Deep-Feature-Based Approach to Marine Debris Classification

The global community has recognized an increasing amount of pollutants entering oceans and other water bodies as a severe environmental, economic, social issue. In addition to prevention, one the key measures in addressing marine pollution is cleanup debris already present environments. Deployment machine learning (ML) deep (DL) techniques can automate waste removal, making process more efficient. This study examines performance six well-known convolutional neural networks (CNNs), namely VGG19, InceptionV3, ResNet50, Inception-ResNetV2, DenseNet121, MobileNetV2, utilized feature extractors according three different extraction schemes for identification classification underwater debris. We compare network (NN) classifier trained on top CNN when extractor (1) fixed; (2) fine-tuned given task; (3) fixed during first phase training afterward. general, fine-tuning resulted better-performing models but much computationally expensive. overall best NN showed Inception-ResNetV2 with accuracy 91.40% F1-score 92.08%, followed by InceptionV3 extractor. Furthermore, we analyze conventional ML classifiers’ features extracted CNNs. Finally, show that replacing classifier, such support vector (SVM) or logistic regression (LR), further enhance new data.