Simultaneous mass estimation and class classification of scrap metals using deep learning

While deep learning has helped improve the performance of classification, object detection, and segmentation in recycling, its potential for mass prediction not yet been explored. Therefore, this study proposes a system with without feature extraction selection, including principal component analysis (PCA). These methods are evaluated on combined Cast (C), Wrought (W) Stainless Steel (SS) image dataset using state-of-the-art machine algorithms prediction. After that, best framework is DenseNet classifier, resulting multiple outputs that perform both classification The proposed architecture consists neural network backpropagation (BPNN) prediction, which uses up to 24 features extracted from depth images. method obtained 0.82 R2, 0.2 RMSE, 0.28 MAE regression 95% C&W test DenseNet+BPNN+PCA model. DenseNet+BPNN+None model selected (None) used CW&SS data had lower 80% (0.71 0.31 0.32 MAE). presented monitoring composition waste streams optimize robotic pneumatic sorting systems by providing better understanding physical properties objects being sorted.