Underwater target detection method and system based on scene self-adaption

Abstract

The invention discloses an underwater target detection method and system based on scene self-adaption, and belongs to the field of computer vision and deep learning. According to the method, a scene classification model is introduced at the front end of a detection model, and accurate recognition of the underwater environment is achieved. And based on the classification result, the target detection model most suitable for the current scene is adaptively selected, so that effective adaptation to different underwater environments is realized, and the accuracy and robustness of underwater target detection are remarkably improved. According to the dynamic selection mechanism, the detection system can flexibly adjust the detection strategy according to different scene characteristics, and the detection performance of the model in a complex underwater environment is remarkably improved. Therefore, accurate positioning and identification of the underwater target are realized, the problem of unstable detection performance in a complex underwater environment in the prior art is effectively solved, and the method not only improves the detection accuracy, but also enhances the adaptability of the system to different underwater scenes.

Description

Technical Field

[0001] The present invention belongs to the field of computer vision and deep learning, and relates to a scene-adaptive underwater target detection method and system. Background Art

[0002] Underwater target detection mainly adopts target detection methods based on deep learning. The typical detection framework includes two core components: feature extraction network and detection head. In terms of feature extraction, feature pyramid network (FPN) enhances the expression ability of features at different levels through top-down feature transfer and realizes multi-scale feature fusion. In the target detection stage, the mainstream method adopts a two-stage detection strategy, the region proposal network (RPN) generates candidate boxes, and R-CNN performs accurate classification and position regression. Traditional methods face serious challenges in underwater environments. Underwater images generally have problems such as blur, distortion and uneven lighting. These complex imaging environments make it difficult to accurately extract target features. Existing research mostly focuses on a single detection framework, lacking in-depth discussion on the fusion of classification and detection tasks. In the complex and changeable underwater environment, a single detection model is difficult to adapt to the changes in different scenes, resulting in unstable detection performance. Summary of the invention

[0003] In view of the above problems, the present invention proposes a scene-adaptive underwater target detection method and system.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] The present invention proposes a scene-adaptive underwater target detection method, comprising:

[0006] Acquire an underwater target environment image, classify the underwater target environment image, and acquire an underwater target environment classification image;

[0007] Construct an underwater target detection model based on ResNet network and DAFPN feature fusion structure;

[0008] The underwater target environment classification image is input into the underwater target detection model to obtain images of different categories and their corresponding positions, thereby realizing the detection of underwater images.

[0009] Preferably, the underwater target environment image is classified using a target image classification network to obtain an underwater target environment classification image;

[0010] The target image classification network first receives an underwater target environment image and performs initial feature extraction through a convolutional layer. After downsampling through a Max Pool layer, it is sequentially processed through two first Res Block residual blocks, and deep feature learning is performed through multiple second Res Block residual blocks. The features finally obtained are aggregated through an Avg Pool average pooling layer, and an underwater target environment classification image is output through an FC (3) fully connected layer. The underwater target environment classification image includes a uniform illumination scene image, a non-uniform illumination scene image, and an image enhancement scene image.

[0011] Preferably, the first Res Block residual block and the second Res Block residual block include a Batch Norm layer, a 3×3 convolution layer and a ReLU activation function, and retain the original feature information through skip connections.

[0012] Preferably, a gated network is used to receive underwater target environment classification images, and an underwater target detection model matching the current scene style is adaptively selected through a mapping relationship.

[0013] Preferably, the underwater target detection model includes a feature extraction module, a feature enhancement module and a detection head module;

[0014] Input the underwater target environment classification image to the Backbone feature extraction module, gradually extract features through convolution operations, and generate feature maps of three different scales; input the three feature maps of different scales to the DAFPN feature enhancement module, adjust the channel of each feature map through 1×1 convolution, and then realize the fusion of features of different scales through upsampling and downsampling operations, use 3×3 convolution for feature enhancement, and generate the corresponding enhanced feature map; send the enhanced feature map to the Detection Head module, the classification branch of the detection head module predicts the category information of the target, and the regression branch of the detection head module predicts the location information of the target, and finally outputs the location coordinates of the detection box and the corresponding category prediction result.

