A Convolutional Recurrent Neural Network and Feature Pyramid Network based Tactile Sensing for Perceptual Intelligence Recognition System

Nowadays, in the Artificial Intelligence (AI) community, developing an efficient artificial sensory system is important for efficient interaction between human-machine. The existing models for perceptual intelligence recognition system did not process the sequential data in real time and doesn’t capture complex contextual relationships in the tactile data. Therefore, this paper proposes Convolutional Recurrent Neural Networks and Feature Pyramid Network (CRNN-FPN) for recognizing the perceptual intelligence systems on tactile sensor which effectively captured the complex contextual relationships and processed the sequential data in real time. The input tactile sensor contains 3 layers they are soft contact interface, shell and a Complementary Metal-Oxide-Semiconductor (CMOS) image sensor; then, CRNN is applied for extracting the features from tactile sensor. After that, extracted features are sent for backbones, where, ShuffleNet, Cross Stage Partial Network (CSPNet), and swim transformer are used to obtain feature maps. These feature maps are fused with FPN and its outputs are then passed over a decoder through four channels where, each channel contains two 3×3 convolutional layers and one 1×1 convolutional layer. In the end, the output of each channel produces localization, information, angle posture information and normal force information. The proposed CRNN-FPN achieved better results in terms of accuracy 99.23 % and precision of 98.32 % when compared with existing transfer learning model.