Behavior State Recognition of Complex Wingbeat Patterns Targets Based on Bi-LSTM

Radar serves as a crucial tool for observing airborne biological targets, with wingbeat frequency being a vital parameter for distinguishing between different species of organisms. The wingbeat patterns of airborne biological targets are diverse. To improve the extraction of wingbeat frequencies from radar signals, it is necessary to identify the behavioral states of different wingbeat patterns and eliminate signals that cannot be used for wingbeat extraction. This paper presents an intelligent recognition algorithm based on wingbeat pattern behavioral states. Firstly, Doppler frequency modeling is performed on wingbeat pattern targets, and then simulated data is input into a Bi-LSTM network for pattern recognition training. Finally, the algorithm's performance is validated using field radar measurements after performing short-time Fourier transformation on the measured data. The validation results based on radar experimental data indicate that this algorithm can effectively recognize wingbeat pattern behavioral states.