We study the problem of underwater debris classification and removal by remotely operated vehicles. This task is particularly important for subsea oil and gas fields exploitation. The classification of underwater debris is a challenging and difficult problem because of the complexity of underwater environments. We investigate four different algorithms based on deep convolutional neural networks for detecting and classifying marine debris. The proposed techniques are built on Keras and Tensorflow using Python programming environment. To train the algorithm for detection, various dataset information containing different types of marine debris have been established. Four distinct classifier and activation function combinations have been compared experimentally. The dataset is consist of fifteen category. The suggested approach is a modified VGGNet-16 trained on the dataset. The use of a sigmoid classifier and the Relu activation function to categories marine improves classification accuracy. The overall result indicates that classification accuracy on the testing set is satisfactory.