Analysis of Deep Learning Based Garbage Detection in Water Bodies

The pollution of water bodies is an important problem for society. Harmful substances like nitrogen, lead, pesticides and oil spills affect aquatic life. The visualization of garbage gets difficult resulting in poor image resolution in water bodies. Object detection methodologies used in marine litter detection such as background subtraction and frame difference cannot distinguish the shape of garbage and background due to low resolution of images. The deep learning with transfer learning-based methodologies like ResNet50 and VGG19 handles the low-resolution images and performs effectively the garbage classification and detection. The main objective of the proposed research work is to contribute marine ecosystems maintenance by reviewing and analyzing the garbage detection and classification in water bodies. It reviews the significance of garbage detection in water bodies, open problems in garbage detection, role of deep learning in garbage detection, working of standard convolutional neural networks such as VGG19(Visual Geometry Group) and ResNet50 (Residual Network). The CNN, VGG19 and ResNet50 models are experimented with the AquaTrash dataset. Accuracy is utilized as the performance metric leveraged to represent performance of the methodologies. The experimental result shows that VGG achieved an improved garbage detection compared to others.