Journal of Cybersecurity and Information Management JCIM 2690-6775 2769-7851 10.54216/JCIM https://www.americaspg.com/journals/show/3911 2019 2019 Research Scholar, Department of Electrical & Electronics and Communication Engineering, GITAM (Deemed to be University), Visakhapatnam, Andhra Pradesh, India Maddukuri Maddukuri Professor, Department of Electrical & Electronics and Communication Engineering, GITAM (Deemed to be University) Visakhapatnam, Andhra Pradesh, India J. B. Seventline This paper discusses about implementing Machine Learning Models with the Marine_Pulse dataset. This is about side-scan sonar images that have four groups: Engineering Platform (EP), PipelineCable (PC), Sea Bed Surface (SBS), and Underwater Residual Mound (URM). This manuscript performed some difficult feature extraction and classification methods using the DenseNet-DNN framework. This paper delves deeply into the implementation of the DenseNet121 Dropout, DenseNet201 Dropout, DenseNet201 Enhanced Dropout, and DenseNet201 Transfer Learning models. It investigates how these models perform on feature extraction and classification using a DNN. We enhanced the performance and reduced overfitting by applying a dropout to DenseNet121 and DenseNet201 and by adding transfer learning (TL) to DenseNet201, respectively. The models were evaluated based on the accuracy, precision, recall, F1-score, specificity, and classification errors of the training and testing samples. We observed that DenseNet201 Enhanced Dropout outperformed the other models, achieving the highest accuracy of 95.79%. DenseNet201 Dropout followed this achievement with an accuracy of 94.74% and DenseNet121 Dropout with an accuracy of 92.11%. DenseNet201 Transfer Learning, on the other hand, had the worst accuracy (92.11%).  Specificity is a measure of how well the model represents negative examples correctly. The maximum specificity was observed in DenseNet201 Enhanced Dropout (98.38%). DenseNet201 Dropout follows it at 97.96% and DenseNet121 Dropout at 97.18%. The smallest specificity was reported on DenseNet201 TL with 96.60%. This result demonstrates that our keys can generalize well and that they maintain high classification accuracy on the test data. 2025 2025 187 208 10.54216/JCIM.160213 https://www.americaspg.com/articleinfo/2/show/3911