Water contamination analysis in IoT enabled aquaculture using deep learning based AODEGRU

Abstract

Water contamination presents a significant challenge in aquaculture, impacting the sustainability of ecosystems and the health of aquatic organisms. Precisely assessing water contamination levels is crucial for effective monitoring and safeguarding aquatic life within the aquaculture industry. Traditional methods for evaluating water contamination are characterized by their costliness, time-consuming nature, and susceptibility to errors. Integrating computer technologies such as Artificial Intelligence (AI), the Internet of Things (IoT), and Data Analytics offers promising potential in addressing this issue. Nevertheless, current deep learning solutions have limitations related to data variability, interpretability, and performance. To address these limitations, this study proposes a comprehensive framework that incorporates IoT-based data collection and data segregation techniques to enhance the accuracy of water contamination classification in aquaculture. Real-time data collected through IoT devices, encompassing parameters like temperature, pH levels, dissolved oxygen, nitrate concentration, and other water quality indicators, enables a holistic evaluation of water quality. By considering predefined acceptable ranges for aquatic life, this framework calculates a water contamination index, facilitating the classification of data into categories such as contaminated and non-contaminated. To ensure robust classification, the study introduces an innovative attention-based model known as the Ordinary Differential Equation Gated Recurrent Unit (AODEGRU). This attention mechanism directs the model's focus towards salient features associated with water contamination, while the AODEGRU architecture captures temporal patterns within the data. Experimental results underscore the effectiveness of the proposed model. It demonstrates its superiority with high performance, achieving an accuracy rate of approximately 98.69% on a publicly available dataset and an impressive 99.89% accuracy on a real-time dataset, clearly outperforming existing methodologies.