The procedure of labeling a picture is image classification, and deep learning is well suited to this area. This is because images are metaphysical and can use the parallel structure to learn a variety of characteristics. In this research, we define three basic arrangements for the Convolution Neural Network model. One setup is straightforward, and the other two are advanced variations on the fundamental level, which uses the overtaking avoidance method. CIFAR-10, a dataset of 60000 image sets of 10 different kinds of products, has been trained and tested. Based on the various performance matrices, comparisons of model variants indicate that the accuracy of the model may be significantly enhanced by employing the drop-out regularization and also, fewer batches could also get better results than bigger ones.