[0015] Preferably, when the feature enhancement module performs feature extraction, the DAFPN includes multiple CGR modules and feature fusion units. The CGR module extracts features through convolution operations and establishes connections between different scales; the feature fusion unit adopts an adaptive fusion strategy to integrate features at different levels.

[0016] Preferably, the detection head module includes an RPN module, an R-CNN module and a sample reweighting mechanism; RPN generates candidate boxes, and R-CNN performs precise classification and position regression; the sample reweighting mechanism dynamically adjusts the sample weights according to the prediction error of RPN, and outputs results including target categories and position coordinates.

[0017] The present invention proposes an underwater target detection system based on scene adaptation, comprising:

[0018] An image acquisition module, wherein the image acquisition module is used to acquire an underwater target environment image, classify the underwater target environment image, and acquire an underwater target environment classification image;

[0019] A model building module, wherein the model building module is used to build an underwater target detection model based on a ResNet network and a DAFPN feature enhancement structure;

[0020] The image processing module is used to input the underwater target environment classification image into the underwater target detection model to obtain images of different categories and corresponding positions, thereby realizing the detection of underwater images.

[0021] A terminal device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of a scene-adaptive-based underwater target detection method are implemented.

[0022] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a scene-adaptive underwater target detection method.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] The present invention proposes a method for underwater target detection based on scene adaptation. The method introduces a scene classification model before detection to achieve accurate recognition of the underwater environment. Based on the classification result, the target detection model that best suits the current scene is adaptively selected. This dynamic selection mechanism enables the detection system to flexibly adjust the detection strategy according to the characteristics of different scenes, significantly improving the detection performance of the model in complex underwater environments. Through the accurate recognition of the scene by the classification model and the dynamic selection of the detection model, effective adaptation to different underwater environments is achieved, significantly improving the accuracy and robustness of underwater target detection. Therefore, the present invention achieves accurate positioning and identification of underwater targets, and effectively solves the problem of unstable detection performance of the prior art in complex underwater environments. This method not only improves the detection accuracy, but also enhances the system's adaptability to different underwater scenes.

[0025] The present invention proposes a scene-adaptive underwater target detection system, which divides the system into an image acquisition module, a model building module and an image processing module, obtains images of different categories and their corresponding positions, and realizes the detection of underwater images. The modularization concept is adopted to make each module independent of each other, which is convenient for unified management of each module. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0027] Figure 1 This is a flow chart of the underwater target detection method based on scene adaptation of the present invention.

[0028] Figure 2 It is an architecture diagram of the scene recognition network proposed in the present invention.

[0029] Figure 3 It is an architecture diagram of the target detection network of the present invention.

[0030] Figure 4 It is the overall network architecture diagram of the present invention.

[0031] Figure 5 This is a diagram of the underwater target detection system based on scene adaptation of the present invention.

[0032] Figure 6 The figure is a schematic diagram of the structure of an electronic device of the present invention. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0034] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0035] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0036] In the description of the embodiments of the present invention, it should be noted that if the terms "upper", "lower", "horizontal", "inner", etc. indicate an orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the product of the invention is usually placed when in use, it is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0037] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0038] The present invention proposes a scene-adaptive underwater target detection method, which improves the scene adaptability and detection accuracy of the detection model under different data styles by integrating scene classification and detection tasks. The detection method can adaptively select the optimal detection model according to the style of the input image, thereby achieving accurate and robust target detection in a variety of scenes. Figure 1 As shown, the steps of the detection method are as follows:

[0039] S1, obtaining an underwater target environment image, classifying the underwater target environment image, and obtaining an underwater target environment classification image;

[0040] The underwater target environment image is classified using a target image classification network to obtain an underwater target environment classification image;

[0041] The target image classification network first receives an underwater target environment image and performs initial feature extraction through a convolutional layer. After downsampling through a Max Pool layer, it is sequentially processed through two first Res Block residual blocks, and deep feature learning is performed through multiple second Res Block residual blocks. The features finally obtained are aggregated through an Avg Pool average pooling layer, and an underwater target environment classification image is output through an FC (3) fully connected layer. The underwater target environment classification image includes a uniform illumination scene image, a non-uniform illumination scene image, and an image enhancement scene image.

