Vision Based Floating Garbage Classification using SIFT

Pollution caused due to unregulated dumping of waste in water bodies has been a problem for some time now. Water Pollution contributes to a lot of environmental as well as social hazards and poses a risk to the public health and well-being of humans. Major Steps are being taken to limit this type of pollution but it’s far from being completely tackled. This study proposes a vision-based approach to identifying different types of water pollutants. Classification-based machine learning algorithms like KNN, Random Forest, XGBoost, SVM, etc. are used on a custom-made dataset containing 1500 images of waste. Feature extraction of all the images is done using the SIFT feature extraction algorithm which is then fed to the classifiers. Machine learning techniques like Dimensionality reduction, K fold and Grid Search Cross Validation are used to improve the accuracy and optimize model performance. The proposed model boasts a highest accuracy of 95 percent with a ROC-AUC score of 0.97. The result of the proposed study is an efficient classification of floating waste in different target classes.