Predictive Analytics on Maritime object detection and Classification Accuracy Using Multi-Layer Perceptron in Comparison With Residual Networks (ResNets)

Our study investigates and compares the efficacy of Multi-Layer Perceptrons (MLPs) and Residual Networks (ResNets) for maritime object detection, with a specific emphasis on identifying plastic and garbage pollution. The objective is to determine the most effective neural network architecture for enhancing accuracy in maritime safety and environmental monitoring. In this research, both MLP and ResNet algorithms were tested, with each being executed over 10 iterations (N=10). We evaluated these algorithms across two separate groups, incorporating a total of 100 samples in the analysis. To ensure the robustness of our findings, we utilized the Gpower statistical test, setting the power to 85% with α=0.05, which aligns with standard statistical practices for detecting significant differences or effects. The results indicated that MLPs achieved an accuracy rate of 93.86%, surpassing ResNets, which had an accuracy of 91.73%, in the context of maritime object detection focusing on plastic and garbage. While MLPs demonstrated superior performance, it is crucial to consider factors beyond mere accuracy, such as the adaptability of the models to varying environmental conditions and practical implementation challenges. To sum up, our results highlight the superior detection accuracy of Multi- Layer Perceptrons (93.86%) over Residual Networks (91.73%) when it comes to identifying marine waste, specifically plastic and garbage. However, choosing amongst these models should also include how well they can be applied in real-world scenarios and how well they can adapt to various marine conditions.