A Multi-Indexes and Non-Invasive Fish Health Assessment System with Deep Learning and Impedance Sensing

Abstract

Monitoring the health status of economically valuable fish during waterless transportation poses challenges. To address this issue, we developed a multi-indexes and non-invasive assessment system (MiNiAS) leveraging wearable electrical impedance sensors and deep learning technology. The system integrates multiple stress indexes to measure the health status of live fish and successfully validates the accuracy of singular spectrum analysis combined with deep learning and the k-nearest neighbor (SSA-DL-KNN) health classification model. Evaluation metrics such as MAE, MAPE, and RMSE indicate the CNN-LSTM model's robust performance in predicting fish stress indexes. The KNN algorithm demonstrates high accuracy in classifying health levels, with 96% and 98% accuracy for large and small fish, respectively. Hourly assessment accuracy exceeds 90%. Our findings provide valuable insights into non-invasive health assessment for live fish, offering a significant technical reference for fishery applications.

Keywords: Electrical impedance, Stress indexes, Noninvasive assessment, Deep learning, KNN classification