Research on SeaTreasure Target Detection Technology Based on Improved YOLOv7-Tiny
With the increasing market demand for sea treasures such as sea cucumbers, sea urchins and shells, the global aquaculture industry is booming. However, the complexity of underwater environments leads to problems such as missed and misdetection of sea treasures in the traditional manual inspection pr...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11031423/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | With the increasing market demand for sea treasures such as sea cucumbers, sea urchins and shells, the global aquaculture industry is booming. However, the complexity of underwater environments leads to problems such as missed and misdetection of sea treasures in the traditional manual inspection process. To address this challenge, this paper proposes a target detection algorithm for underwater sea treasures called UPA-YOLO, which aims to achieve accurate and efficient detection of underwater treasures and accelerates the inference through model transformation to enable the deployment of the detection model in edge devices. First, based on the YOLOv7-Tiny network, the MAFPN neck structure is used to replace the ELAN structure to achieve the multi-scale capture of semantic information of underwater sea treasures, and to enhance the UPA-YOLO model to accurately locate the targets of underwater sea treasures; second, the P2ELAN module is constructed and added to the backbone network, which makes use of the redundancy information in the feature map and dynamically adjusts the convolution kernel to adapt to data The P2ELAN module is added to the backbone network, using the redundant information in the feature map, dynamically adjusting the convolutional kernel to adapt to the lack of data, reducing the number of parameters in the model, and introducing the MSCA attention mechanism to inhibit the complex and changeable background features underwater, to improve the semantic feature extraction ability of the UPA-YOLO model for underwater targets, adding the MPDiou loss function to the improved algorithm model and completing the data validation of the detection model; finally, based on the TensorRT acceleration framework, the optimisation of the target detection Finally, based on the TensorRT acceleration framework, the target detection model is optimised, and the Jetson Nano edge device is used to complete the localisation deployment and realise the real-time target detection task of underwater sea treasures. Experimental results demonstrate that our optimized algorithm achieves a mAP0.5 of 79.5%, representing a 2.0 percentage point improvement over the baseline YOLOv7-tiny model (77.5%) while maintaining computational efficiency at 45.2 GFLOPs. When deployed on Jetson Nano edge devices, the model achieves an inference speed of 86 ms per image, corresponding to a <inline-formula> <tex-math notation="LaTeX">$4\times $ </tex-math></inline-formula> acceleration compared to the unoptimized version. This performance enables practical real-time detection of marine products in field applications. |
|---|---|
| ISSN: | 2169-3536 |