Similarity Analysis of Waste Images Through Feature Extraction

Large-scale waste disposal is a process that requires significant effort and cost. Recently, various countries have been attempting to automate the process to reduce these costs. In particular, detecting the similarity of waste objects in the automation of waste disposal processes is an important issue. In this study, we used the SIFT (Scale-Invariant Feature Transform) algorithm for image feature extraction to extract features of waste objects. We also extracted color information in the HSV (Hue, Saturation, Value) color space for image similarity analysis. The extracted features were validated using XGBoost and Artificial Neural Networks. We used a dataset of 81,072 images of large-scale waste for training and evaluation. We conducted additional experiments to compare the traditional method of SIFT matching and threshold application with machine learning-based models and to evaluate the contribution of HSV to machine learning performance. In threshold-based measurements, SIFT and HSV achieved accuracies of 0.82 each. In machine learning, XGBoost achieved an accuracy of 0.93, and the Artificial Neural Network using CNN achieved an accuracy of 0.93. This represents a maximum performance improvement of 10% compared to the method using only SIFT.