CN-121982502-A - YOLOv 11-based underwater fishing net detection method and YOLOv-based underwater fishing net detection device

CN121982502ACN 121982502 ACN121982502 ACN 121982502ACN-121982502-A

Abstract

The invention provides an underwater fishing net detection method and device based on YOLOv11, the method comprises the steps of carrying out feature extraction on an input underwater image to be detected through a main feature extraction network in an improved YOLOv detection model to obtain an initial feature image, carrying out multi-scale feature fusion on the initial feature image through a neck network in the detection model to obtain a fusion feature image, carrying out fishing net detection on the fusion feature image through a detection head network in the detection model to obtain a detection result of a fishing net target, and adopting a GSConv module, a GSBottleneck bottleneck layer and a VoV-GSCSP module by the detection head network to improve the reasoning speed and detection efficiency of the model, strengthen the integration capability of the model on the features of different scales of the fishing net, improve the robustness of the model in a complex underwater scene and maintain higher detection precision on the premise of improving the efficiency.

Inventors

WANG CHONG
Zhang Zhengrun
LI HAO
SONG LIFEI

Assignees

武汉理工大学

Dates

Publication Date: 20260505
Application Date: 20251229

Claims (10)

1. The YOLOv-based underwater fishing net detection method is characterized by comprising the following steps of: Extracting features of an input underwater image to be detected through a trunk feature extraction network in an improved YOLOv detection model to obtain an initial feature map, wherein the trunk feature extraction network adopts a MobileNetV network to replace the trunk feature extraction network of an original YOLOv11, and carrying out multi-scale feature fusion on the initial feature map through a neck network in the improved YOLOv detection model to obtain a fusion feature map, wherein the neck network adopts a Slim-Neck structure to replace the neck network of the original YOLOv; And carrying out fishing net detection on the fusion characteristic map through a detection head network in the improved YOLOv detection model to obtain a detection result of a fishing net target, wherein the detection head network adopts a GSConv module, a GSBottleneck bottleneck layer and VoV-GSCSP modules to replace the original YOLOv detection head network.
2. The YOLOv 11-based underwater fishing net detection method as defined in claim 1, further comprising, prior to the feature extraction of the input underwater image by the trunk feature extraction network in the modified YOLOv11 detection model: The method comprises the steps that an original underwater image is acquired through an underwater robot, and the underwater robot is provided with a monocular camera and an auxiliary light source; And adopting an improved dark channel prior enhancement algorithm to perform enhancement processing on the original underwater image and the color to generate the underwater image to be detected.
3. The method for detecting an underwater fishing net based on YOLOv a 11 as defined in claim 2, wherein said performing defogging and color enhancement processing on said original underwater image using an improved dark channel prior enhancement algorithm to generate said underwater image to be detected comprises: Performing channel self-adaptive dark channel estimation processing based on minimum values of a blue channel and a green channel on an input original underwater image to obtain a preliminary transmissivity image; performing bright area extraction and background light estimation processing on the preliminary transmittance map to obtain a background light value containing red channel proportion correction; performing Sobel edge weight-based guided filtering treatment on the preliminary transmittance map to obtain an optimized transmittance map; Performing relative depth calculation and red channel adaptive gain compensation processing based on the optimized transmissivity graph to obtain a compensated image, wherein the gain value of the red channel adaptive gain compensation is smaller than the adaptive upper limit related to transmissivity; and sequentially performing white balance adjustment, local contrast enhancement based on CLAHE and texture sharpening on the compensated image to obtain the underwater image to be detected.
4. The method for detecting the underwater fishing net based on YOLOv th 11 of claim 1, wherein the MobileNetV network is of a reverse residual structure and comprises a1×1 convolution extended layer, a depth separable convolution layer and a1×1 convolution reduced layer, and the step of extracting features of the input underwater image through a main feature extraction network in the improved YOLOv detection model to obtain an initial feature map comprises the following steps: Performing channel expansion processing on the underwater image to be detected through the 1 multiplied by 1 convolution expansion layer to obtain a high-dimensional feature map; Performing spatial feature extraction processing on the high-dimensional feature map through the depth separable convolution layer to obtain a spatial feature map; and carrying out channel compression processing on the space feature map through the 1X 1 convolution reduction layer to obtain the initial feature map.
5. The YOLOv 11-based underwater fishing net detection method as defined in claim 1, wherein the fishing net detection of the fused feature map by the detection head network in the modified YOLOv11 detection model includes: performing feature transformation on the fusion feature map by using the GSConv module to obtain a first feature map; Inputting the first feature map into the GSBottleneck bottleneck layer for feature compression and cross-layer information fusion to obtain a second feature map; Inputting the second characteristic diagram into the VoV-GSCSP module for multi-scale characteristic strengthening and integration to obtain a strengthened multi-scale characteristic diagram for generating a detection result of the fishing net target.
6. The method for detecting an underwater fishing net based on YOLOv th claim 11, wherein the performing feature transformation on the fusion feature map by using the GSConv module to obtain a first feature map includes: dividing the fusion feature map into a first path feature and a second path feature; performing a standard convolution operation on the first path feature and a depth separable convolution operation on the second path feature; Splicing the first path characteristics after the standard convolution operation and the second path characteristics after the depth separable convolution operation in the channel dimension to obtain first splicing characteristics; And executing channel shuffling operation on the first spliced feature to generate the first feature map.
7. The method for detecting the underwater fishing net based on YOLOv and 11 of claim 5, wherein the GSBottleneck bottleneck layer comprises a preset GSConv module, the inputting the first feature map into the GSBottleneck bottleneck layer for feature compression and cross-layer information fusion to obtain a second feature map comprises: Inputting the first feature map into a preset GSConv module in the GSBottleneck bottleneck layer for feature compression processing to obtain basic transformation features; And carrying out residual connection and channel integration processing on the basic transformation characteristics to obtain the second characteristic diagram.
8. The YOLOv-based underwater fishing net detection method as defined in claim 1, wherein the detection head network is constructed based on a SSDLite framework, and the convolutional network in the SSDLite framework comprises a depth separable convolutional network.
9. The method for detecting an underwater fishing net based on YOLOv th 11 th claim 5, wherein inputting the second feature map into the VoV-GSCSP module for multi-scale feature enhancement and integration, the method for obtaining an enhanced multi-scale feature map comprises: Dividing the second feature map into a main path feature and an auxiliary path feature; performing standard convolution operation on the main path characteristics; performing standard convolution operation and lightweight bottleneck layer processing on the auxiliary path characteristics; Splicing the main path characteristics after standard convolution operation and the auxiliary path characteristics after light bottleneck layer processing in the channel dimension to obtain second splicing characteristics; and performing 3X 3 convolution operation on the second spliced characteristic to obtain a reinforced multi-scale characteristic diagram.
10. YOLOv 11-based underwater fishing net detection device is characterized by comprising: the extraction unit is used for extracting the characteristics of the input underwater image to be detected through a trunk characteristic extraction network in the improved YOLOv detection model to obtain an initial characteristic diagram, and the trunk characteristic extraction network adopts a MobileNetV network to replace the trunk characteristic extraction network of the original YOLOv; The fusion unit is used for carrying out multi-scale feature fusion on the initial feature map through a neck network in the improved YOLOv detection model to obtain a fusion feature map, and the neck network adopts a Slim-Neck structure to replace the neck network of the original YOLOv; The detection unit is used for carrying out fishing net detection on the fusion characteristic map through a detection head network in the improved YOLOv detection model to obtain a detection result of a fishing net target, wherein the detection head network adopts a GSConv module, a GSBottleneck bottleneck layer and VoV-GSCSP module to replace the original YOLOv detection head network.

