BAOS-CNN: A novel deep neuroevolution algorithm for multispecies seagrass detection.

Area of Science:

• * Marine Biology
• * Computer Science
• * Artificial Intelligence

Background:

• * Deep learning, particularly Convolutional Neural Networks (CNNs), shows promise for seagrass image identification.
• * Manual architectural engineering and hyperparameter tuning of CNNs are resource-intensive.
• * Automating CNN design is crucial for advancing marine ecological monitoring.

Purpose of the Study:

• * To propose a Deep Neuroevolutionary (DNE) model for automating CNN architectural engineering and hyperparameter tuning.
• * To introduce a novel metaheuristic algorithm, Boosted Atomic Orbital Search (BAOS), for optimizing CNNs.
• * To evaluate the performance of the proposed BAOS-CNN model in seagrass image identification.

Main Methods:

• * Development of the Boosted Atomic Orbital Search (BAOS) algorithm, an enhancement of the Atomic Orbital Search (AOS) incorporating Lévy flight.
• * Implementation of a Deep Neuroevolutionary (DNE) model (BAOS-CNN) utilizing the BAOS algorithm.
• * Training and evaluation of the BAOS-CNN model on a multi-species seagrass dataset and comparison with six other optimization algorithms.

Main Results:

• * The BAOS-CNN model achieved the highest overall accuracy (97.48%) among seven evolutionary-based CNN models on the multi-species seagrass dataset.
• * State-of-the-art overall accuracy was obtained on the 'DeepSeagrass' dataset (92.30% for four classes, 93.5% for five classes).
• * The proposed method effectively automates CNN design for complex ecological image analysis.

Conclusions:

• * The proposed BAOS-CNN model offers an effective automated approach for CNN architectural engineering and hyperparameter tuning.
• * This method significantly improves the accuracy of seagrass image identification and mapping.
• * The developed algorithm holds potential for revolutionizing marine ecological research through advanced AI applications.