THE QUESTIONS OF APPLYING ARTIFICIAL NEURAL NETWORKS IN DEVELOPING A MACHINE FOR ACCEPTING CONTAINERS
The paper describes the results of developing an original reverse vending machine on the basis of IoT and neural networks. These automatic machines shall perform the assigned functions (to recognize containers – bottles of PET and aluminum cans), to be safe (not to endanger health and the property of the user), but at the same time time to be cost-effective in terms of production and service. These machines are commonly developed on the basis of IoT controllers and single-board computers. They have very limited memories and computing capabilities, however allow to use a big list of specialized devices for project implementation (cameras, barcode scanners, spectrometers, etc.). For lowering the price cost an attempt to abandon the use of expensive electronics was made, by using only visual recognition of an object on the basis of neural networks. Some of the most popular networks based on Keras were selected for this purpose and trained over Tensorflow and Caffe: LeNet, AlexNet, SqueezeNet. Comparative tests were carried out and conclusions were drawn.
Authors: A. N. Kokoulin, A. I. Tur, A. A. Yuzhakov
Direction: Automation and Control
Keywords: Reverse vending machine, artificial neural networks, convolutional neural network, recycling
View full article