and fine-tune the decomposed low-rank matrix to adapt to specific downstream tasks; Figure 3 As shown, input , for the pre-trained weight matrix and , using low-rank decomposition to constrain its update; that is ; ; in, , , and order During training, freeze the weight matrix and , only for A and B The parameters in are updated with gradients.

[0030] This example uses uniform distribution initialization A Parameters, and B All parameters of are set to zero. Compared with the method of directly adjusting all weight parameters, fine-tuning greatly reduces the computational cost, and good fine-tuning effects can be achieved by adjusting a small number of parameters.

[0031] Due to the problems of color shift, low illumination and image blur in underwater images, it is difficult for the image encoder to accurately capture the edge details and complete semantic information of the turtle, which in turn affects the image segmentation accuracy. To this end, this paper implements a turtle feature selective enhancement module for the underwater blurry image segmentation task. The module consists of two core parts: a key feature selection module and a multi-scale feature generator. Figure 4 As shown in Figure 3, the multi-stage selection mechanism of the key feature selection module can realize the interactive fusion of multi-level features and significantly enhance the semantic information of underwater images. The selective enhancement process is as follows: ; ; ; ; ; in, represents the features generated by the encoder, express go through Features after downsampling; application of upper-layer features Generate selectivity coefficient Used to highlight key features; selectivity coefficient With low-level feature maps Perform element-wise multiplication and combine with high-level feature maps Add together to get the updated feature map After layer-by-layer iteration, the multi-level feature maps after selective enhancement are fused into feature information with rich semantic information, and Restore the channel dimension and generate the final fusion feature map ;right Convolution and pooling operations are used to reconstruct feature maps at different scales, generating 5 feature maps of different scales for subsequent instance segmentation tasks.

[0032] The turtle instance segmentation head structure is as follows Figure 5As shown in the figure, it includes three stages: candidate region generation, precise alignment, and multi-task prediction. First, the region generation network generates candidate regions on the multi-scale feature map. Specifically, the region generation network introduces three anchor boxes of preset sizes and aspect ratios at each feature point, and simultaneously predicts the confidence of the target existence and the regression offset of the bounding box for each anchor box; all candidate boxes are screened through non-maximum suppression, and 1,000 high-quality candidate regions are retained as input for subsequent processing; the region alignment module accurately aligns the candidate regions with the feature map to avoid the quantization error caused by traditional alignment methods. This module uses bilinear interpolation to obtain the feature values ​​of four sub-pixel points at each sampling position, and combines the maximum pooling operation to extract fixed-size regional features; the aligned features are sent to the multi-task branch network to complete the detection and segmentation tasks respectively. One branch is responsible for target classification and bounding box regression, which is used to further identify the target category and refine its spatial position; the other branch uses a fully convolutional network structure to generate a high-resolution binary mask corresponding to the target to achieve pixel-level instance segmentation. This design not only takes into account the accuracy of detection and segmentation, but also has strong generalization ability, and is suitable for target segmentation tasks in complex backgrounds in underwater images.

[0033] As one or more implementation methods, after the network structure of this embodiment is built, it is necessary to design a loss function to train the network. The loss function of the underwater turtle instance segmentation network consists of the region proposal network loss, the detection head classification loss, the detection head bounding box regression loss and the segmentation mask loss; the total loss function for: ; The candidate region generation network is responsible for generating candidate regions. Its losses include classification loss (foreground / background) and bounding box regression loss. The former uses cross entropy loss and the latter uses smooth L1 loss. The loss function is as follows: ; ; ; ; in, are anchor labels, representing background and foreground respectively; is the probability of being predicted as the foreground; is the normalization term; are the predicted bounding box parameters, is the true bounding box parameter; is the balance coefficient; is the number of positive samples, which plays a normalization role.

[0034] The classification loss of the detection head is used to predict the target category of the candidate area, and the multi-classification cross entropy loss function used is for ;in, is the number of candidate regions, which plays a normalization role. is the true category label of the candidate region. is the probability of predicting the true category.

