Convolutional Neural Networks based multi-object recognition from a RGB image

With the flow of time, the application of different kinds of intelligent systems in many sectors like security, medical operations, detecting critical diseases, space researches, industrial heavy works, automated vehicles, and many others are increasing all over the of the world, and an intelligent system works by the use of its ability of image recognition. Furthermore, image recognition has been a notable subject in the scope of digital systems in most of the works we do today. Moreover, high-dimensional data from the physical world is obtained in order to produce statistical or representative knowledge by image recognition. In this study, a mutated image recognition technique has been recommended. For the work, different objects from different images were recognized by using a trained convolutional neural network and also the accuracy of the convolutional neural network was measured to examine the performance of the system. The evaluation of the accuracy of image recognition is the main continuation of this work which was done by modifying the system of updating weights in back-propagation. We used small filters of convolution in our work. We tried in this work to demonstrate that by modifying the weight updating system a vital advancement can be reached for greater efficiency in image recognition and applying it the convolutional neural network. Inception-v3, which is trained for the ImageNet Large Visual Recognition Challenge was used in our work and we found our results to be better. It was also pointed that the correctness of the outcomes from our designed system is far better than the state-of-art standards.