[0042] The first Res Block residual block and the second Res Block residual block contain a Batch Norm layer, a 3×3 convolution layer, and a ReLU activation function, and retain the original feature information through skip connections.

[0043] S2. Build an underwater target detection model based on ResNet network and DAFPN feature enhancement structure;

[0044] A gated network is used to receive underwater target environment classification images, and through the mapping relationship, an underwater target detection model that matches the current scene style is adaptively selected.

[0045] The underwater target detection model includes a feature extraction module, a feature enhancement module and a detection head module;

[0046] The underwater target environment classification image is input into the Backbone feature extraction module, and features are gradually extracted through convolution operations to generate feature maps of three different scales. The three feature maps of different scales are input into the DAFPN feature enhancement module, and each feature map is first subjected to 1×1 convolution for channel adjustment, and then upsampling and downsampling operations are used to achieve the fusion of features of different scales, and 3×3 convolution is used for feature enhancement to generate the corresponding enhanced feature map. The enhanced feature map is sent to the Detection Head module, and the classification branch of the detection head module predicts the category information of the target, and the regression branch of the detection head module predicts the location information of the target, and finally outputs the location coordinates of the detection box and the corresponding category prediction result.

[0047] When the feature enhancement module performs feature extraction, DAFPN contains multiple CGR modules and feature fusion units. The CGR module extracts features through convolution operations and establishes connections between different scales; the feature fusion unit adopts an adaptive fusion strategy to integrate features at different levels.

[0048] The detection head module includes an RPN module, an R-CNN module and a sample reweighting mechanism; RPN generates candidate boxes, and R-CNN performs precise classification and position regression; the sample reweighting mechanism dynamically adjusts the sample weights according to the prediction error of RPN, and outputs results including target categories and position coordinates.

[0049] S3. Input the underwater target environment classification image into the underwater target detection model to obtain images of different categories and corresponding positions, thereby realizing the detection of underwater images.

[0050] The following is a detailed description of this method:

[0051] Step 1. Dataset construction and preprocessing: Based on the UTDAC2020 dataset, this paper uses radial masking technology and CycleGAN to generate non-uniform illumination and enhanced images, and constructs an extended dataset containing a variety of typical underwater scenes. The dataset covers the following scene types: uniform illumination scene (Uniform), non-uniform illumination scene (Non-uniform), and image enhancement scene (Augmented).

[0052] In terms of dataset construction, the present invention uses radial gradient masking technology to process the UTDAC2020 dataset according to the characteristics of different underwater scenes, and generates a simulated deep-sea non-uniform imaging dataset. This technology can effectively simulate the light attenuation and scattering characteristics in deep-water environments, providing more realistic training data for the model. On the other hand, an object-guided dual adversarial contrast learning method is used to construct an enhanced dataset for shallow sea scenes. This method significantly improves the visibility of the target by generating detection-friendly enhanced images, providing better training samples for the detection model.

[0053] Step 2: Style classification: The input image is passed through the scene classification model for style discrimination. The scene classification model uses the ResNet network as the backbone network. After feature extraction, pooling layer and full connection layer, it can classify the image into one of three styles: augmented, non-uniform and uniform. This classification can help the subsequent detection model select appropriate parameters and feature layers to adapt to the image characteristics under different styles.

