Comparison of Different DNN Models for Classification of Different Sports Images

In today’s world, sports have become an important part of almost everyone’s life. Any kind of sports, not only it entertains people but also helps to maintain physical and mental health of oneself. In this fast-growing world, sports industry is also evolving with same pace. Due to development in the sports field, various new sports are being introduced, which most of the people are not aware about. Also, there are few sports which are getting newly introduced, but have very similar features like other sports which already exist. So, classifying such sports activity through images can be very confusing. But with the help of deep learning techniques, it has become easy to understand what is happening inside an image. It has now possible to extract the features from complex images by using Deep Neural Network (DNN) techniques. So, using DNN techniques different sports recognition and classification is possible. In this paper, we have proposed a comparative analysis of different DNN algorithms for recognition and classification of different sports images and coming to a conclusion about which model is best suited for deployment on the embedded systems. We have compared ResNet-50, Efficient Net , and MobileNet-V2 models. Based on the parameters like accuracy, model size, parameters generated, and F1 score. In this paper, F1 score obtained for different models like ResNet-50, Efficient Net , and MobileNet-V2 are 97%, 95% and 94% respectively. Different models can be chosen according to the requirements. The Efficient Net model gave the best accuracy among all and the MobileNet-V2 model consumed the least space. Details of the models and comparative study are discussed further in the paper.