Description

YOLOv 11-based underwater fishing net detection method and YOLOv-based underwater fishing net detection device Technical Field The invention relates to the technical field of underwater image detection, in particular to an underwater fishing net detection method and device based on YOLOv < 11 >. Background An underwater unmanned vehicle (Unmanned Underwater Vehicle, UUV) serves as a highly intelligent operation platform and plays an increasingly important role in the fields of marine exploration, environment monitoring, underwater facility inspection, maintenance and the like. The safety navigation of UUV is the precondition of executing various tasks. However, in complex real marine environments, especially offshore and aquaculture areas, there are a large number of fishing nets that are abandoned or destroyed by fishing or aquaculture activities. The fishing nets are often in a suspended or semi-suspended state, have weak visual characteristics and irregular forms, and are extremely easy to wind a propeller or a sensor of the UUV, so that equipment is damaged, tasks are failed, and even safety accidents are caused. Therefore, the real-time and accurate detection of the fishing net in the underwater environment is realized, and the method has important practical significance for autonomous obstacle avoidance and safe navigation of the UUV. In the related art, a target detection algorithm based on deep learning is mostly adopted, particularly models represented by YOLO (You Only Look Once), SSD (Single Shot MultiBox Detector) and fast R-CNN are adopted, however, the method has obvious defects that on one hand, the mainstream deep learning model is high in computational complexity and large in parameter quantity and needs powerful computing platform support, which contradicts limited airborne computing resources of UUV and other embedded equipment, on the other hand, the models are not designed aiming at underwater optical imaging characteristics, and the robustness of the model to the problem of degradation of underwater images (such as fog blurring and color cast) is insufficient, so that the detection performance in a complex underwater environment is reduced, particularly the detection precision and recall rate of fine, semitransparent and low-contrast targets such as fishing nets are difficult to meet practical requirements, and the detection efficiency and the robustness of the underwater fishing net target detection are reduced. Disclosure of Invention In view of the above, it is necessary to provide a method and a device for detecting underwater fishing net based on YOLOv a to solve the technical problems of low detection efficiency and robustness caused by detecting the underwater fishing net target by using the existing YOLOv a model. In order to solve the technical problem, in a first aspect, the present invention provides a method for detecting an underwater fishing net based on YOLOv a, comprising: Extracting features of an input underwater image to be detected through a trunk feature extraction network in an improved YOLOv detection model to obtain an initial feature map, wherein the trunk feature extraction network adopts a MobileNetV network to replace the trunk feature extraction network of the original YOLOv; Performing multi-scale feature fusion on the initial feature map through a neck network in the improved YOLOv detection model to obtain a fused feature map, wherein the neck network adopts a Slim-Neck structure to replace the neck network of the original YOLOv; And carrying out fishing net detection on the fusion characteristic map through a detection head network in the improved YOLOv detection model to obtain a detection result of a fishing net target, wherein the detection head network adopts a GSConv module, a GSBottleneck bottleneck layer and VoV-GSCSP modules to replace the original YOLOv detection head network. In a possible implementation manner, before the feature extraction of the input underwater image by the trunk feature extraction network in the improved YOLOv detection model, the method further includes: The method comprises the steps that an original underwater image is acquired through an underwater robot, and the underwater robot is provided with a monocular camera and an auxiliary light source; And adopting an improved dark channel prior enhancement algorithm to perform enhancement processing on the original underwater image and the color to generate the underwater image to be detected. In one possible implementation manner, the defogging and color enhancement processing is performed on the original underwater image by adopting a modified dark channel prior enhancement algorithm, so as to generate the underwater image to be detected, which includes: Performing channel self-adaptive dark channel estimation processing based on minimum values of a blue channel and a green channel on an input original underwater image to obtain a preliminary transmissivity image; perform