Deep learning computer vision for the separation of Cast- and Wrought-Aluminum scrap

Enhanced Plastic Recycling using RGB+Depth Fusion with MassFaster and MassMask R-CNN

“…As shown in bold in Table II, the best object detection results are generally obtained by combining RGB and ED images, which show, in the case of Faster R-CNN, an increase of 10% in AP50:95 when compared to RGB images. Previous work has shown that mid-to-late fusion of RGB and depth images can improve object recognition [23], [27]. For the Mask R-CNN, there is a slight difference between whether the RGB and RGBD or RGBED inputs are used.…”

“…However, for RGBD fusion, the RGB and D images are fed into two separate subnetworks with the same architecture. Then, the parameters of layers C2 and C3 are concatenated in both subnetworks to obtain a uniform output [23]. In the following step, ResNet50_FPN_RGB, ResNet50_FPN_RGBD, or ResNet50_FPN_RGBED will denote either the ResNet50_FPN architecture for pure RGB images or the modified ResNet50_FPN for the fusion of RGB and D/ED images.…”

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

“…Both systems were implemented in an automatic sorting machine, showing a sorting performance average accuracy of up to 94.71% ± 1.69 (Zhang et al, 2021). In an earlier study, the presented research built on the classification of C&W Al by evaluating five CNN Deep Learning models and two transfer learning methods (Díaz-Romero et al, 2021). This study showed that the fusion of RGB and 3D images at the last layer of the DenseNet network improves the classification of the evaluated dataset.…”

“…A dataset of 120 C, 428 W Al scrap samples and 134 SS samples of different shapes (e.g., compact, bar, sheet, pipe, and irregular) with a mass distribution between 5 to 200 grams (g) was collected from a Belgian recycling facility. The Wrought and Cast pieces were used in a previous study to classify Al scraps (Díaz-Romero et al, 2021). The 548 Al samples (C&W) were collected randomly from the Twitch fraction.…”

“…In this regard, X-Ray Fluorescence (XRF) or Laser-Induced Breakdown Spectrometry (LIBS), and/or machine vision technologies, using X-Ray Transmission, and/or Color and Depth cameras are considered to encompass substantial potential to sort Al based on the alloying elements, e.g. C from W alloys, and to obtain higher purity output fractions (Díaz-Romero et al, 2021). Combining these technologies with a pneumatic valve block and/or a robotic gripping system opens the possibility of developing robust and cost-efficient systems for sorting scrap Al.…”