[0054] Classification network architecture: The classification network is based on the MMpretrain framework. For scene recognition, a backbone network based on ResNer18 is designed. The residual block has two 3×3 convolutional layers, followed by a batch normalization layer and a ReLU activation function. You can skip these two convolutional layers and add the input directly before the final ReLU activation function. Figure 2As shown in the figure. The processing flow of the model structure starts from the input stage (64 filters). First, the underwater image is received and the initial feature extraction is performed through the 7×7 convolution layer. After downsampling through the Max Pool layer, the feature processing is performed through two Res Block residual blocks in succession. Then, in the feature extraction backbone network, deep feature learning is performed through multiple Res Block residual blocks. Each residual block contains a combination of Batch Normalization layer, 3×3 convolution layer, and ReLU activation function, and the original feature information is retained through jump connections. At the same time, the number of feature channels gradually increases from 128 filters to 256 filters, and finally reaches 512 filters. In the output stage, the features are aggregated through the Avg Pool average pooling layer, and finally three different types of prediction results are output through the FC (3) fully connected layer: Uniform (uniform distribution), Non_Uniform (non-uniform distribution) and Augmented (enhanced). This structural design improves the feature expression ability through residual connection and multi-scale feature extraction, and enhances the adaptability of the model to different scenarios by using three-branch output. The entire network effectively improves the performance of underwater target detection through this multi-scale feature extraction and fusion method. The network contains the following key components: the input layer processes the initial feature extraction, the maximum pooling layer reduces the feature map size, and the batch normalization layer standardizes the feature distribution. The deep residual network contains multiple residual modules. The input of the residual module is directly superimposed on the output of the module after identity mapping. This residual function is easier to optimize and can greatly deepen the number of network layers. There are two layers in the residual block, as shown in formula (1), where σ represents the nonlinear function ReLu.

[0055] F(x)=W 2σ (W1x)(1)

[0056] F(x) represents the feature extraction and transformation function of the entire network, W 2σ represents the weight parameter of the second layer network, W1 represents the weight parameter of the first layer network, and x represents the input feature map. Then through a short-circuit connection and the second ReLU, the output y1 is obtained as shown in formula (2).

[0057] y1=F(x,{W i})+x (2)

[0058] y1 represents the output feature map after feature enhancement, W i Represents the set of learnable weight parameters in the network. When the input and output dimensions need to be changed, a linear transformation w can be performed on x in the short-circuit connection, as shown in formula (3).

[0059] y2=F(x,{Wi})+W s x (3)

[0060] W s Represents the linear transformation weight in the short-circuit connection. The purpose of y2 is to enhance the model's detection ability for underwater targets through feature enhancement and short-circuit connection.

[0061] Step 3, Gated network: The gated network is one of the core components of the present invention, and is responsible for dynamically selecting the optimal target detection model based on the output of the classification model. Specifically, the classification model first identifies the scene style of the input image and inputs the identification result into the gated network. After receiving the style category information, the gated network adaptively selects the target detection model that best matches the current scene style through a preset mapping relationship. For example, when the classification model identifies a deep-sea scene, the gated network will select a detection model specifically optimized for the deep-sea environment; when a shallow-sea scene is identified, a detection model that is more suitable for the shallow-sea environment will be selected. This dynamic selection mechanism ensures that the detection system can always use the detection model that best suits the current scene, effectively improving the detection performance and adaptability of the system in different underwater environments.

[0062] Step 4, target detection: For each style of data, the present invention trains independent target detection models respectively. These detection models are built based on ResNet and DAFPN structures to adapt to the characteristics of different data styles. During the detection process, the feature extraction stage generates a multi-scale feature map, and then the detection head performs target positioning and classification at each scale. In addition, RetinaRPN and ROI Align modules are integrated in each detection model to enhance the feature learning effect and positioning accuracy of the target area. In order to further improve the robustness of detection, the present invention introduces the BoostingReweighting strategy in the detection head, and performs specific optimization on data of different styles, so that the detection model maintains good generalization ability on data sets with large style changes.

