The proposed work presents an advanced and efficient framework for ship detection in remote sensing imagery by integrating deep learning techniques with feature- driven visual analysis. The proposed method combines the strengths of a Convolutional Neural Network (CNN) for precise ship body detection and a feature-based approach for the accurate identification of ship wakes. Recognising the challenges posed by complex sea backgrounds, including cloud cover, foam, and occlusions, the method introduces a robust pre-processing pipeline to segment the foreground target from noisy surroundings. To address limitations of earlier approaches confined to clear and isolated scenarios, a multiscale feature extraction strategy is incorporated, allowing effective detection of ships with diverse sizes and orientations. A sea-region modelling step based on statistical analysis enhances the ability to distinguish wakes from background patterns, which is critical for inferring sailing direction and identifying partially visible or cloud-covered vessels. Crucially, a Particle Swarm Optimisation (PSO)-based refinement mechanism is integrated into the detection pipeline to fine-tune bounding box predictions generated by the CNN. This hybrid CNN-PSO approach significantly improves localisation accuracy and reduces false positives, especially in densely populated maritime scenes. Through extensive experimentation on large-scale remote sensing datasets, the proposed system achieves a remarkable detection accuracy of 99%, outperforming traditional techniques and establishing its potential for real-world maritime surveillance applications