Comparative analysis of marine debris simulation using ensemble learning with XGBoost and deep convolutional neural networks

Marine ecosystems, wildlife, and human activities are seriously threatened by marine garbage. Deep learning-based systems for categorisation can automate classifying distinct types of marine debris from photos or video recordings, allowing for more effective and precise monitoring and assessment of debris levels in different maritime circumstances. DL is a useful tool that can help with environmental conservation efforts by categorising marine waste. To improve classification accuracy, sensitivity, and specificity for different types of marine debris, we investigate the use of ensemble learning approaches in this work and used for execution in Python. We compare three distinct implementations of the powerful gradient boosting method XGBoost with innovative deep convolutional neural networks: XGBoost with Adam and GoogleNet optimiser, XGBoost with VGG19 and Adam optimiser, and XGBoost with ResNet and Adam optimiser. The XGBoost algorithm and feature extraction from three different pre-trained CNN architectures, GoogLeNet, VGG19, and ResNet, are used in this study to examine the effectiveness of classifying maritime debris. We highlight the outstanding results obtained by combining ResNet + Adam with XGBoost, a reliable and effective method for classifying maritime trash and producing an accuracy of 91%, specificity of 0.88, and sensitivity of 0.91, respectively.