The vast oceans harbor immense potential for sustainable development, brimming with resources crucial for human well-being. Aquaculture plays a vital role in boosting both the fishing economy and national economies as a whole. As a pillar industry for coastal regions, it drives economic progress and paves the way for a brighter future.
However, aquaculture demands advanced technological solutions to drive its growth. Process automation is a key piece of this puzzle, and artificial intelligence emerges as a powerful tool to achieve it.
A team of researchers from Guilin University of Electronic Technology (China) published a study in the journal Scientific Reports, proposing a new underwater object detection network: WBi-YOLOSF. This new tool performs automatic classification and detection of aquatic products, improves production efficiency in the aquaculture industry, and promotes its economic development.
Challenges and the Rise of AI
Underwater environments pose unique challenges for object detection. Low-quality images plagued by factors such as turbidity, low contrast, and noise significantly hinder traditional methods. To overcome these obstacles, researchers have harnessed the power of deep learning.
WBi-YOLOSF, a novel underwater object detection network, represents a significant breakthrough. Based on the YOLO framework, it features several groundbreaking innovations:
- AU-BiFPN: This novel feature extraction structure enhances the network’s ability to extract crucial information, addressing the bottleneck of gradient information flow.
- Data augmentation: To compensate for the limitations of real-world data acquisition, WBi-YOLOSF utilizes data augmentation techniques, ultimately leading to greater accuracy, particularly for small object detection.
- Swarm intelligence optimization: A novelty in underwater object detection models, WBi-YOLOSF leverages swarm intelligence algorithms to optimize the network’s hyperparameters, speeding up training convergence and improving accuracy. This eliminates the need for manual configuration, allowing the network to learn optimal parameters on its own.
- Focal loss function: To further enhance the accuracy of dense object detection, the network employs the focal loss function, effectively focusing on areas that require improvement.
WBi-YOLOSF: An Underwater Neural Network to Boost Productivity
Researchers have developed a new underwater object detection neural network called WBi-YOLOSF. This innovative technology enables automatic classification and detection of aquatic products, a giant step toward optimizing production efficiency in the aquaculture industry.
To train this model, an extensive dataset comprising images of 15 different aquatic products was created. However, underwater environmental conditions present challenges in image quality. To overcome this limitation, an underwater image enhancement algorithm was implemented, resulting in a more robust dataset.
Overcoming Underwater Detection Challenges
A critical aspect of underwater object detection is the precise identification of multiple small targets, a problem that has traditionally led to high false positive and false negative rates. To address this challenge, a new data augmentation technique was developed, further enriching the training set.
Optimized Neural Network Architecture
The WBi-YOLOSF network architecture is based on a novel feature extraction network called AU-BiFPN. This structure solves the gradient degradation problem associated with deep networks, significantly improving multi-scale feature fusion. Additionally, the swarm intelligence algorithm was integrated to optimize the model’s hyperparameters, accelerating training and notably increasing accuracy.
Impressive Results
The results obtained are promising. The WBi-YOLOSF network achieves an average accuracy of 0.982 and a processing speed of 203 frames per second, outperforming other models in both metrics. This means rapid and accurate identification of aquatic products, laying the foundation for a fully automated aquaculture system.
Conclusion
The WBi-YOLOSF technology represents a significant advancement in the application of artificial intelligence to aquaculture. Its ability to detect underwater objects with high precision and speed has the potential to transform the aquaculture industry, increasing production, reducing costs, and ensuring a sustainable future for this vital sector.
At the heart of this advancement is a neural network called WBi-YOLOSF, designed to quickly and accurately identify different aquatic species underwater. This technology is crucial for automating processes in aquaculture, leading to greater efficiency, cost reduction, and an increase in sustainable production.
How This Innovation Works
- Enhanced underwater vision: Scientists have overcome underwater vision challenges, such as poor image quality and difficulty distinguishing small objects.
- Accelerated learning: The neural network has been optimized to learn faster and more accurately thanks to new data processing techniques and intelligent algorithms.
- Precise and rapid detection: WBi-YOLOSF can identify up to 15 different aquatic species with 98.2% accuracy and at a speed of 203 images per second.
Impact on Aquaculture
This technology has the potential to transform the aquaculture industry by enabling:
- Task automation: From feeding to harvesting, tasks can be performed by machines more efficiently.
- Resource optimization: Better management of resources, such as food and space, can lead to greater sustainability.
- Production increase: Rapid identification of species allows for more effective management of farms.
In summary, this development, combined with the use of YOLO to detect the quality and freshness of shrimp or deformities in salmon, represents a significant step toward a more modern, profitable, and environmentally friendly aquaculture.
The study was funded by the National Natural Science Foundation of China; Guangxi Innovation-Driven Development Special Fund; Guangxi Key Laboratory of Wireless Wideband Communication and Signal Processing; Guangxi Bagui Scholar; and Innovation Project of GUET Graduate Education.
Contact
Yujie Mu
School of Computer and Information Security, Guilin University of Electronic Technology
Guilin, 541004, China
Email: muyujie2021@163.com
Reference (open access)
Jiang, L., Mu, Y., Che, L., & Wu, Y. (2024). WBi-YOLOSF: Improved feature pyramid network for aquatic real-time target detection under the artificial rabbits optimization. Scientific Reports, 14(1), 1-27. https://doi.org/10.1038/s41598-024-68878-7
