A Comparative Study of Deep Learning Models for Recognition of Poisonous Foods for Dogs

Abstract

Convolutional neural network (CNN) models are widely applied in various areas including image classification, machine translation, autonomous driving, natural language processing, face recognition, recommendation systems, among others. This work investigated and compared different deep convolutional neural network models for image classification on a custom dataset. The models were trained on a dataset composed of seven image classes. The images were collected from various sources and a dataset for training the CNN models was created. The images included fruits, vegetables, and chocolates, which are considered poisonous to dogs for which Labrador Retrievers are used as a case study. Among the trained models, the Xception model showed the best performance, with a testing accuracy of 95 %. Other notable models with high performance included InceptionV3, InceptionResNetV2, MobileNetV2 and VGG-16 with testing accuracy of 93.5 %, 94.4 %, 92.0 % and 91.5% respectively. The trained models were able to easily recognize the food classes that are considered poisons for Labrador Retrievers on independent user images, with very high accuracy.

Keywords

• cnn
• image processing
• machine learning
• classification