Image category classification using 12-Layer deep convolutional neural network

In comparison to human vision, it's hard for systems to understand images and figure them out on their own. In the modern world, image processing is mostly done by convolutional neural networks that have been learned over time. As a result, our system categorizes real-time natural colour photographs using deep learning. In the majority of convolutional networks, the activation function is often a form of the rectified linear unit, which is prone to vanishing and exploding gradient problems. Though numerous studies have proposed solutions for resolving this issue, there has yet to be an efficient and feasible approach. To overcome this issue in our method, we combined Rectified Linear Unit (ReLU) variations without altering the activation functions or adding more layers. There are 12 hidden layers in the proposed network, and ten separate image categories were made from the CIFAR10 data set. In the experiment, we got 89.13 percent accuracy which is best when we compared our model to Alex net, Google net, and Resnet18. This shows that our model is better than convolutional neural networks with rectified linear units at categorizing natural pictures.