[0063] Object detection network architecture: The object detection network adopts a "backbone-neck-heads" structure, such as Figure 3The figure shows the target detection network architecture of the present invention. The model adopts an end-to-end detection architecture. First, an underwater image is input into the Backbone feature extraction module, and features are gradually extracted through a multi-layer convolutional network to generate three feature maps C3, C4 and C5 of different scales (corresponding to shallow to deep features respectively); then these feature maps are input into the DAFPN feature enhancement module, where each feature map is first subjected to a 1×1 convolution for channel adjustment, and then upsampling and downsampling operations are used to achieve the fusion of features of different scales, and 3×3 convolutions are used for feature enhancement to generate corresponding enhanced feature maps P3, P4 and P5; finally, these enhanced feature maps are sent to the Detection Head module, which includes two branches, classification and regression. The classification branch is responsible for predicting the category information of the target, and the regression branch is responsible for predicting the location information of the target, and finally outputs the location coordinates of the detection box and the corresponding category prediction results.

[0064] In the feature extraction stage, the network uses a backbone network to perform multi-level feature extraction on the input image. The extracted features are processed by DAFPN, which contains multiple convolutional layers (CGR modules) and feature fusion units. The CGR module extracts features through convolution operations and establishes connections between different scales. The feature fusion unit adopts an adaptive fusion strategy to effectively integrate features at different levels. In the detection head part, the network contains two submodules, RPN and R-CNN. RPN is responsible for generating candidate boxes, while R-CNN performs accurate classification and position regression. The detection head also contains a sample reweighting mechanism to dynamically adjust the sample weights according to the prediction error of RPN. The final output includes detection results such as target category and position coordinates. The entire network structure achieves accurate detection of underwater targets through multi-scale feature extraction and fusion, as well as adaptive adjustment of the detection head. The structural design fully considers the particularity of the underwater environment and can effectively handle target detection tasks in complex scenes.

[0065] Step 5, result output: By organically combining style classification and object detection, the framework of the present invention can adaptively select the best detection model according to the style of the input data. The experimental results show that, as shown in Table 1, the performance comparison results of six detection models (Boosting-RCNN, UDMDet_III, TOOD, GCC_NET, VFNET and SI-MEM proposed in this invention) in three underwater scenes (uniform scene, non-uniform scene and enhanced scene). By comparing the two evaluation indicators of AP (average precision) and AP50 (average precision when the IOU threshold is 0.5), it can be seen that the SI-MEM model proposed in this paper has achieved the best performance in all scenes: 48.3% AP and 83.4% AP50 in uniform scene, 44.8% AP and 79.7% AP50 in non-uniform scene, and 45.5% AP and 80.6% AP50 in enhanced scene. These data are significantly better than other comparison models, verifying the effectiveness of the multi-scene adaptive detection framework proposed in this paper.

[0066] Table 1

[0067]

[0068] Therefore, the present invention realizes multi-scene target detection through the above steps: scene classification, the input image is identified by ImageClassifier for scene type, Gating Network selects the corresponding detector according to the classification result, Uniform Detector is selected to process uniform illumination scenes, Non_Uniform Detector is selected to process non-uniform illumination scenes, and Augmented Detector is selected to process enhanced scenes. The detection process is as follows Figure 4 As shown, the algorithm first initializes the pre-trained model, selects the corresponding detector according to the scene type, performs target detection and outputs the results.

[0069] Table 2

[0070]

[0071] As shown in Table 2, the algorithm describes the workflow of a multi-scene object detection framework: first, the pre-trained classification model and the detector sets of three scenes (enhanced scene, non-uniform scene, uniform scene) are initialized, and then the classification model is used to perform scene style recognition on the input image. According to the recognition result, the most matching detector is dynamically selected for each image (such as Daug for enhanced scene, Dnon for non-uniform scene, Duni for uniform scene, and Duni is used by default when it cannot be judged). Finally, the selected detector is used to detect the object in the image and output the result, realizing a detector selection mechanism based on scene adaptation.

