An integration of ensemble deep learning with hybrid optimization approaches for effective underwater object...

Area of Science:

• Computer Vision
• Deep Learning
• Marine Biology

Background:

• Underwater object detection (UOD) is crucial for marine environmental studies and species protection.
• Existing terrestrial object recognition methods struggle with underwater challenges like small, obstructed, and clustered objects, and limited computational resources.
• Recent advancements in deep learning (DL) and computer vision (CV) have significantly improved image-based UOD.

Purpose of the Study:

• To develop an advanced technique for underwater object detection and classification.
• To address the limitations of current UOD methods in challenging underwater environments.
• To improve the accuracy and robustness of underwater object recognition.

Main Methods:

• Proposed the Underwater Object Detection and Classification Utilizing the Ensemble Deep Learning Approach and Hybrid Optimization Algorithms (UODC-EDLHOA) technique.
• Employed EfficientNetB7 with an attention mechanism for feature extraction and YOLOv9 for object detection.
• Utilized an ensemble of deep neural network (DNN), deep belief network (DBN), and long short-term memory (LSTM) for UOD, with hyperparameter tuning via hybrid Siberian tiger and sand cat swarm optimization (STSC).

Main Results:

• The UODC-EDLHOA model demonstrated robust classification performance on the UOD dataset.
• Achieved a superior accuracy of 92.78% in underwater object detection and classification.
• Outperformed existing techniques in validation experiments.

Conclusions:

• The UODC-EDLHOA technique offers a significant advancement in underwater object detection and classification.
• The hybrid optimization and ensemble deep learning approach effectively addresses underwater imaging challenges.
• This method holds great potential for enhancing maritime environmental monitoring and underwater species protection.