Eventos Anais de eventos
COBEM 2021
26th International Congress of Mechanical Engineering
DEEP CONVOLUTIONAL NEURAL NETWORKS FOR IMAGE CLASSIFICATION: A CASE STUDY IN AN ELECTRIC UTILITY WAREHOUSE
Submission Author:
Paulo Piratelo , PR
Co-Authors:
Paulo Piratelo, Rodrigo Negri de Azeredo, Eduardo Yamao, Gabriel Maidl, Rafael Martini Silva, Laércio de Jesus, Renato Penteado, Leandro dos Santos Coelho, Gideon Leandro
Presenter: Paulo Piratelo
doi://10.26678/ABCM.COBEM2021.COB2021-0981
Abstract
Warehouse management has proven to be fundamental for improvements in the productivity and organization of companies, bringing several benefits in controlling the flow of products and operations. As convolutional neural networks (CNNs) are great tools of deep learning to operate classification tasks in computer vision, six models of architectures were tested in a classification assignment through a red-green-blue (RGB) image dataset to classify equipment, allocated in the warehouse of an electric utility. Thus, the present work aims to compare the performance of these models in the asset identification process. The dataset consisted of 565 images was built in local, in an uncontrolled environment, representing real challenges that occur in many warehouses. SqueezeNet obtained the best results, reaching an accuracy of 97.5% and a F1 score of 97.4% in the test set. This comparison can be used in order to guide future works on automated and intelligent inventory management solutions on the electric maintenance field.
Keywords
Convolutional neural networks, Deep learning, Computer Vision, Warehouse Management, electric utility
