Analyzing the Impact of Integrated Signal and Image Processing Techniques on Pattern Recognition and Decision Accuracy in Complex Systems

Integrated signal and image processing is being used more to enhance pattern recognition and decision quality in complex sensing conditions, but the quality of benefit of combining underwater sonar and camera data to ensure continuous localization is context-dependent. This work quantifies the effect of feature-level fusion of multibeam sonar and monocular camera imagery on diver localization accuracy. A publicly available underwater dataset wassynchronized to produce 1,193 paired samples across four experimental sets. Each frame was converted to grayscale, resized to 64×64 pixels, and represented using global intensity statistics and regional mean features. Three Random Forest regression models were trained under identical conditions: sonar-only, camera-only, and fused features. Localization error was evaluated using Euclidean distance and standard regression metrics. Sonar-only processing produced the highest error, camera-only significantly reduced error, and multimodal fusion achieved the lowest overall RMSE and MAE. The greatest improvements from fusion were observed in scenarios with higher spatial variability, while in visually favorable conditions, camera-only performance remained competitive. These findings indicate that integrating acoustic and visual features enhances overall localization robustness, though the magnitude of improvement depends on environmental characteristics. Adaptive fusion strategies may further improve stability in real-world underwater systems.