A Spatial-Spectral Classification Method Based on Deep Learning for Controlling Pelagic Fish Landings in Chile.

Area of Science:

• Marine Biology and Ecology
• Artificial Intelligence and Machine Learning
• Fisheries Science and Management

Background:

• Overexploitation of fish stocks threatens marine ecosystems and the fishing industry.
• Accurate monitoring of fish landings is crucial for sustainable resource management and quota enforcement.
• Current methods for fish species identification can be labor-intensive and prone to errors.

Purpose of the Study:

• To develop and evaluate a deep learning-based spatial-spectral method for classifying five important pelagic fish species.
• To assess the potential of this method for automated monitoring of fish landings in the Chilean fishing industry.
• To improve the accuracy and efficiency of fish species identification for fisheries management.

Main Methods:

• A convolutional neural network (CNN) architecture with two processing channels was employed.
• The CNN processed both Red-Green-Blue (RGB) images and visible and near-infrared (VIS-NIR) reflectance spectra of fish samples.
• Five pelagic species, including *Engraulis ringens*, *Merluccius gayi*, *Strangomera bentincki*, *Normanichthtys crockeri*, and *Stromateus stellatus*, were classified.

Main Results:

• The proposed deep learning model achieved classification accuracy exceeding 94% across all performance metrics.
• The spatial-spectral approach demonstrated superior performance compared to existing state-of-the-art techniques.
• The method successfully differentiated between targeted and non-targeted fish species.

Conclusions:

• The developed deep learning method shows significant potential for automated, accurate fish species classification.
• This technology can aid in the effective monitoring of fish landings and ensure compliance with fishing quotas.
• Implementing this method can contribute to the sustainable management of marine resources.