[0035] The bounding box regression loss of the detection head is used to fine-tune the bounding box of the positive sample. The smooth L1 loss is used to measure the difference between the predicted bounding box parameters and the true bounding box, and to optimize the model so that it can generate more accurate target positioning results, that is, ;in, and are the predicted and true bounding box parameters respectively; is the number of positive samples.

[0036] The segmentation mask loss uses pixel-by-pixel binary cross entropy to measure the difference between the predicted mask and the true mask. For each positive sample, only the binary mask loss of the category to which it belongs is calculated, thereby optimizing the model to generate more accurate segmentation results; its segmentation mask loss function is for: ; in, is the true mask at position ( h , w ) label; is the predicted mask probability; is the total number of pixels of positive samples.

[0037] During training, data augmentation was performed using random horizontal flipping, random scaling, and random cropping. The experimental environment was Python 3.10 (Ubuntu 20.04), PyTorch 2.1.1, and Cuda 12.1. All experiments were conducted using an NVIDIA GeForce RTX 4090PLUS GPU. Training parameters were set as follows: AdamW optimizer, initial learning rate: 0.0001, weight decay rate: 0.1, batch size: 4, training epochs: 24; image resolution: 1024×1024.

[0038] This example uses a specially collected underwater turtle image dataset to train the proposed segmentation network and evaluates its performance on the test set of this dataset. The experimental results are shown in Figure 2. Figure 6 As shown, the proposed network performs well in both object detection and instance segmentation tasks. On the test set of the turtle dataset, the bounding box average precision AP of the proposed network is boxReaching 88.8%, the average mask precision AP mask The accuracy reached 87.7%, which verified that the network has good detection and segmentation capabilities for targets in complex underwater environments.

[0039] This embodiment improves the adaptability and training efficiency of the network by designing a low-rank decomposition and update mechanism for weights, and innovatively designs a turtle feature selective enhancement module to effectively enhance the network's perception of fuzzy image features and improve the segmentation accuracy of underwater turtles.

[0040] Example 2 The second embodiment of the present invention introduces a turtle instance segmentation system based on weighted low-rank decomposition and feature enhancement.

[0041] like Figure 7 A turtle instance segmentation system based on weighted low-rank decomposition and feature enhancement is shown, including: An image acquisition module, fully configured to acquire underwater sea turtle images; The feature capture module is fully configured to build an image encoder based on a weighted low-rank decomposition and update mechanism, and to capture the global features of the acquired turtle images using the image encoder and the multi-head self-attention mechanism; A feature enhancement module is configured to selectively enhance the obtained global features of the turtle image to obtain a multi-scale reconstructed feature map; The instance segmentation module is configured to generate candidate regions on the obtained multi-scale reconstructed feature map, align the multi-scale reconstructed feature map on the generated candidate regions, and perform detection and segmentation of the aligned multi-scale reconstructed feature map through a multi-task branch network to complete turtle instance segmentation.

[0042] The detailed steps are the same as those of the turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement provided in Example 1, and will not be repeated here.

[0043] Example 3 A third embodiment of the present invention provides a computer-readable storage medium.

[0044] A computer-readable storage medium stores a program thereon, which, when executed by a processor, implements the steps of the turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement as described in the first embodiment of the present invention.

[0045] The detailed steps are the same as those of the turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement provided in Example 1, and will not be repeated here.

[0046] Example 4 A fourth embodiment of the present invention provides an electronic device.

[0047] An electronic device includes a memory, a processor, and a program stored in the memory and running on the processor. When the processor executes the program, it implements the steps of the turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement as described in Example 1 of the present invention.

[0048] The detailed steps are the same as those of the turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement provided in Example 1, and will not be repeated here.

[0049] Example 5 A fifth embodiment of the present invention provides a computer program product.

[0050] A computer program product includes software code, wherein the program in the software code executes the steps in the turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement as described in Example 1 of the present invention.

