A Lean Convolutional Neural Network for Vehicle Classification
Jonathan Jesus Sanchez-Castro, Julio Cesar Rodriguez-Quinonez, Luis Roberto Ramírez-Hernández, Guillermo Galaviz, Daniel Hernández Balbuena, Gabriel Trujillo‐Hernández, Wendy Flores‐Fuentes, Paolo Mercorelli, Wilmar Hernández, Oleg Sergiyenko, Felix Fernando Gonzalez-Navarro · 2020
Image classification is an important task in machine vision, in which vehicle classification is used for different applications like traffic analysis, autonomous driving, security, among others. Recent studies made with Convolutional Neural Networks (CNN) have shown that these networks have surpassed older algorithms like Support Vector Machine (SVM) and K-Nearest Neighbor (KNN) in terms of accuracy, speed, and resources management. Even though that CNN have better accuracy and speed they still are heavy in resource consumption on computers which makes them not suitable to deploy on an embedded platform. This paper proposes a lean CNN that has a smaller number of parameters and still maintaining the best accuracy possible on vehicle classification.