Comparison of Self Developed Deep CNN Architecture with State of Art Existing Techniques

The growth of consistent and tough image identification systems is a main challenge in artificial intelligence and computer vision. Understanding to identify objects requires classifier that are able to process relevant information under image variation like scaling, rotations, light and background changes, etc. Deep Convolutional Neural Networks can be trained using standard image datasets and own datasets but an optimized Deep CNN systems designed for more accuracy with minimum training images.

In our system, major finding includes various training methods for deep CNN architectures on existing and our own developed dataset which is the collection of real images. In our enhanced method proved that our deep CNN architecture get better accuracy and quality (attribute) properties with minimum training.