Algae, with their abundance of nutrients and health benefits, are a rising star in the food industry. Ensuring consistent quality requires efficient classification and processing; however, traditional processing methods, often relying on manual inspection, struggle to keep pace with growing demand and quality standards. This is where cutting-edge technology steps in, offering a revolutionary upgrade.
A new study published by scientists from Jimei University and the Fujian Provincial Key Laboratory of Food Microbiology and Enzyme Engineering applied computer vision based on deep transfer learning to identify lower seaweeds, including those from the third and fourth harvests and impure algae in this work.
Artificial Intelligence for Macroalgae Classification
Various artificial intelligence technologies have been employed in seaweed research and aquaculture for identification, biomass estimation, model construction for predicting biomass production, disease identification, etc.
Chen et al., (2022) used machine learning methods to monitor macroalgae in the intertidal zone; while Acebo et al., (2020) employed data mining techniques to develop a model to predict seaweed production.
Detecting Imperfections in Seaweed
Seaweed quality depends on identifying impurities and collecting them at the optimal time. Unfortunately, current processes are slow, subjective, and prone to human errors. Imagine sorting through thousands of delicate strands, distinguishing subtle differences in texture and color: a tedious and error-prone task.
Accurately classifying seaweed by harvest period, identifying inferior quality, and detecting impurities are crucial for maintaining quality and optimizing production speed. However, current methods often lack the speed, accuracy, and objectivity required for large-scale operations.
Enter deep learning, the AI superhero
This research introduces a turning point: computer vision based on deep transfer learning. Imagine an artificial intelligence trained with mountains of data, equipped to recognize even the subtlest nuances of seaweed. Here’s how it works:
- Master models: Pre-trained AI models like YOLOv8 and YOLOv5, perfected on vast visual datasets, form the foundation.
- Fine-tuning for seaweed specificity: These models are then “reprogrammed” to focus on specific types of algae and relevant impurities for the processing plant.
- Real-time classification and detection: The trained model analyzes live images, automatically classifies algae by harvest period (including identifying inferior third and fourth harvests) and detects four common types of impurities.
The results: a quantum leap in efficiency and quality
The research demonstrates notable success:
- Increased accuracy: Harvest period classification achieved a superior accuracy of 93.5%, a 16% improvement over traditional methods.
- Impurity detection: YOLOv8n achieved an outstanding average precision of 99.14% in impurity detection.
- Speed and efficiency: Processing a single image takes just 4.3 ms, ensuring real-time analysis on production lines.
- Compact power: At less than 6 MB, these AI brains are powerful and lightweight.
Benefits for the seaweed processing industry
This study showcases the immense potential of AI to revolutionize seaweed processing. The benefits are numerous:
- Improved quality control: Objective and automated inspection ensures consistent, high-quality seaweed products.
- Enhanced production efficiency: Faster and more accurate analysis leads to higher processing speed and yield.
- Economic benefits: Waste reduction, higher quality, and faster production translate into significant cost savings and profit gains.
- Smarter seaweed industry: This study paves the way for intelligent automation, propelling the entire industry towards greater efficiency and sustainability.
Conclusion
By embracing AI-based classification, the seaweed industry can unlock a new era of efficiency, sustainability, and quality. This study represents a significant step towards achieving this vision, paving the way for a smarter and more sustainable future for seaweed production.
The study was funded by the Natural Science Foundation of Fujian Province of China, National Key R&D Program of China, and National Natural Science Foundation of China.
Contact
Honghao Cai
Department of Physics, School of Science, Jimei University, Xiamen, Fujian Province, China
Email: hhcai@jmu.edu.cn
Main Reference
Gao, Z., Huang, J., Chen, J. et al. Deep transfer learning-based computer vision for real-time harvest period classification and impurity detection of Porphyra haitnensis. Aquacult Int (2024). https://doi.org/10.1007/s10499-024-01422-6
Other References
Acebo, J. G., Feliscuzo, L. S., & Romana, C. L. C. S. (2021). Model Development in Predicting Seaweed Production Using Data Mining Techniques. In Advances in Computer, Communication and Computational Sciences: Proceedings of IC4S 2019 (pp. 843-850). Springer Singapore.
Chen J, Li X, Wang K, Zhang S, Li J, Sun M (2022) Assessment of intertidal seaweed biomass based on RGB imagery. PLoS ONE 17(2): e0263416. https://doi.org/10.1371/journal.pone.0263416