[0072] The present invention solves the problem of insufficient generalization ability of traditional single detection models on multi-style data sets through the collaborative work of classification and detection tasks. This framework can be widely used in scenarios such as underwater detection, and is particularly suitable for target detection tasks in complex environments with significant style differences, providing an efficient and stable solution for multi-scene and multi-style target detection. Through the collaborative work of the above-mentioned classification and detection models, the present invention can adaptively select a suitable detection model according to the style of the input data to achieve accurate detection of multi-style and multi-scene data. This method significantly improves the application performance of the detection model in complex environments and provides a new solution for multi-scene target detection tasks.

[0073] Innovations and advantages: Multi-scenario adaptability, targeted processing of different underwater scenes, improved detection accuracy and robustness, modular design, decoupling of classification and detection modules, easy maintenance and optimization, enhanced feature extraction, multi-scale feature fusion, spatial attention mechanism. The present invention effectively solves the problem of underwater multi-scenario target detection through innovative framework design and algorithm implementation, and provides new solutions for research and application in related fields.

[0074] The multi-scenario target detection framework of the present invention significantly improves the adaptability and detection accuracy of the detection model under different data styles by integrating classification and target detection. By training independent detection models on multi-style data sets (enhanced, non-uniform, uniform) respectively, and using the classification module to adaptively select the optimal model, the framework can achieve optimal performance in all types of styles, which not only effectively improves the detection accuracy, but also enhances the robustness of the model, and reduces false detection and missed detection. At the same time, the framework of the present invention has good generalization ability and can adapt to diversified data inputs, and is particularly suitable for complex environments such as underwater detection. In addition, the guidance of the classification module enables each detection to only call the detection model of the corresponding style, reducing the consumption of computing resources and improving the operating efficiency of the system. The invention provides an efficient and stable solution for multi-scenario target detection in complex environments.

[0075] Example 2

[0076] The present invention proposes an underwater target detection system based on scene adaptation, such as Figure 5 As shown, including:

[0077] An image acquisition module, wherein the image acquisition module is used to acquire an underwater target environment image, classify the underwater target environment image, and acquire an underwater target environment classification image;

[0078] A model building module, wherein the model building module is used to build an underwater target detection model based on a ResNet network and a DAFPN feature enhancement structure;

[0079] The image processing module is used to input the underwater target environment classification image into the underwater target detection model to obtain images of different categories and corresponding positions, thereby realizing the detection of underwater images.

[0080] Example 3

[0081] See also Figure 6 As shown, the present invention also provides an electronic device 100 based on a scene-adaptive underwater target detection method; the electronic device 100 includes a memory 101, at least one processor 102, a computer program 103 stored in the memory 101 and executable on the at least one processor 102, and at least one communication bus 104.

[0082] The memory 101 can be used to store the computer program 103. The processor 102 implements the steps of the underwater target detection method based on scene adaptation described in Example 1 by running or executing the computer program stored in the memory 101 and calling the data stored in the memory 101. The memory 101 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data (such as audio data) created according to the use of the electronic device 100, etc. In addition, the memory 101 can include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other non-volatile solid-state storage devices.

[0083] The at least one processor 102 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor 102 may be a microprocessor or any conventional processor, etc. The processor 102 is the control center of the electronic device 100, and uses various interfaces and lines to connect various parts of the entire electronic device 100.

[0084] The memory 101 in the electronic device 100 stores a plurality of instructions to implement a scene-adaptive underwater target detection method, and the processor 102 can execute the plurality of instructions to implement:

[0085] Acquire an underwater target environment image, classify the underwater target environment image, and acquire an underwater target environment classification image;

[0086] Construct an underwater target detection model based on ResNet network and DAFPN feature fusion structure;

[0087] The underwater target environment classification image is input into the underwater target detection model to obtain images of different categories and their corresponding positions, thereby realizing the detection of underwater images.

