Multiview Objects Recognition Using Deep Learning-Based Wrap-CNN with Voting Scheme

Industrial automation effectively reduces the human effort in various activities of the industry. In many autonomous systems, object recognition plays a vital role. Thus, finding a solution for the accurate recognition of objection for the autonomous system is motivated among researchers. In this sense, various techniques have been designed with the support of classifiers and machine learning techniques. But those techniques lack their performance in the case of Multiview object recognition. It is found that a single classifier or machine learning algorithm is not enough to recognize Multiview objects accurately. In this paper, a Wrap Convolutional Neural Network (Wrap-CNN) with a voting scheme is proposed to solve the Multiview object recognition problem and attain better recognition accuracy. The proposed model consists of three phases such as pre-processing, pre-training CNNs and voting schemes. The pre-processing phase is done to remove the unwanted noise. These pre-trained CNN models are used as feature extractors and classify the images into their respective classes. Here, the Wrap-CNN, nine pre-trained CNN are used in parallels, such as Alex Net, VGGNet, GoogLeNet, Inceptionv3, SqueezeNet, ResNet v2, Xception, MobileNetV2 and ShuffleNet. Finally, the output class from the nine predicted classes is chosen based voting scheme. The system was tested in two scenarios, such as images without rotation and with rotation. The overall accuracy is 99% and 93% for without rotation and with rotation recognition, respectively. Ultimately the system proves the effectiveness for the Multiview object recognition, which can be used for the industrial automation system.