Deep Learning Computer Vision for Sorting and Size Determination of Municipal Waste

Melinte, Octavian; Dumitriu, Dan; Mărgăritescu, Mihai; Ancuţa, Paul-Nicolae

doi:10.1007/978-3-030-26991-3_14

Cited by 8 publications

(17 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Although the object detector architecture is more complex than an Image Classification CNN and the accuracy is measured in a different manner, all the models used in this research obtained better results than these classifiers. [7] 81.6 Yes COCO VGG-16 [6] 88.42 No ImageNet ResNet50 [9] 87 No No RecycleNetV4 [8] 81 No No Adam DenseNet121 [8] 95 No ImageNet Adam, SGD Inception-ResNetV2 [8] 87 No ImageNet Adam, SGD…”

Section: Discussionmentioning

confidence: 99%

“…Furthermore, in order to evaluate the generalization of our models outside the TrashNet dataset with real-time images, at the inference time, a custom small set of images has been created, consisting of 25 images of plastic and glass bottles which are completely different from the ones the model has been trained with and tested/validated against. The same evaluation has been performed in our first paper regarding municipal waste sorting using CNN [6]. Due to the fact that the detection of bottles in municipal waste is a process that involves identification of objects that have different degrees of deformation, shapes, position, and transparency, this small set of images has been taken into account for testing at the inference time.…”

Section: Discussionmentioning

confidence: 99%

“…The goal is to obtain a real-time and accurate CNN object detector which will be able to correctly classify and locate municipal waste, in order to be identified and picked up by the autonomous mobile robot system. A preliminary study in the area of waste sorting using deep learning has been performed by our team in a previous paper [6]. More precisely, pre-trained CNN architectures have been evaluated on a custom trash test dataset, created in under laboratory conditions, with a high degree of variance (high degree of transparency, uncommon positions, lower resolution, cropped or blocked images), in order to test the real generalization of the models outside the training and testing dataset.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Deep Convolutional Neural Networks Object Detector for Real-Time Waste Identification

2020

Self Cite

View full text Add to dashboard Cite

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.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Deep Convolutional Neural Networks Object Detector for Real-Time Waste Identification

2020

Self Cite

View full text Add to dashboard Cite

show abstract

“…Finally, a computer program was used to resize/adjust all images to the same resolution. In a later phase of the work it will be necessary to increase the number of images to be treated, reaching 1500 images (see Figure 1) of each type for training purposes [16][17][18].…”

Section: Methodsmentioning

confidence: 99%

“…To find the characteristics that define that image, it is necessary to multiply the array by a set of feature detectors. As a result, you will get the feature map according to the different feature detectors that we have applied to our image [9,17,18].…”

Section: Convolutionmentioning

confidence: 99%

Digitalization, Circular Economy and Environmental Sustainability: The Application of Artificial Intelligence in the Efficient Self-Management of Waste

Alonso

Forradellas

Morell³

et al. 2021

Sustainability

View full text Add to dashboard Cite

The great advances produced in the field of artificial intelligence and, more specifically, in deep learning allow us to classify images automatically with a great margin of reliability. This research consists of the validation and development of a methodology that allows, through the use of convolutional neural networks and image identification, the automatic recycling of materials such as paper, plastic, glass, and organic material. The validity of the study is based on the development of a methodology capable of implementing a convolutional neural network to validate a reliability in the recycling process that is much higher than simple human interaction would have. The method used to obtain this better precision will be transfer learning through a dataset using the pre-trained networks Visual Geometric Group 16 (VGG16), Visual Geometric Group 19 (VGG19), and ResNet15V2. To implement the model, the Keras framework is used. The results conclude that by using a small set of images, and thanks to the later help of the transfer learning method, it is possible to classify each of the materials with a 90% reliability rate. As a conclusion, a model is obtained with a performance much higher than the performance that would be reached if this type of technique were not used, with the classification of a 100% reusable material such as organic material.

show abstract

Visual Product Inspection Based on Deep Learning Methods

Kuric

Kandera

Klarák

et al. 2020

Lecture Notes in Mechanical Engineering

View full text Add to dashboard Cite

Deep Learning Computer Vision for Sorting and Size Determination of Municipal Waste

Cited by 8 publications

References 4 publications

Deep Convolutional Neural Networks Object Detector for Real-Time Waste Identification

Deep Convolutional Neural Networks Object Detector for Real-Time Waste Identification

Digitalization, Circular Economy and Environmental Sustainability: The Application of Artificial Intelligence in the Efficient Self-Management of Waste

Visual Product Inspection Based on Deep Learning Methods

Contact Info

Product

Resources

About