This paper presents an extensive research carried out for enhancing the performances of convolutional neural network (CNN) object detectors applied to municipal waste identification. In order to obtain an accurate and fast CNN architecture, several types of Single Shot Detectors (SSD) and Regional Proposal Networks (RPN) have been fine-tuned on the TrashNet database. The network with the best performances is executed on one autonomous robot system, which is able to collect detected waste from the ground based on the CNN feedback. For this type of application, a precise identification of municipal waste objects is very important. In order to develop a straightforward pipeline for waste detection, the paper focuses on boosting the performance of pre-trained CNN Object Detectors, in terms of precision, generalization, and detection speed, using different loss optimization methods, database augmentation, and asynchronous threading at inference time. The pipeline consists of data augmentation at the training time followed by CNN feature extraction and box predictor modules for localization and classification at different feature map sizes. The trained model is generated for inference afterwards. The experiments revealed better performances than all other Object Detectors trained on TrashNet or other garbage datasets with a precision of 97.63% accuracy for SSD and 95.76% accuracy for Faster R-CNN, respectively. In order to find the optimal higher and lower bounds of our learning rate where the network is actually learning, we trained our model for several epochs, updating the learning rate after each epoch, starting from 1 × 10−10 and decreasing it until reaching 1 × 10−1.
Taking as an example six main rivers that drain the western flank of the Eastern Carpathians, a conceptual model has been developed, according to which fluvial bed sediment bimodality can be explained by the overlapping of two grain size distribution curves of different origins.Thus, for Carpathian tributaries of the Siret, coarse gravel joins an unimodal distribution presenting a right skewness with enhanced downstream fining. The source of the coarse material distributions is autohtonous (by abrasion and hydraulic sorting mechanisms). A second distribution with a sandy mode is, in general, skewed to the left. The source of the second distribution is allohtonous (the quantity of sand that reaches the river-bed through the erosion of the hillslope basin terrains). The intersection of the two distributions occurs in the area of the 0·5-8 mm fractions, where, in fact, the right skewness (for gravel) and left skewness (for sand) histogram tails meet. This also explains the lack of particles in the 0·5-8 mm interval. For rivers where fine sediment sources are low, the 0·5-8 mm fractions have a higher proportion than the fractions under 1 mm.For the Siret River itself, bed sediment bimodality is greatly enhanced due to the fact that the second mode is more than 25% of the full sample. As opposed to its tributaries, the source of the first mode, of gravel, is allohtonous to the Siret river, generated by the massive input of coarse sediment through the Carpathian tributaries, while the second mode, of the sands, is local. In this case we can also observe that the two distributions of particles of different origins overlap in the 0·5-8 mm fraction domain, creating the illusion of 'particle lack' in the fluvial bed sediments.acquisition of a comprehensive database in order to better understand the diversity of situations in the field that may involve the process of river-bed material diminution. This opinion is shared by many authors (Sambrook Smith and Ferguson, 1996;Rice, 1998;Gomez et al., 2001) and we also sustain it.For 10 years we have focused our attention on rivers in the drainage basin of the Siret, an important affluent of the Danube in the Romanian territory. We took as examples the experience of many authors (Brierley and Hickin) in their research on downstream variation in grain size on a single river or a river sector; we also thought that a spatial approach of the variability of the river-bed material on many rivers in a river system of over 43 000 km 2 would bring an important understanding in this field. A similar approach has been taken by authors such as Yatsu (1955), Knighton (1980), Ibbeken and Schleyer (1991) and many others. This method proves to be difficult due to the fact that volumetric sampling in river gravel-beds is a significant stumbling block for those that study the phenomena. For instance, in the higher part of the Carpathian rivers that we have sampled, the weight of the sample in situ was more than 1000 kg, which implied an extraordinary effort for the team (see Figure 3 below).In conc...
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.