Fig. 6: Recognition of flapping frequency motion and wing morphology in the feathered flapping wing robot.

a Flapping period image of the robot with feather-PVDF biohybrid mechanoreceptor, T represents a complete flapping period. b Voltage signal of the robot undergoing approximately linear sweep frequency motion from 0% to 100% of power (blue). c Irregularly variable frequency motion voltage signal within the power range of the robot (green). d Schematic diagram of transfer learning and motion recognition process in the robot. The light blue blocks represent the network structure in Fig. 4, retaining all network layers except the fully connected layer and using the voltage signal from (b) of approximately linear sweep frequency motion from 0% to 100% power for weight training of the fully connected layer (transfer learning). A new network is obtained, and the voltage signal from irregularly variable frequency motion (c) is used to test the prediction performance of the transfer network. e Time-history curves of test and identified values for linear variable frequency motion (sliding step = 0.01 s, number of samples n = 6501) and irregular variable frequency motion (sliding step = 0.01 s, number of samples n = 4151). Source data are provided as a Source Data file. f Images of different wing morphologies, including elliptical wing (red, I), high-lift wing (blue, II), and soaring wing (brown, III), representing the flight states of birds in different environments. g Motion voltage signals of different wing morphologies. Source data are provided as a Source Data file. h Feature extraction and principal component analysis of the original data, resulting in a three-dimensional feature distribution map (number of samples n = 180). i Confusion matrix of wing morphology classification obtained through CNN network training.