Marine aquaculture has experienced explosive growth in China and other parts of the world over the past few decades, becoming a crucial economic pillar. However, this rapid and uncontrolled expansion has led to severe environmental problems, such as water pollution, loss of biodiversity, and obstruction of maritime routes.
To ensure the sustainability of this industry and protect our coastal ecosystems, it is imperative to implement an effective and precise monitoring system for marine farming areas. Accurate monitoring of these areas is essential to ensure their sustainability and protect marine resources.
Researchers from the South China Sea Institute of Planning and Environment Research, the Key Laboratory of Marine Environmental Survey Technology and Application, the Technology Innovation Center for South China Sea Remote Sensing, Surveying and Mapping Collaborative Application, and Henan University of Urban Construction evaluated the capabilities of two well-known YOLO models, YOLOv5 and YOLOv7, to detect offshore aquaculture areas based on different high-resolution optical remote sensing images.
The Revolution of Remote Sensing
Traditionally, marine aquaculture monitoring relied on visual interpretation of satellite images, a labor-intensive process dependent on human expertise. However, the enormous volume of data generated annually has surpassed the capacity for manual analysis. This is where artificial intelligence comes into play.
Machine learning techniques such as support vector machines, decision trees, and random forests have been applied with some success, but they require deep domain knowledge to define relevant image features.
The Solution: Artificial Intelligence to the Rescue
Deep learning has revolutionized the field of computer vision, offering a promising alternative to traditional machine learning. By eliminating the need to manually define features, deep learning models can automatically extract relevant information from images.
Models such as DeepLab-v3+, U-Net, and convolutional neural networks have proven effective in detecting marine aquaculture areas. Specialized models like LFCSDN have even achieved outstanding results.
However, one technology that has shown particular promise is the YOLO (You Only Look Once) algorithm. Originally designed for object detection in general images, YOLO has been successfully adapted to identify various marine objects, such as ships.
This study focuses on the application of the YOLOv5 and YOLOv7 models, known for their speed and accuracy, to identify offshore aquaculture areas using high-resolution satellite images.
YOLOv5 vs. YOLOv7: Exceptional Performance
The results obtained were surprising. YOLOv5 outperformed YOLOv7 in all performance indicators, including precision, recall, mean average precision (mAP), and F1 score. This suggests that YOLOv5 might be the preferred option for offshore aquaculture monitoring applications based on satellite images.
The Resolution Factor: A Double-Edged Sword
One of the most innovative aspects of this study was the exploration of using super-resolution (SR) methods to enhance the spatial resolution of satellite images and assess whether this influenced the accuracy of the YOLO models. The Real-ESRGAN method was used to increase the clarity and resolution of images before applying the detection models.
Surprisingly, the results indicated that despite improving image clarity, SR methods had a negative effect on the performance of YOLO models for detecting marine aquaculture areas. This finding suggests that enhancing image resolution does not always result in better performance of deep learning models, at least in the context of object detection in marine environments.
Conclusions and Implications
The study demonstrates the potential of YOLO models to revolutionize offshore aquaculture monitoring. YOLOv5, in particular, stands out as an accurate and efficient tool for identifying these areas in satellite images. However, it is essential to consider the unintended effects of super-resolution techniques on model performance.
This work represents a significant step toward more sustainable management of marine aquaculture. By providing a reliable methodology for monitoring these areas, effective strategies can be developed to mitigate environmental impacts and ensure the health of coastal ecosystems.
The study was funded by the Director’s Foundation of South China Sea Bureau of Ministry of Natural Resources; the Marine Economy Special Project of the Guangdong Province. The APC was funded by the Science and Technology Project of Guangdong Forestry Administration (2024): Monitoring and Ecological Value Assessment of Coastal Wetland Resources in the Guangdong Province.
Reference (open access)
Dong, D., Shi, Q., Hao, P., Huang, H., Yang, J., Guo, B., & Gao, Q. (2024). Intelligent Detection of Marine Offshore Aquaculture with High-Resolution Optical Remote Sensing Images. Journal of Marine Science and Engineering, 12(6), 1012. https://doi.org/10.3390/jmse12061012
