The Journal of Science and Engineering Applications (JSEA)

Abstract

Image classification is identified as a basic problem in the domain of computer vision as well as artificial intelligence. Such a problem plays a significant role in facilitating various applications in the real world. Although deep neural networks have proven to be effective in solving various image recognition problems, the efficiency of such models is normally limited by the occurrence of overfitting as well as the limited ability to generalize. Moreover, the models have limited optimal training. Thus, the purpose of this work is to develop a robust image classification system based on the idea of not using overly deep models. Instead, the work emphasizes the significance of using appropriate training techniques. Through experiments performed on the standard image classification dataset, the efficiency of the approach is proven by attaining the highest accuracy in the test dataset as 90.92%, surpassing the efficiency of other models.