Sonar Objective Detection Based on Dilated Separable Densely Connected CNNs and Quantum-Behaved PSO Algorithm.

Area of Science:

• Ocean exploration technology
• Artificial intelligence in marine science
• Signal processing for underwater acoustics

Background:

• Underwater sonar objective detection is crucial for ocean exploration.
• Complex environments pose challenges for existing sonar detection methods.
• Need for robust and accurate sonar detection techniques.

Purpose of the Study:

• To propose an advanced sonar objective detection method using DS-CNNs and QPSO.
• To enhance feature extraction and reduce information loss in sonar image analysis.
• To optimize neural network parameters for improved detection accuracy and robustness.

Main Methods:

• Developed dilated separable convolution kernels to expand receptive fields and improve feature extraction.
• Introduced a multisampling pooling (MS-pooling) operation to minimize feature loss and restore image resolution.
• Utilized the quantum-behaved particle swarm optimization (QPSO) algorithm to optimize neural network weight parameters.

Main Results:

• The proposed DS-CNNs achieved a detection accuracy of 96.98% on a sonar image dataset.
• Demonstrated superior detection effects compared to existing methods.
• Exhibited enhanced robustness in complex underwater environments.

Conclusions:

• The DS-CNNs combined with QPSO offer a highly effective solution for underwater sonar objective detection.
• The method significantly improves detection accuracy and robustness.
• This approach advances capabilities in ocean exploration and marine surveillance.