Fish vitality assessment under adversity stress based on multi-sensing fusion and deep learning techniques

Abstract

Accurately assessing the vitality of fish during Cryogenic Waterless Live Transportation (CWLT) is crucial for ensuring their survival rate and transport quality. To address this, this study developed a multi-sensing fusion measurement system, integrating the impedance sensor and image sensor to capture fish impedance, phase angle data and visual image data, which can provide reliable data support for vitality assessment. A multi-input deep learning model for cross-modal data fusion, called the whale optimization algorithm-based dual-channel convolutional network (WOA-DConvNet), was developed to assess fish vitality. The study first conducted a correlation analysis between impedance and phase angle data using blood index data, revealing the relationship between different frequency signals and fish vitality status. Cluster analysis and variance analysis were then performed to examine the significant differences in vitality states. The results show significant differences (P < 0.05) in impedance and phase angle data between different vitality levels. Finally, the WOA-DConvNet model was used to evaluate the vitality of pearl gentian grouper during the live transport. The results showed that the model achieved excellent performance, achieving precision, recall, f1 score and accuracy of 91.81 %, 91.74 %, 91.76 % and 91.71 %. The core goal of this research is to explore the integrated application of artificial intelligence and advanced sensing technologies in the live transport process, aiming to develop a more precise and intelligent vitality assessment method and promote scientific monitoring and management during live fish transportation.