Answers are grounded in this paper and cite their sources.
Computational study demonstrates high predictive accuracy in underwater IoT monitoring, suggesting robust analytics for sustainable aquaculture.
Growing needs for the management of sustainable water resources and improvement of aquaculture production efficiency have promoted the development of IoT-based underwater monitoring systems. Conventional approaches based on simple sensor networks are subject to severe constraints, such as limited scalability, high energy consumption, low accuracy in dynamic scenarios, and poor support for data processing. In addition, prior art models are sensitive to noise; they cannot deal with missing data, and they fail in modeling long-range dependence among aquatic parameters. To this end, a novel UT-based UDW analytics framework is introduced by combining WT, AE and TN. The WPT further decomposes an intricate underwater signal into sub-time–frequency components at different scales to extract non-stationary features effectively. Autoencoders learn compact latent representations by encoding and decoding non-linear features from noisy sensor data, while Transformer Networks learn long-range temporal dependencies via self-attention mechanisms for predictive accuracy in sequential prediction. The proposed hybrid method facilitates better feature learning, noise reduction, and generalization under various underwater environmental scenarios. The results of the experiments demonstrate better performance than the state-of-the-art methods such as ResNet, YOLOv7, and MLCNN, with an accuracy of ∼99.5%, which brings corresponding enhancements in the precision (≈98.9%), recall (≈98.7%), and F1-score (≈98.8%) in important underwater prediction utilization.
No takes yet. Share an insight, caveat, or question.
Loey et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: