Detection of Plastic Waste in Ocean Using Machine Learning Based Bi- LSTM With Triplet Attention Mechanism

Underwater pollution is more frequent in recent times and the river that carrying tons of plastic which merged into the sea and pollute the aquatic ecosystem. Globally it is somewhat difficult to track the flow of plastic waste from river and submerged into ocean. The Plastic Overshoot Day (POD) report reported that India will release 391,879 tonnes of microplastics in water bodies. This is crucial and need to take preventive steps to stop expanding of this quantity. Underwater plastic waste detection is quite challenging task and government should spend huge amount to equip underwater robots to grasp waste and collect it. The light attenuation in ocean causes the colour cast, low visibility and weak illumination of underwater captured images. To assist underwater robots in recognising and classifying plastic waste, a computational technique is required. This study proposes a comparative analysis of deep learning (DL) algorithms such as Multi-Layer Perceptron (MLP), Convolutional Neural Network (CNN), U-Net, Bi-LSTM, and Bi-LSTM with Triplet attention Mechanism to classifying underwater plastic waste. To boost the performance of classification, the proposed study adopts a pre-processing technique namely total variation denoising (TVD) with $\mathbf{Z}$ score normalisation technique. Both unmanned aerial vehicle (UAE) captured images and satellite images are used for this study. The models were designed and developed to detect plastic waste under the epipelagic layer of the ocean. Evaluating the performance of selected modes was done with parameters namely accuracy, precision, F1 score, and PSNR.