Comparing deep learning and support vector machines for autonomous waste sorting
- Resource Type
- Conference
- Authors
- Sakr, George E.; Mokbel, Maria; Darwich, Ahmad; Khneisser, Mia Nasr; Hadi, Ali
- Source
- 2016 IEEE International Multidisciplinary Conference on Engineering Technology (IMCET) Multidisciplinary Conference on Engineering Technology (IMCET), IEEE International. :207-212 Nov, 2016
- Subject
- Bioengineering
Communication, Networking and Broadcast Technologies
Components, Circuits, Devices and Systems
Computing and Processing
Power, Energy and Industry Applications
Robotics and Control Systems
Signal Processing and Analysis
Neurons
Support vector machines
Convolution
Sorting
Training
Biological neural networks
Computers
- Language
Waste sorting is the process of separating waste into different types. The current trend is to efficiently separate the waste in order to appropriately deal with it. The separation must be done as early as possible in order to reduce the contamination of waste by other materials. The need to automate this process is a significant facilitator for waste companies. This research aims to automate waste sorting by applying machine learning techniques to recognize the type of waste from their images only. Two popular learning algorithms were used: deep learning with convolution neural networks (CNN) and support vector machines (SVM). Each algorithm creates a different classifier that separates waste into 3 main categories: plastic, paper and metal using only 256 × 256 colored png image of the waste. The accuracies of the two classifiers are compared in order to choose the best one and implement it on a raspberry pi 3. The pi controls a mechanical system that guides the waste from its initial position into the corresponding container. However, in this paper we only compare the two machines learning techniques and implement the best model on the pi in order to measure its speed of classification. SVM achieved high classification accuracy 94.8% while CNN achieved only 83%. SVM also showed an exceptional adaptation to different types of wastes. NVIDIA DIGITS was used for training the CNN while Matlab 2016a was used to train the SVM. The SVM model was finally implemented on a Raspberry pi 3 where it produced quick classification, taking on average 0.1s per image.