Object Classification System Using Convolutional Neural Network for Several Environmental Conditions

At this time, autonomous mobile robots are used to move groceries. Therefore, object classification is required to move groceries to avoid collisions with other objects. Object classification using a convolutional neural network (CNN) can significantly increase mobile robot autonomy. Therefore, this paper investigates object classification performance using CNN for several environmental conditions. The research method is computer simulation using several steps: image collection, image annotation, and training. SSD Mobilenetv2FPNLite was used as a CNN model for object classification. The objects for the classification using the pre-trained CNN are cars, motorcycles, persons, and goods rickshaws. The results enabled us to estimate the effectiveness of using pre-trained CNN for classifying different objects. The result also shows that objects can be recognized with a classification accuracy of 100%.