[0051] The detailed steps are the same as those of the turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement provided in Example 1, and will not be repeated here.

[0052] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk drives, CD-ROMs, optical storage devices, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention may be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0053] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0054] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0055] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0056] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0057] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

[0058] The above description is merely a preferred embodiment of this embodiment and is not intended to limit this embodiment. Those skilled in the art will readily appreciate that this embodiment may be modified and varied in various ways. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this embodiment shall be within the scope of protection of this embodiment.

Claims

1. A turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement, characterized in that: include: Acquire underwater turtle images; An image encoder is constructed based on a weighted low-rank decomposition and update mechanism. The global features of the acquired turtle images are captured using the image encoder and multi-head self-attention mechanism. The obtained global features of the turtle image are selectively enhanced to obtain a multi-scale reconstructed feature map; Candidate regions are generated on the obtained multi-scale reconstructed feature map, and the multi-scale reconstructed feature map is aligned on the generated candidate regions. The aligned multi-scale reconstructed feature map is detected and segmented through a multi-task branch network to complete the turtle instance segmentation.

2. A turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement as claimed in claim 1, characterized in that: The constructed image encoder has a built-in weighted low-rank decomposition update module; in the weighted low-rank decomposition update module, the weights are subjected to low-rank decomposition through the weighted low-rank decomposition and update mechanism to obtain the decomposed low-rank matrix, and the update is completed.

3. The turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement as claimed in claim 1, characterized in that: In feature selective enhancement, a turtle image feature map is obtained based on the global features and selectivity coefficients of the turtle image. The obtained turtle image feature map is iterated layer by layer to obtain a multi-level feature map, and the obtained multi-level feature maps are fused to obtain a fused feature map. Convolution and pooling are used to reconstruct the obtained fused feature map at different scales to obtain a multi-scale reconstructed feature map.

4. The turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement as claimed in claim 1, characterized in that: The turtle instance segmentation includes a candidate region generation stage, an alignment stage, and a multi-task prediction stage; wherein, in the candidate region generation stage, the region generation network introduces an anchor box of preset size on the feature point, and based on the confidence of the anchor box prediction target and the regression offset of the bounding box, non-maximum suppression is used to screen all candidate boxes to obtain candidate regions, completing the candidate region generation stage.

5. The turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement as claimed in claim 4, characterized in that: In the alignment stage, bilinear interpolation is used to obtain feature values ​​at the sampling positions, and the maximum pooling operation is combined to extract regional features of fixed size; the aligned features are sent to the multi-task branch network to complete the detection and segmentation tasks of the scale-reconstructed feature map respectively.

6. The turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement as claimed in claim 1, characterized in that: The loss function of the turtle instance segmentation includes the region proposal network loss function, the detection head classification loss function, the detection head bounding box regression loss function and the segmentation mask loss function.

7. A turtle instance segmentation system based on weighted low-rank decomposition and feature enhancement, characterized in that: include: An image acquisition module, fully configured to acquire underwater sea turtle images; The feature capture module is fully configured to build an image encoder based on a weighted low-rank decomposition and update mechanism, and to capture the global features of the acquired turtle images using the image encoder and the multi-head self-attention mechanism; A feature enhancement module is configured to selectively enhance the obtained global features of the turtle image to obtain a multi-scale reconstructed feature map; The instance segmentation module is configured to generate candidate regions on the obtained multi-scale reconstructed feature map, align the multi-scale reconstructed feature map on the generated candidate regions, and perform detection and segmentation of the aligned multi-scale reconstructed feature map through a multi-task branch network to complete turtle instance segmentation.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement as described in any one of claims 1 to 6 are implemented.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the program, the steps of the turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement are implemented as described in any one of claims 1 to 6.

10. A computer program product comprising software code, characterized in that The program in the software code executes the steps of the turtle instance segmentation method based on weighted low-rank decomposition and feature enhancement as described in any one of claims 1-6.

Images