[0088] Example 4

[0089] If the module / unit integrated in the electronic device 100 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory and read-only memory (ROM, Read-Only Memory).

[0090] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, 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 storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0091] 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 flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, 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 flowchart and / or block diagram. 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.

[0092] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0093] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A scene-adaptive underwater target detection method, characterized in that: include: Acquire an underwater target environment image, classify the underwater target environment image, and acquire an underwater target environment classification image; Construct an underwater target detection model based on ResNet network and DAFPN feature fusion structure; The underwater target environment classification image is input into the underwater target detection model to obtain images of different categories and their corresponding positions, thereby realizing the detection of underwater images.

2. The underwater target detection method based on scene adaptation according to claim 1 is characterized in that: The underwater target environment image is classified using a target image classification network to obtain an underwater target environment classification image; The target image classification network first receives an underwater target environment image and performs initial feature extraction through a convolutional layer. After downsampling through a Max Pool layer, it is sequentially processed through two first Res Block residual blocks, and deep feature learning is performed through multiple second Res Block residual blocks. The features finally obtained are aggregated through an Avg Pool average pooling layer, and an underwater target environment classification image is output through an FC (3) fully connected layer. The underwater target environment classification image includes a uniform illumination scene image, a non-uniform illumination scene image, and an image enhancement scene image.

3. The underwater target detection method based on scene adaptation according to claim 2 is characterized in that: The first ResBlock residual block and the second Res Block residual block contain a Batch Norm layer, a 3×3 convolution layer, and a ReLU activation function, and retain the original feature information through skip connections.

4. The underwater target detection method based on scene adaptation according to claim 1, characterized in that: A gated network is used to receive underwater target environment classification images, and through the mapping relationship, an underwater target detection model that matches the current scene style is adaptively selected.

5. The underwater target detection method based on scene adaptation according to claim 1, characterized in that: The underwater target detection model includes a feature extraction module, a feature enhancement module and a detection head module; The underwater target environment classification image is input into the Backbone feature extraction module, and features are gradually extracted through convolution operations to generate feature maps of three different scales. The three feature maps of different scales are input into the DAFPN feature enhancement module, and each feature map is subjected to channel adjustment through 1×1 convolution. Then, features of different scales are fused through upsampling and downsampling operations, and features are enhanced using 3×3 convolution to generate corresponding enhanced feature maps. The enhanced feature maps are sent to the DetectionHead module, and the classification branch of the detection head module predicts the category information of the target, and the regression branch of the detection head module predicts the location information of the target, and finally outputs the location coordinates of the detection box and the corresponding category prediction results.

6. The underwater target detection method based on scene adaptation according to claim 5, characterized in that: When the feature enhancement module performs feature extraction, DAFPN contains multiple CGR modules and feature fusion units. The CGR module extracts features through convolution operations and establishes connections between different scales; The feature fusion unit adopts an adaptive fusion strategy to integrate features at different levels.

7. The scene-adaptive underwater target detection method according to claim 5, characterized in that: The detection head module includes an RPN module, an R-CNN module and a sample reweighting mechanism; RPN generates candidate boxes, and R-CNN performs precise classification and position regression; the sample reweighting mechanism dynamically adjusts the sample weights according to the prediction error of RPN, and outputs results including target categories and position coordinates.

8. A scene-adaptive underwater target detection system, characterized in that: include: An image acquisition module, wherein the image acquisition module is used to acquire an underwater target environment image, classify the underwater target environment image, and acquire an underwater target environment classification image; A model building module, wherein the model building module is used to build an underwater target detection model based on a ResNet network and a DAFPN feature enhancement structure; The image processing module is used to input the underwater target environment classification image into the underwater target detection model to obtain images of different categories and corresponding positions, thereby realizing the detection of underwater images.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the underwater target detection method based on scene adaptation as claimed in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the underwater target detection method based on scene adaptation as claimed in any one of claims 1 to 7 are implemented.

Images