Improving The Performance of Object Detection in Underwater with Advanced Deep Learning Based Noise Reduction Techniques

In order to identify objects that could appear suspicious, this investigation suggests an extensive method that includes pre-processing, ‎enhancement, and noise elimination from submarine images. The method integrates the strengths of SIFT (Scale-Invariant Feature ‎Transform) for feature extraction, CLAHE (Contrast Limited Adaptive Histogram Equalization) for enhancing underwater images to ‎accurately identify features, and DnCNN (Deep Convolutional Neural Network) for effective noise removal through dedicated training. The ‎method incorporates edge enhancement, color correction, brightness modulation, and preparatory processing and de-noising. The purpose of ‎these enhancements is to visually highlight suspicious objects in the underwater images. To calculate the effectiveness of the proposed ‎method, testing has been conducted on a variety of underwater image datasets with various settings and suspicious objects. Analysis was ‎done in contrast to current methods for pre-processing, noise reduction, and object recognition. The results were evaluated using measurable ‎performance metrics as SSIM, MSE, and PSNR. The experiment's findings demonstrate that the combined approach outperforms the ‎individual methods, resulting in a higher rate of detection the 71.2% in Dn-CNN and 61.1 in CLAHE, by making suspicious items appear ‎larger and improving the overall clarity and quality of underwater images. Numerous applications in military operations, offshore ‎technology, maritime research and rescue, aquaculture, and other domains make it an essential sensor technology.