An improved dung beetle optimization with recurrent convolutional neural networks for efficient detection and classification of undersea water object images

• Published in: Earth science informatics Vol. 17; no. 4; pp. 3651 - 3671
• Main Authors: Jasmine, J. Jeno, Raja, S. Edwin, Muniraj, R., Jarin, T.
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01-08-2024; Springer Nature B.V
• Subjects: Algorithms; Artificial neural networks; Beetles; Classification; Clustering; Computer vision; Deep learning; Deep sea; Deep sea environments; Dung; Earth and Environmental Science; Earth Sciences; Earth System Sciences; Effectiveness; False alarms; Feature extraction; Image enhancement; Image filters; Image quality; Image segmentation; Information Systems Applications (incl.Internet); Machine learning; Marine resources; Neural networks; Object recognition; Ontology; Operational hazards; Optimization; Performance evaluation; Robustness; Sensitivity analysis; Simulation and Modeling; Space Exploration and Astronautics; Space Sciences (including Extraterrestrial Physics; Underwater; Image quality enhancement; Adaptive spectral clustering; Improved dung beetle optimization; Anisotropic diffusion filter; Undersea water object detection; And Recurrent convolutional neural network
• ISSN: 1865-0473; 1865-0481
• DOI: 10.1007/s12145-024-01358-8

The exploration of the underwater environment has become increasingly important due to the utilization and development of deep-sea resources in recent years. To overcome the hazards of high-pressure deep-sea conditions, autonomous underwater operations have become essential, with intelligent computer vision playing a pivotal role.This study proposes a novel deep-learning model for the effective detection and classification of underwater object images (UWOI). The model addresses the challenge of low-quality, weak illumination, and noise in underwater images by employing an Anisotropic Diffusion Filter (ADF) during pre-processing. To enhance segmentation accuracy, the model utilizes Adaptive Spectral Clustering (ASC). Textural and statistical features are then extracted using the Gray Level Co-occurrence Matrix (GLCM) for robust feature representation. Finally, the proposed model leverages an Improved Dung Beetle Optimization (IDBO) algorithm in conjunction with a Recurrent Convolutional Neural Network (RCNN) for UWOI detection and classification. Extensive evaluations demonstrate that the proposed model achieves significantly improved performance compared to previous methods, attaining superior results in terms of accuracy, Dice score, sensitivity, Structural Similarity Index (SSIM), and specificity. The proposed method consistently demonstrates strong performance in both specificity and sensitivity compared to existing methods, with specificity ranging from 94 to 97% across iterations (10–100), exceeding existing methods (SDCS, UCPS, AEA-QoS, and RLOD), and sensitivity ranging from 94% to 96.65% across iterations (10–50), with the value rising to 96.65% at the 100th iteration. Overall, the findings suggest that the proposed method achieves both high true positive rates (specificity) and low false negative rates (sensitivity), indicating its effectiveness in correctly identifying true targets and minimizing false alarms compared to existing methods. This work contributes to the advancement of underwater object recognition by offering a robust and efficient deep-learning approach.